{"version":1,"source":"repository","updatedAt":"2026-09-25","totalRecordCount":58,"recordCount":58,"filters":{"q":"","repo":"","area":"","dimension":"","id":""},"summary":{"version":1,"source":"repository","updatedAt":"2026-09-25","recordCount":58,"claimCount":80,"repositories":["manaflow-ai/cmux","swiftlang/swift","teamleaderleo/glaeda"],"countsByConfidence":{"measured":41,"observed":27,"derived":10,"projected":2},"countsByDimension":{"critical-path":30,"paid-compute":2,"compute":25,"developer-loop":5,"reliability":7,"storage":4,"observability":2,"replacement-cost":2,"cache-transfer":1,"control-plane":1,"cache":1},"headlineClaims":[{"recordId":"cmux-14314","repository":"manaflow-ai/cmux","title":"Start macOS compile admission beside the fast Linux jobs","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14314#issuecomment-5825460838","id":"admission-gate-wait","dimension":"critical-path","metric":"run start to compile admission job created","direction":"reduction","value":121,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=62 before, n=16 after, 2026-09-24/25","population":"pull_request ci.yml runs, first attempts","note":"160.5 s to 39.5 s median; p90 236 s to 77.5 s. One late-night hour after merge; re-measure on a weekday. Job start and compile start also moved (282 s to 60 s, 430 s to 242 s), but pool routing changes merged in the same hour confound those.","headline":true,"additive":false,"overlapGroup":"cmux-admission-start"},{"recordId":"cmux-14318","repository":"manaflow-ai/cmux","title":"ci: place each PR macOS job on a free owned mini, overflow the rest","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14318#issuecomment-5826976943","id":"pr-admission-owned-share","dimension":"paid-compute","metric":"share of attempt-1 PR compile admissions placed on owned Mac minis","direction":"shift","value":49,"unit":"percent of admissions","recurrence":"per-event","confidence":"measured","measurementWindow":"before 2026-09-24 03:46Z to 14:59Z (n=223, 0 owned), after 2026-09-25 01:13Z to 04:37Z (n=53, 26 owned)","population":"attempt-1 PR compile admission jobs in ci.yml","note":"0% to 49%, over one 3.5-hour window of US night. Owned-placed admissions waited 2 s median and 3 s p90 for a runner; the other 27 overflowed to Blacksmith. Of the 26 owned attempt-1 jobs, 5 succeeded, 11 failed and 10 were cancelled (mostly superseded pushes). The failures mix real build failures with owned-runner refusals, which the retry path (cmux#14312, cmux#14325) re-runs as attempt 2; this measurement keeps attempt 1 only, so a refused job counts as owned and its refusal plus re-queue time is not in the wait. Runner speed after start is not compared here.","headline":true,"additive":false,"overlapGroup":"cmux-pr-routing"},{"recordId":"cmux-14333","repository":"manaflow-ai/cmux","title":"Stop portal lifecycle tests waiting 12 s each on login(1)","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14333#issuecomment-5827083427","id":"portal-suite-duration","dimension":"compute","metric":"TerminalWindowPortalLifecycleTests suite time per CI run","direction":"reduction","value":166,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"before 2026-09-24 15:16Z to 2026-09-25 03:29Z (n=41, median 173.6 s), after 2026-09-25 04:00Z to 04:21Z (n=2, 4.5 s and 7.3 s)","population":"ci.yml app-host unit test shard 1 jobs","note":"173.6 s median minus the larger after value (7.3 s). Only two post-merge runs existed at recording time, so this stays observed. Shard 1's job dropped more than the suite did (581 s median to 319 s and 360 s), so part of that drop comes from other merges and is not claimed. Shard 4 was already the longest app-host job, so this saves shard 1 runner time rather than shortening the app-host lane.","headline":true,"additive":false,"overlapGroup":null},{"recordId":"cmux-14334","repository":"manaflow-ai/cmux","title":"Stop the CLI no-socket regression waiting 90 s for an app it never starts","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14334#issuecomment-5827083266","id":"no-socket-test-duration","dimension":"compute","metric":"testLaunchCapableCommandsReachTheirDispatchPathWithoutLiveImplicitSocket duration","direction":"reduction","value":88,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 16:29Z to 2026-09-25 01:03Z (n=28), after 2026-09-25 01:16Z to 02:43Z (n=5)","population":"ci.yml app-host unit test shard 4 jobs that ran the full CMUXCLIErrorOutputRegressionTests suite","note":"90.4 s to 2.4 s median; every before run fell in 90.2 to 90.6 s and every after run in 2.35 to 2.54 s. Six runs that ran the test in under 1 s (a different selection), three on each side, are excluded. The test runs inside the parallel 'Run unit tests' step, so its time is not all wall time.","headline":true,"additive":false,"overlapGroup":"cmux-app-host-shard4-duration"},{"recordId":"cmux-14346","repository":"manaflow-ai/cmux","title":"Replay input times onto an owned Mac's kept DerivedData","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14346#issuecomment-5826980830","id":"warm-owned-whole-module-rebuilds","dimension":"compute","metric":"share of warm owned compile admissions that rebuilt the whole cmux module","direction":"reduction","value":35,"unit":"percentage points","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 23:15Z to 2026-09-25 02:31Z (n=11), after 2026-09-25 03:18Z to 04:37Z (n=15)","population":"Warm compile admissions on owned glaeda Macs (kept DerivedData adopted), jobs with an Xcode timing summary","note":"82% (9 of 11) to 47% (7 of 15). The task counts are bimodal, so the median (2615 to 1236 tasks) sits on the boundary job of the lower cluster and one more whole-module job would move it back near 2480; it is not claimed. Jobs 7 to 52 changed inputs from the kept build compiled 1 to 177 tasks; jobs 71 to 187 inputs away still rebuilt the module, because the kept build is the previous PR's, not the merge base. The PR's stamp bump made each Mac's first job after merge cold (18 of 35 owned admissions cold after, 13 of 25 before), a one-time cost not netted here.","headline":true,"additive":false,"overlapGroup":"cmux-owned-warm-admission"},{"recordId":"cmux-14349","repository":"manaflow-ai/cmux","title":"Build cmuxTests without the compilation cache so it rebuilds incrementally","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14349","id":"cmuxtests-test-edit-wall-time","dimension":"critical-path","metric":"cmuxTests build time after a one-test-file body edit","direction":"reduction","value":108,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36081880621, 2026-09-25 01:29Z to 02:08Z: 139 s with the cache on (129 s in a second series), 31 s with it off for cmuxTests","population":"Incremental cmux-unit rebuilds over admission DerivedData on blacksmith-12vcpu-macos-26, Xcode 26.6","note":"One canary run, one edit per mode. Uses 139 s minus 31 s; the second cache-on series gives 98 s. Only pays off when an admission rebuilds on adopted DerivedData. Adding a top-level declaration in the app still recompiles every cmuxTests file with or without the cache (117 s in the same run). Of the remaining 31 s, 25.7 s was cmuxTests' own emit-module job, which cmux#14364 then removed.","headline":true,"additive":false,"overlapGroup":"cmux-cmuxtests-rebuild"},{"recordId":"cmux-14359","repository":"manaflow-ai/cmux","title":"Build the app without the compilation cache before Xcode 26.6","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14359","id":"macos15-app-body-edit-recompiles","dimension":"compute","metric":"cmux target SwiftCompile tasks for a body-only edit to one app file","direction":"reduction","value":5217,"unit":"tasks/build","recurrence":"per-run","confidence":"observed","measurementWindow":"two edits each way in run 36086596738 (5,219 and 5,224 before, 2 and 2 after), confirmed by two edits in run 36090019330 (2 and 2)","population":"Incremental cmux builds on the Blacksmith macOS 15 pool (Xcode 26.3)","note":"Two edits each way, one runner each. Only pays off in PR admission when it adopts DerivedData to rebuild on: the macOS 15 seeds (cmux#14315, cmux#14358) merged the same morning, and seeds built before this fix used the cache for cmux, so the first adoption after a stale seed can still rebuild the app fully. Xcode 26.6 pools were already incremental and are unchanged.","headline":true,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"},{"recordId":"cmux-14360","repository":"manaflow-ai/cmux","title":"Run the CLI pipe regressions in compile admission, drop the separate lane","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14360#issuecomment-5826986283","id":"cli-lane-runner-time","dimension":"compute","metric":"macOS runner time of the removed cli-pipe-regressions job","direction":"reduction","value":3.28,"unit":"runner-minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"lane jobs n=41, 2026-09-25 00:46Z to 03:58Z; smoke step n=54 before, n=15 after (03:58Z to 04:47Z)","population":"ci.yml pull_request runs that routed to the CLI lane (41 of 137 in the window); 18 on glaeda minis, 23 on Blacksmith macOS","note":"Median lane job 197 s, gross, on runs that routed to the CLI lane. The mean lane job was 382 s, so the median is the conservative figure. Averaged over all 137 PR runs in the window this is about 59 s per run. This removal also ends the cli-pipe-resolve-time claim in cmux-14257.","headline":true,"additive":false,"overlapGroup":"cmux-cli-lane"},{"recordId":"cmux-14364","repository":"manaflow-ai/cmux","title":"Build cmuxTests without emitting its Swift module","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14364","id":"cmuxtests-test-edit-wall-time","dimension":"critical-path","metric":"cmuxTests build time after a one-test-file body edit","direction":"reduction","value":27.4,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36089490735, 2026-09-25 03:25Z to 03:40Z: 35.3 s and 31.6 s with the module emitted, 4.2 s and 4.1 s without","population":"Incremental cmux-unit rebuilds over admission DerivedData on blacksmith-12vcpu-macos-26, Xcode 26.6","note":"Uses the smaller before (31.6 s) minus the larger after (4.2 s). The whole cmux-unit rebuild went from 47 to 55 s to 19 s in the same series, which contains this saving, so it is not claimed separately. A standalone-driver series that still emitted the module took 36.4 s, so the gain comes from dropping the module. Only pays off when an admission rebuilds on adopted DerivedData, and the first admissions after the merge rebuild cmuxTests once because its arguments changed. Too few post-merge admissions existed at recording time (29 minutes after the merge) to measure it there. cmux#14378 (merged 04:40Z) applies the same settings to the local scripts/test-unit.sh.","headline":true,"additive":false,"overlapGroup":"cmux-cmuxtests-rebuild"},{"recordId":"glaeda-1190","repository":"teamleaderleo/glaeda","title":"worktree reclaim: delete local branches whose work has landed (--branches)","evidence":"https://redirect.github.com/teamleaderleo/glaeda/pull/1190#issuecomment-5826944682","id":"landed-branches-deleted","dimension":"developer-loop","metric":"landed local branches deleted by the hourly reclaim on Air Blue","direction":"reduction","value":255,"unit":"branches","recurrence":"one-time","confidence":"measured","measurementWindow":"first four branch-pass runs after merge, 2026-09-24 20:03 to 2026-09-25 00:15 EDT","population":"Local branches in the Git repositories under ~/Projects on Air Blue","note":"Backlog clearance, capped at 64 per run, and eligible branches remained, so the total keeps rising. Before, these were only removed by a manual git branch -D. Big Red runs the same job; its receipts were not read here.","headline":true,"additive":false,"overlapGroup":"glaeda-worktree-reclaim"},{"recordId":"glaeda-1209","repository":"teamleaderleo/glaeda","title":"fleet-cas: tailnet addressing, launchd wrapper, upstream error logging","evidence":"https://redirect.github.com/teamleaderleo/glaeda/pull/1209#issuecomment-5826438514","id":"deployed-cross-machine-hits","dimension":"reliability","metric":"cacheable tasks hit through a deployed node reading another mini's store","direction":"reliability","value":4192,"unit":"tasks/build","recurrence":"per-run","confidence":"observed","measurementWindow":"one fresh full-app build on cmux11s before merge, 2026-09-25","population":"Fresh full cmux app builds on Manaflow minis using the deployed fleet-cas node (Xcode 26.6)","note":"Before, the deployed (launchd) node could not reach the store at all, so deployed cross-machine reads served nothing; lab reads in glaeda-1162 and glaeda-1180 ran from ssh, which the block does not affect. The 326 s build time is not claimed as a reduction: there is no same-host, same-Xcode cold baseline for it (756 s in glaeda-1162 was Xcode 26.3).","headline":true,"additive":false,"overlapGroup":"glaeda-fleet-cas-fresh-build"},{"recordId":"cmux-14089","repository":"manaflow-ai/cmux","title":"Prune expired R2 cache archives, never one a latest pointer names","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14089","id":"r2-storage-freed-first-run","dimension":"storage","metric":"R2 cache bytes deleted by the first prune","direction":"reduction","value":117.2,"unit":"GiB","recurrence":"one-time","confidence":"measured","measurementWindow":"run 35972452822, 2026-09-24","population":"manaflow-ai/cmux CI cache bucket","note":"One-time reclaim. The steady-state effect is capping growth at about one day of per-commit archives, not measured here.","headline":true,"additive":false,"overlapGroup":"cmux-r2-cache-storage"},{"recordId":"cmux-14188","repository":"manaflow-ai/cmux","title":"Seed DerivedData on the 12 vCPU macOS 26 pool","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14188#issuecomment-5825462059","id":"seed-queue-wait","dimension":"critical-path","metric":"seeder job queue wait","direction":"reduction","value":1326,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=6 before, n=16 after, 2026-09-24","population":"seed-derived-data.yml runs on main","note":"1,342 s to 16 s median. After 14:00Z the 12 vCPU pool queued too (621 s and 663 s), once PR admissions began preferring it.","headline":true,"additive":false,"overlapGroup":"cmux-seed-freshness"},{"recordId":"cmux-14226","repository":"manaflow-ai/cmux","title":"Build focused runs from compile admission's DerivedData seed","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14226#issuecomment-5825517035","id":"e2e-build-6vcpu","dimension":"critical-path","metric":"E2E app build step on 6 vCPU macOS 26","direction":"reduction","value":453,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=27 before, n=16 after, 2026-09-24/25","population":"All focused test-e2e.yml builds on 6 vCPU macOS 26","note":"888 s to 435 s median. The before side includes 7 builds that had already adopted main's DerivedData or an early admission seed.","headline":true,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"},{"recordId":"cmux-14255","repository":"manaflow-ai/cmux","title":"Start macOS checkouts from main's git objects","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14255","id":"admission-checkout-time","dimension":"critical-path","metric":"compile admission checkout plus git seed restore","direction":"reduction","value":36,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"80-run median before; 1 admission job after","population":"macOS compile admission","note":"48 s median to 12 s (4 s checkout plus 8 s restore). One after-sample per lane.","headline":true,"additive":false,"overlapGroup":"cmux-macos-checkout"},{"recordId":"cmux-14257","repository":"manaflow-ai/cmux","title":"Restore SwiftPM's manifest cache for package resolves","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14257","id":"admission-resolve-time","dimension":"critical-path","metric":"admission Resolve Swift packages step","direction":"reduction","value":46,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 11:00Z to 16:40Z (n=21), after 17:34Z to 18:40Z (n=12), 2026-09-24","population":"macOS 26 compile admissions","note":"63 s to 17 s median.","headline":true,"additive":false,"overlapGroup":"cmux-swiftpm-resolve"},{"recordId":"cmux-14275","repository":"manaflow-ai/cmux","title":"Seed each Swift job width admission compiles at","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14275#issuecomment-5824983225","id":"seeded-admission-module-reemits","dimension":"compute","metric":"SwiftEmitModule tasks in the cmux scheme of a test-only seeded admission","direction":"reduction","value":62,"unit":"tasks/admission","recurrence":"per-run","confidence":"observed","measurementWindow":"one admission before (run 36040706124) and one after (run 36077356582), both 6 vCPU","population":"Seeded macOS compile admissions on Blacksmith macOS 26 pools","note":"62 to 0, one run each. Holds when a seed of the runner's width exists. Owned 14-CPU minis have none yet (cmuxterm-hq#590). cmuxTests still recompiles fully after adoption (cmux#14320).","headline":true,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"},{"recordId":"cmux-14282","repository":"manaflow-ai/cmux","title":"Keep the runner account out of the SwiftPM manifest cache key","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/14282#issuecomment-5823586250","id":"mini-admission-resolve-time","dimension":"critical-path","metric":"admission Resolve Swift packages step on owned minis","direction":"reduction","value":36,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"before 19:07 to 20:40Z; after 21:55 to 22:55Z (n=4), 2026-09-24","population":"Compile admissions on owned glaeda minis","note":"52 to 56 s before, 13 to 16 s after; 36 s is the conservative gap (52 minus 16). cli-pipe-regressions on a mini still missed (41 s, n=1) because its workspace path differs.","headline":true,"additive":false,"overlapGroup":"cmux-swiftpm-resolve"},{"recordId":"glaeda-1162","repository":"teamleaderleo/glaeda","title":"Measure the fleet compilation cache mini to mini on the full cmux app","evidence":"https://github.com/teamleaderleo/glaeda/pull/1162","id":"fresh-full-app-build","dimension":"developer-loop","metric":"fresh full cmux app build with a filled fleet cache","direction":"reduction","value":624,"unit":"seconds/event","recurrence":"per-event","confidence":"observed","measurementWindow":"one cold and one cached build, 2026-09-24","population":"Fresh builds on a Manaflow M4 Pro mini","note":"756 s to 132 s. Lab measurement, not CI. For the full app the store and reader were on the same mini; cross-machine was shown on the CmuxSettingsUI chain, and for the full app in glaeda-1180.","headline":true,"additive":false,"overlapGroup":"glaeda-fleet-cas-fresh-build"},{"recordId":"cmux-13935","repository":"manaflow-ai/cmux","title":"Run the CmuxMobileShell package tests serially","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13935","id":"serial-test-time-lower-bound","dimension":"critical-path","metric":"package-test-wall-time-reduction-lower-bound","direction":"reduction","value":655,"unit":"seconds/run","recurrence":"per-run","confidence":"derived","measurementWindow":"same-machine parallel/serial comparison","population":"1,060 CmuxMobileShell package tests","note":"Parallel execution exceeded 12 minutes without finishing; serial execution completed in 65 seconds. The stored 655-second reduction is therefore a lower bound.","headline":true,"additive":false,"overlapGroup":null},{"recordId":"cmux-13902","repository":"manaflow-ai/cmux","title":"Move focused E2E dispatches to the low-queue macOS pool","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13902","id":"e2e-queue-p90-gap","dimension":"critical-path","metric":"focused-e2e-queue-p90-reduction","direction":"reduction","value":82.3,"unit":"minutes","recurrence":"none","confidence":"observed","measurementWindow":"60 dispatches from 2026-09-22T21:19Z to 2026-09-23T05:18Z","population":"33 macOS 15 and 25 macOS 26 focused E2E dispatches","note":"Observed queue p90 was 83.3 minutes on macOS 15 and 1.0 minute on macOS 26. Execution median was about four minutes slower on macOS 26, and the test mixes differed, so the comparison is operational rather than causal.","headline":true,"additive":false,"overlapGroup":"cmux-macos-queue-2026-09"},{"recordId":"cmux-13749","repository":"manaflow-ai/cmux","title":"Read app-host artifacts over parallel range requests","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13749","id":"artifact-transfer-baseline","dimension":"compute","metric":"artifact-transfer-job-minutes-per-full-suite","direction":"baseline","value":60,"unit":"job-minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"two referenced CI runs","population":"Seven app-host product consumers in a full-suite run","note":"The baseline was about eight minutes per consumer; the PR reports the transfer step as roughly 98 percent transport time and moves the path to concurrent ranges with verification and fallback.","headline":true,"additive":false,"overlapGroup":null},{"recordId":"cmux-13646","repository":"manaflow-ai/cmux","title":"Throttle push-triggered Nightly builds by published commit age","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13646","id":"nightly-push-savings-low-edge","dimension":"compute","metric":"projected-macos-runner-minutes-avoided-per-day","direction":"reduction","value":450,"unit":"runner-minutes/day","recurrence":"daily","confidence":"projected","measurementWindow":"seven-day replay","population":"push-triggered Nightly publication work","note":"The PR projects roughly 450–600 macOS minutes/day avoided at the two-hour default. This record stores the lower edge of that stated range; the PR also gives a roughly 280-minute/day lower-bound scenario.","headline":true,"additive":false,"overlapGroup":"cmux-nightly-capacity-2026-09"},{"recordId":"cmux-13810","repository":"manaflow-ai/cmux","title":"Measure CI waste on a schedule","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13810","id":"unchanged-head-repeat-runs","dimension":"observability","metric":"runs-beyond-first-on-unchanged-head","direction":"baseline","value":9763,"unit":"runs/window","recurrence":"one-time","confidence":"observed","measurementWindow":"first real report, approximately 23.1 hours of a 24-hour window","population":"2,424 workflow/head pairs in the fetched run set","note":"The report found 9,763 runs beyond the first for a head that did not change. This is an observation to investigate, not a claim that every repeated run is removable.","headline":true,"additive":false,"overlapGroup":"cmux-control-plane-runs-2026-09"},{"recordId":"cmux-13703","repository":"manaflow-ai/cmux","title":"Remove full-history checkout and unconditional Zig setup from iOS uploads","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13703","id":"ios-upload-setup-savings","dimension":"compute","metric":"macos-minutes-avoided-per-upload","direction":"reduction","value":16,"unit":"runner-minutes/event","recurrence":"per-event","confidence":"derived","measurementWindow":"real upload baseline plus local shallow-checkout timing","population":"iOS TestFlight/App Store upload runs when the pinned GhosttyKit prebuilt exists","note":"The PR states about 16–17 macOS minutes saved per upload. This record stores the lower edge of that range.","headline":true,"additive":false,"overlapGroup":"cmux-ios-upload-capacity-2026-09"},{"recordId":"cmux-13652","repository":"manaflow-ai/cmux","title":"Measure CMUX macOS CI capacity and replacement-cost exposure","evidence":"https://github.com/teamleaderleo/scrapbook/blob/main/work/impact/sources/cmux-actions-90d-2026-09-23.json","id":"warp-replacement-monthly-floor-90d","dimension":"replacement-cost","metric":"warp-list-price-equivalent-floor-per-30-day-month","direction":"baseline","value":52437.36,"unit":"USD/30-day-month","recurrence":"none","confidence":"derived","measurementWindow":"90-day Actions export; Warp price checked 2026-09-23","population":"All 1,966,401 self-hosted macOS minutes priced at WarpBuild's cheapest documented 6-vCPU macOS rate","note":"Floor valuation only: 655,467 equivalent minutes per 30 days at $0.08/min. No 12-vCPU premium is applied. This is replacement-cost exposure, not a claim that every minute was actually billed to Warp or saved by one contributor.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-90d-2026-09"},{"recordId":"cmux-13908","repository":"manaflow-ai/cmux","title":"Compile the E2E test product once and reuse it on reruns","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13908","id":"rerun-compile-avoided","dimension":"critical-path","metric":"compile-minutes-avoided-per-failed-dispatch-rerun","direction":"reduction","value":15.3,"unit":"minutes/event","recurrence":"per-event","confidence":"derived","measurementWindow":"60 dispatches over about 8 hours","population":"Reruns of a failed focused E2E dispatch while the retained product remains available","note":"Fresh dispatches do not claim lower total runner minutes; this figure is the measured compile portion removed from a rerun by the new job split.","headline":true,"additive":false,"overlapGroup":null},{"recordId":"cmux-13962","repository":"manaflow-ai/cmux","title":"Report recorded failures even when app-host results are incomplete","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13962","id":"hidden-recorded-failures","dimension":"reliability","metric":"recorded-failures-hidden-in-observed-run","direction":"reliability","value":148,"unit":"failures/run","recurrence":"one-time","confidence":"measured","measurementWindow":"one referenced full-suite run","population":"Recorded XCTest and Swift Testing failures in the incomplete app-host shards","note":"The fix changes diagnostics, not the fail-closed verdict: incomplete runs remain failed while recorded failures are now named.","headline":true,"additive":false,"overlapGroup":null},{"recordId":"glaeda-1133","repository":"teamleaderleo/glaeda","title":"Reclaim eligible linked worktrees with --apply","evidence":"https://redirect.github.com/teamleaderleo/glaeda/pull/1133#issuecomment-5826944522","id":"worktrees-reclaimed-backlog","dimension":"storage","metric":"idle linked worktrees removed in the first apply runs","direction":"reduction","value":248,"unit":"worktrees","recurrence":"one-time","confidence":"measured","measurementWindow":"three apply runs, 2026-09-23 03:00 to 03:15 EDT","population":"Git repositories under ~/Projects on Air Blue and Big Red","note":"Exact count from receipts; each reclaim is recoverable from the recorded commit and branch. Free space rose by 11.4 GiB (sum of per-repository before and after readings, so other host activity can move it).","headline":true,"additive":false,"overlapGroup":"glaeda-worktree-reclaim"},{"recordId":"glaeda-1140","repository":"teamleaderleo/glaeda","title":"Add glaeda-disk: where the disk went, and reclaim idle build output","evidence":"https://redirect.github.com/teamleaderleo/glaeda/pull/1140#issuecomment-5826944345","id":"air-blue-idle-output-deleted","dimension":"storage","metric":"idle build output deleted by glaeda-disk on Air Blue","direction":"reduction","value":52.8,"unit":"GiB","recurrence":"one-time","confidence":"measured","measurementWindow":"32 receipt lines, 2026-09-23 11:57 to 2026-09-24 18:50 EDT","population":"Air Blue (Leo's MacBook Air)","note":"Gross bytes deleted, not sustained free space: DerivedData and scratch regrow when the next build or session needs them. 35.5 GiB of it came after merge; the other 17.3 GiB was deleted by the PR's own build before merge. The minis run the same pressure job but have not been under pressure (194 to 270 GiB free against a 69.1 GiB threshold), so there are no mini receipts.","headline":true,"additive":false,"overlapGroup":"air-blue-disk-reclaim"},{"recordId":"cmux-13797","repository":"manaflow-ai/cmux","title":"Seed the test compilation cache from one clean build","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13797","id":"compilation-seed-download-reduction","dimension":"cache-transfer","metric":"compilation-seed-download-reduction-per-pr","direction":"reduction","value":1.5,"unit":"GiB/run","recurrence":"per-run","confidence":"measured","measurementWindow":"scheduled seed comparison","population":"pull requests restoring the shared test compilation-cache seed","note":"The clean seed is about 3.5 GiB instead of roughly 5.0 GiB, so every pull request restores about 1.5 GiB less.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"},{"recordId":"cmux-13644","repository":"manaflow-ai/cmux","title":"Debounce macOS admission on superseded pull-request pushes","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13644","id":"cancelled-macos-minutes-avoided","dimension":"compute","metric":"runner-minutes-avoided-per-day","direction":"reduction","value":590,"unit":"runner-minutes/day","recurrence":"daily","confidence":"derived","measurementWindow":"24-hour replay","population":"Superseded pull-request runs that would otherwise start macOS work","note":"Replay estimate at the default 180-second delay; the same analysis measured a 1.4-minute median increase to the green signal.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-2026-09"},{"recordId":"cmux-13658","repository":"manaflow-ai/cmux","title":"Route pull-request macOS jobs to sponsored capacity","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13658","id":"paid-capacity-shift","dimension":"paid-compute","metric":"runner-minutes-shifted-from-paid-capacity-per-day","direction":"shift","value":5900,"unit":"runner-minutes/day","recurrence":"daily","confidence":"measured","measurementWindow":"24 hours","population":"Pull-request macOS overflow work in the sampled CMUX CI day","note":"This moves workload onto sponsored capacity; it does not claim the compute disappeared.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-2026-09"},{"recordId":"cmux-13645","repository":"manaflow-ai/cmux","title":"Trigger the persistent Mac router once per CI run","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13645","id":"router-runs-avoided","dimension":"control-plane","metric":"router-runs-avoided-per-day","direction":"reduction","value":5611,"unit":"runs/day","recurrence":"daily","confidence":"derived","measurementWindow":"24-hour measured baseline","population":"persistent-macos-router workflow invocations","note":"The measured baseline was about 6,864 router runs/day; the once-per-CI-run trigger was expected to produce about 1,253/day, an 82% reduction.","headline":true,"additive":false,"overlapGroup":"cmux-control-plane-runs-2026-09"},{"recordId":"cmux-13754","repository":"manaflow-ai/cmux","title":"Seed compilation cache on the runner pull requests restore from","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13754","id":"cold-pr-compile-baseline","dimension":"critical-path","metric":"cold-compile-minutes-per-affected-pr","direction":"baseline","value":19.9,"unit":"minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"referenced pull-request run","population":"Pull-request admission before the cache producer and consumer runner layouts matched","note":"The PR explicitly left warm compile time unmeasured until the next scheduled seed.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"},{"recordId":"cmux-13132","repository":"manaflow-ai/cmux","title":"Move Computer Use runtime out of the 3,700-file app module","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13132","id":"process-identity-incremental-build-reduction","dimension":"developer-loop","metric":"warm-incremental-build-time-reduction","direction":"reduction","value":25.6,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same-machine warm incremental builds","population":"comment-only edit to AgentPIDProcessIdentity.swift before and after moving it from the app target to CmuxFoundation","note":"The measured build fell from 38.9 seconds to 13.3 seconds and compiled zero app-target Swift files after the move.","headline":true,"additive":false,"overlapGroup":"cmux-local-build-loop-2026-09"},{"recordId":"cmux-13131","repository":"manaflow-ai/cmux","title":"Reuse one warm DerivedData across development tags","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13131","id":"same-commit-warm-derived-data","dimension":"developer-loop","metric":"same-commit-tag-build-time-reduction","direction":"reduction","value":917.3,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same-machine build matrix","population":"new development tag at the same commit after a warm build","note":"Fresh DerivedData took 953 seconds; a new tag using the already-warm shared DerivedData took 35.7 seconds.","headline":true,"additive":false,"overlapGroup":"cmux-local-build-loop-2026-09"},{"recordId":"cmux-13160","repository":"manaflow-ai/cmux","title":"Stop pull requests from evicting shared cache seeds","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13160","id":"github-cache-churn-baseline","dimension":"cache","metric":"github-cache-occupancy-observed","direction":"baseline","value":37.5,"unit":"GB","recurrence":"one-time","confidence":"measured","measurementWindow":"cache inventory at 2026-09-20 05:55 UTC","population":"27 GitHub cache entries visible to the repository","note":"The oldest macOS entry was only five minutes old and the shared Debug compilation seed was already missing. Pull-request jobs had 14 save+restore cache steps and wrote 1–3 GB PR-scoped entries.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"},{"recordId":"cmux-13060","repository":"manaflow-ai/cmux","title":"Seed the Debug compilation cache from main","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13060","id":"cold-admission-baseline","dimension":"critical-path","metric":"cold-compile-seconds-per-admission-run","direction":"baseline","value":800,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"referenced compile-admission run","population":"macOS compile admission from an empty compilation cache","note":"The PR deliberately did not quote an unmeasured CI savings number before the scheduled seed had run.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"},{"recordId":"cmux-13097","repository":"manaflow-ai/cmux","title":"Parallelize slow and history-dependent CI guards","evidence":"https://redirect.github.com/manaflow-ai/cmux/pull/13378","id":"guard-critical-path-reduction","dimension":"critical-path","metric":"linux-guard-critical-path-reduction","direction":"reduction","value":261.2,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"accepted before/after workflow runs","population":"Linux guard path gating macOS admission","note":"Measured guard wall time fell from 385.5 seconds to 124.3 seconds, about a 68% reduction.","headline":true,"additive":false,"overlapGroup":"cmux-linux-guard-latency-2026-09"}],"additiveTotals":[],"accountingRule":"Only claims explicitly marked additive are summed. Baselines, capacity shifts, overlapping savings, and projections stay separate."},"records":[{"id":"cmux-14314","repository":"manaflow-ai/cmux","kind":"pull-request","number":14314,"title":"Start macOS compile admission beside the fast Linux jobs","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","critical-path"],"summary":"Compile admission waited for the static stage and the admission gate before starting. It now starts right after change detection, with a hold step that keeps consumers behind the fast Linux gate.","evidence":{"label":"CMUX PR 14314 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14314#issuecomment-5825460838","basis":"Measured after merge from GitHub Actions job and step timings (read-only), posted on the PR: run start to admission job created, median 160.5 s (n=62, 09-24 20:39Z to 09-25 00:26Z) before, 39.5 s (n=16, 00:30Z to 01:39Z) after; p90 236 s to 77.5 s. Examples 36056381503, 36077835830."},"claims":[{"id":"admission-gate-wait","dimension":"critical-path","metric":"run start to compile admission job created","direction":"reduction","value":121,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=62 before, n=16 after, 2026-09-24/25","population":"pull_request ci.yml runs, first attempts","note":"160.5 s to 39.5 s median; p90 236 s to 77.5 s. One late-night hour after merge; re-measure on a weekday. Job start and compile start also moved (282 s to 60 s, 430 s to 242 s), but pool routing changes merged in the same hour confound those.","headline":true,"additive":false,"overlapGroup":"cmux-admission-start"}],"sourcePath":"work/impact/records/cmux-14314.md"},{"id":"cmux-14318","repository":"manaflow-ai/cmux","kind":"pull-request","number":14318,"title":"ci: place each PR macOS job on a free owned mini, overflow the rest","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","critical-path","runner-routing"],"summary":"PR macOS jobs now go to the owned Mac minis first, one job at a time, and overflow to Blacksmith only when no mini is free. This record covers the routing series: owned pools first (cmux#14237), per-job placement (this PR), rolling over to the next pool when one is full (cmux#14323), overflowing to macOS 15 sooner (cmux#14319), and one fleet retry for a refused owned job (cmux#14312, cmux#14325). In the first night after the series landed, about half of PR compile admissions were placed on a mini within seconds.","evidence":{"label":"CMUX PR 14318 measurement comment","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14318#issuecomment-5826976943","basis":"PR comment: attempt-1 'macos / macOS compile admission' jobs from ci.yml pull_request runs, wait = job started_at minus job created_at. 09-24 03:46Z to 14:59Z: n=223, 0% on owned minis, median 120 s, p90 1535 s. 09-24 12Z to 14Z (after cmux#14205): n=38, median 22.5 s, p90 323 s. 09-25 01:13Z to 04:37Z: n=53, 49% on owned minis, median 3 s, p90 229 s; owned-placed median 2 s, p90 3 s (n=26); Blacksmith-placed median 48 s, p90 404 s (n=27)."},"claims":[{"id":"pr-admission-owned-share","dimension":"paid-compute","metric":"share of attempt-1 PR compile admissions placed on owned Mac minis","direction":"shift","value":49,"unit":"percent of admissions","recurrence":"per-event","confidence":"measured","measurementWindow":"before 2026-09-24 03:46Z to 14:59Z (n=223, 0 owned), after 2026-09-25 01:13Z to 04:37Z (n=53, 26 owned)","population":"attempt-1 PR compile admission jobs in ci.yml","note":"0% to 49%, over one 3.5-hour window of US night. Owned-placed admissions waited 2 s median and 3 s p90 for a runner; the other 27 overflowed to Blacksmith. Of the 26 owned attempt-1 jobs, 5 succeeded, 11 failed and 10 were cancelled (mostly superseded pushes). The failures mix real build failures with owned-runner refusals, which the retry path (cmux#14312, cmux#14325) re-runs as attempt 2; this measurement keeps attempt 1 only, so a refused job counts as owned and its refusal plus re-queue time is not in the wait. Runner speed after start is not compared here.","headline":true,"additive":false,"overlapGroup":"cmux-pr-routing"},{"id":"pr-admission-queue-wait","dimension":"critical-path","metric":"PR compile admission queue wait (job created to started), median","direction":"reduction","value":19.5,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 12:00Z to 14:59Z (n=38), after 2026-09-25 01:13Z to 04:37Z (n=53)","population":"attempt-1 PR compile admission jobs in ci.yml","note":"22.5 s to 3 s median; p90 323 s to 229 s; mean 138 s to 70 s. The baseline is conservative: it starts after cmux#14205 (least-queued Blacksmith pool, 12:13Z) and had about half the macOS job demand of the after window (31 to 41 vs 55 to 89 jobs per hour). At matched demand before cmux#14205 the median was 312 s (n=159), but most of that gap belongs to cmux#14205 and a single capped 6vcpu pool, so it is not claimed here. The after window is 3.5 hours of US night, and owned slots were raised again at 04:32Z. Reruns are excluded because their job created_at can follow started_at. Attempt 1 only: for an owned job that was refused and re-run, the extra refusal and re-queue time is not counted, which flatters the after median.","headline":false,"additive":false,"overlapGroup":"cmux-pr-routing"}],"sourcePath":"work/impact/records/cmux-14318.md"},{"id":"cmux-14333","repository":"manaflow-ai/cmux","kind":"pull-request","number":14333,"title":"Stop portal lifecycle tests waiting 12 s each on login(1)","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","testing"],"summary":"TerminalWindowPortalLifecycleTests freed each terminal surface milliseconds after it spawned login(1), which still ignores SIGHUP while starting, so every close waited out Ghostty's full 12 s SIGHUP grace. tearDown now SIGKILLs the processes on the surface's PTY first. The suite went from about 174 s to about 6 s per CI run.","evidence":{"label":"CMUX PR 14333 CI comment","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14333#issuecomment-5827083427","basis":"Comment with the table: ci.yml 'app-host unit tests (1/7)' jobs, suite time summed from the test case durations in each job log. Median 173.6 s (range 127.2 to 197.2, 10 to 16 of 37 tests over 10 s) across 41 runs from 2026-09-24 15:16Z to 2026-09-25 03:29Z; 4.5 s and 7.3 s (no test over 10 s) in the two runs after the 03:54Z merge (36092666273 on main, 36094056367). Shard 1 job 581 s median before, 319 s and 360 s after."},"claims":[{"id":"portal-suite-duration","dimension":"compute","metric":"TerminalWindowPortalLifecycleTests suite time per CI run","direction":"reduction","value":166,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"before 2026-09-24 15:16Z to 2026-09-25 03:29Z (n=41, median 173.6 s), after 2026-09-25 04:00Z to 04:21Z (n=2, 4.5 s and 7.3 s)","population":"ci.yml app-host unit test shard 1 jobs","note":"173.6 s median minus the larger after value (7.3 s). Only two post-merge runs existed at recording time, so this stays observed. Shard 1's job dropped more than the suite did (581 s median to 319 s and 360 s), so part of that drop comes from other merges and is not claimed. Shard 4 was already the longest app-host job, so this saves shard 1 runner time rather than shortening the app-host lane.","headline":true,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14333.md"},{"id":"cmux-14334","repository":"manaflow-ai/cmux","kind":"pull-request","number":14334,"title":"Stop the CLI no-socket regression waiting 90 s for an app it never starts","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","testing","compute"],"summary":"A CLI regression test deliberately starts no app, yet two of its cases waited out the CLI's full 45 s restore-socket startup timeout before asserting 'still opening'. The CLI now accepts an environment override that can only shorten that timeout, and the test sets it to 1 s. The test went from about 90 s to about 2.4 s in CI.","evidence":{"label":"CMUX PR 14334 CI comment","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14334#issuecomment-5827083266","basis":"Comment with the table: ci.yml 'app-host unit tests (4/7)' jobs, test times read from job logs. testLaunchCapableCommandsReachTheirDispatchPathWithoutLiveImplicitSocket median 90.4 s (n=28, 2026-09-24 16:29Z to 2026-09-25 01:03Z) before, 2.4 s (n=5, 01:16Z to 02:43Z) after; shard 4 'Run unit tests' step median 331 s to 263 s; shard 4 job median 653 s to 590 s. Run ids listed in the comment."},"claims":[{"id":"no-socket-test-duration","dimension":"compute","metric":"testLaunchCapableCommandsReachTheirDispatchPathWithoutLiveImplicitSocket duration","direction":"reduction","value":88,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 16:29Z to 2026-09-25 01:03Z (n=28), after 2026-09-25 01:16Z to 02:43Z (n=5)","population":"ci.yml app-host unit test shard 4 jobs that ran the full CMUXCLIErrorOutputRegressionTests suite","note":"90.4 s to 2.4 s median; every before run fell in 90.2 to 90.6 s and every after run in 2.35 to 2.54 s. Six runs that ran the test in under 1 s (a different selection), three on each side, are excluded. The test runs inside the parallel 'Run unit tests' step, so its time is not all wall time.","headline":true,"additive":false,"overlapGroup":"cmux-app-host-shard4-duration"},{"id":"shard4-unit-step","dimension":"compute","metric":"app-host shard 4 'Run unit tests' step","direction":"reduction","value":68,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same runs: 331 s median (288 to 554) before, 263 s median (254 to 305) after","population":"ci.yml app-host unit test shard 4 jobs that ran the full CMUXCLIErrorOutputRegressionTests suite","note":"Upper bound on the lane effect, recorded as runner time. Shard 4 was usually the longest app-host job (median 653 s before, 590 s after), but shard 1 was close (job median 581 s) and ran longer than shard 4 in some after runs (run 36085886212: 653 s against 628 s), so the app-host lane's wall time fell by less than this step until cmux#14333 shortened shard 1. Other merges landed in the same windows and the after sample is five runs, so not all of the step drop is attributable; the per-test drop is. The PR body placed the test in the 'Run CLI no-socket regressions' step, but it runs in 'Run unit tests'; the no-socket step stayed at 218 to 252 s.","headline":false,"additive":false,"overlapGroup":"cmux-app-host-shard4-duration"}],"sourcePath":"work/impact/records/cmux-14334.md"},{"id":"cmux-14346","repository":"manaflow-ai/cmux","kind":"pull-request","number":14346,"title":"Replay input times onto an owned Mac's kept DerivedData","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"An owned Mac kept its DerivedData between compile admissions, but every job copied a fresh source tree into the canonical root, so swift-driver saw every file as new and recompiled the whole cmux module. Each owned job now records its inputs' content digests and times, and the next job replays them onto byte-identical files after adopting the kept build.","evidence":{"label":"CMUX PR 14346 measurement comment","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14346#issuecomment-5826980830","basis":"Warm owned compile admissions from ci.yml pull_request runs, deduplicated across rerun attempts, PR branch excluded. Before (09-24 23:15Z to 09-25 02:31Z): 0 of 11 replayed, median 2615 cmux SwiftCompile tasks, 9 of 11 whole-module rebuilds, compile step median 384 s (n=9). After (09-25 03:18Z to 04:37Z, before cmux#14375): 15 of 15 replayed, median 1236 tasks, 7 of 15 whole-module rebuilds, compile step median 380 s (n=13)."},"claims":[{"id":"warm-owned-whole-module-rebuilds","dimension":"compute","metric":"share of warm owned compile admissions that rebuilt the whole cmux module","direction":"reduction","value":35,"unit":"percentage points","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 23:15Z to 2026-09-25 02:31Z (n=11), after 2026-09-25 03:18Z to 04:37Z (n=15)","population":"Warm compile admissions on owned glaeda Macs (kept DerivedData adopted), jobs with an Xcode timing summary","note":"82% (9 of 11) to 47% (7 of 15). The task counts are bimodal, so the median (2615 to 1236 tasks) sits on the boundary job of the lower cluster and one more whole-module job would move it back near 2480; it is not claimed. Jobs 7 to 52 changed inputs from the kept build compiled 1 to 177 tasks; jobs 71 to 187 inputs away still rebuilt the module, because the kept build is the previous PR's, not the merge base. The PR's stamp bump made each Mac's first job after merge cold (18 of 35 owned admissions cold after, 13 of 25 before), a one-time cost not netted here.","headline":true,"additive":false,"overlapGroup":"cmux-owned-warm-admission"},{"id":"warm-owned-compile-time","dimension":"critical-path","metric":"warm owned Compile app-host test product step, median","direction":"baseline","value":384,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 2026-09-24 23:15Z to 2026-09-25 02:31Z (n=9 compile successes); after 380 s (n=13)","population":"Warm compile admissions on owned glaeda Macs","note":"Null result for time: 384 s median before, 380 s after. p90 rose from 396 s to 836 s on two contended cmux13s jobs (836 s, 1639 s), unrelated to replay. Jobs that avoided the whole-module rebuild compiled in 91 to 241 s. The PR's expected drop to 50 to 100 s did not hold at the median.","headline":false,"additive":false,"overlapGroup":"cmux-owned-warm-admission"}],"sourcePath":"work/impact/records/cmux-14346.md"},{"id":"cmux-14349","repository":"manaflow-ai/cmux","kind":"pull-request","number":14349,"title":"Build cmuxTests without the compilation cache so it rebuilds incrementally","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"Under the compilation cache, every cmuxTests build rewrote a chained bridging header that all 1,073 test inputs depend on, so a one-test-file edit recompiled the whole test target. The CI compile script now turns the cache off for cmuxTests only, and swift-driver's incremental build recompiles just the edited file. The app and every package keep the cache.","evidence":{"label":"CMUX PR 14349","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14349","basis":"Canary cmux#14249, run 36081880621 (job 107905530013, blacksmith-12vcpu-macos-26, Xcode 26.6), cmux-unit rebuilds over the admission's DerivedData. One-test-file body edit: cache on, cmuxTests 139 s span with SwiftCompile 43 batched tasks at 1,355.3 CPU s (whole rebuild 156 s); a second cache-on series with incremental remarks gave 129 s, 1,299.1 CPU s, and 1,073 inputs invalidated by 'Newer: cmuxTests-...-ChainedBridgingHeader.h'. Cache off for cmuxTests: 31 s span, SwiftCompile 1 task at 0.6 s (whole rebuild 45 s). Read from the job summary and the canary-cmuxtests-probe-1 artifact."},"claims":[{"id":"cmuxtests-test-edit-wall-time","dimension":"critical-path","metric":"cmuxTests build time after a one-test-file body edit","direction":"reduction","value":108,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36081880621, 2026-09-25 01:29Z to 02:08Z: 139 s with the cache on (129 s in a second series), 31 s with it off for cmuxTests","population":"Incremental cmux-unit rebuilds over admission DerivedData on blacksmith-12vcpu-macos-26, Xcode 26.6","note":"One canary run, one edit per mode. Uses 139 s minus 31 s; the second cache-on series gives 98 s. Only pays off when an admission rebuilds on adopted DerivedData. Adding a top-level declaration in the app still recompiles every cmuxTests file with or without the cache (117 s in the same run). Of the remaining 31 s, 25.7 s was cmuxTests' own emit-module job, which cmux#14364 then removed.","headline":true,"additive":false,"overlapGroup":"cmux-cmuxtests-rebuild"},{"id":"cmuxtests-test-edit-cpu","dimension":"compute","metric":"cmuxTests SwiftCompile CPU time after a one-test-file body edit","direction":"reduction","value":1354,"unit":"cpu-seconds/build","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36081880621: SwiftCompile 43 tasks at 1,355.3 s with the cache on, 1 task at 0.6 s with it off for cmuxTests","population":"Incremental cmux-unit rebuilds over admission DerivedData on blacksmith-12vcpu-macos-26, Xcode 26.6","note":"Xcode Build Timing Summary, summed task time across the 12 vCPU runner, not wall time. Same caveats as the wall-time claim.","headline":false,"additive":false,"overlapGroup":"cmux-cmuxtests-rebuild"}],"sourcePath":"work/impact/records/cmux-14349.md"},{"id":"cmux-14359","repository":"manaflow-ai/cmux","kind":"pull-request","number":14359,"title":"Build the app without the compilation cache before Xcode 26.6","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"On the macOS 15 pool (Xcode 26.3), a body-only edit to one app file recompiled every file in the cmux target. Under the compilation cache, Xcode 26.3 rewrites the app's ChainedBridgingHeader.h and bridging PCH on every build with identical bytes and a newer mtime, and the driver then invalidates all 2,615 inputs. The compile script now turns the cache off for the cmux target when Xcode is older than 26.6, and the same edit compiles 2 tasks.","evidence":{"label":"CMUX PR 14359","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14359","basis":"Canary cmux#14351 on blacksmith-6vcpu-macos-15, Xcode 26.3, body-only edits to Sources/Sidebar/GPUSpinner.swift over the admission's DerivedData. Run 36086596738, same job: cache on, cmux-unit rebuilds of 495 s and 448 s with 5,219 and 5,224 cmux SwiftCompile tasks; cache off for cmux, 41 s and 41 s with 2 tasks each. Run 36090019330, the fix as merged through the script: 0 cached Swift jobs for cmux in admission, then 2 tasks per edit in 72 s and 44 s."},"claims":[{"id":"macos15-app-body-edit-recompiles","dimension":"compute","metric":"cmux target SwiftCompile tasks for a body-only edit to one app file","direction":"reduction","value":5217,"unit":"tasks/build","recurrence":"per-run","confidence":"observed","measurementWindow":"two edits each way in run 36086596738 (5,219 and 5,224 before, 2 and 2 after), confirmed by two edits in run 36090019330 (2 and 2)","population":"Incremental cmux builds on the Blacksmith macOS 15 pool (Xcode 26.3)","note":"Two edits each way, one runner each. Only pays off in PR admission when it adopts DerivedData to rebuild on: the macOS 15 seeds (cmux#14315, cmux#14358) merged the same morning, and seeds built before this fix used the cache for cmux, so the first adoption after a stale seed can still rebuild the app fully. Xcode 26.6 pools were already incremental and are unchanged.","headline":true,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"},{"id":"macos15-app-body-edit-wall-time","dimension":"critical-path","metric":"cmux-unit build-for-testing wall time after a body-only app edit","direction":"reduction","value":407,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36086596738: 448 s and 495 s with the cache on, 41 s and 41 s with it off for cmux","population":"Incremental cmux builds on the Blacksmith macOS 15 pool (Xcode 26.3)","note":"Uses the smaller before value (448 s minus 41 s). The fix as merged measured 72 s and 44 s in run 36090019330. Same caveat as the task claim: it needs adopted DerivedData to rebuild on.","headline":false,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"}],"sourcePath":"work/impact/records/cmux-14359.md"},{"id":"cmux-14360","repository":"manaflow-ai/cmux","kind":"pull-request","number":14360,"title":"Run the CLI pipe regressions in compile admission, drop the separate lane","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","compute","capacity"],"summary":"The cli-pipe-regressions lane took its own macOS runner to rebuild the Debug CLI that compile admission already builds, then ran three Python tests and two CmuxFoundation filters. Admission's CLI smoke step now runs the two tests it lacked against its own binary, the CmuxFoundation filters stay covered by swift-package-tests, and the lane is deleted.","evidence":{"label":"CMUX PR 14360 measurement comment","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14360#issuecomment-5826986283","basis":"PR comment: in ci.yml PR runs created 2026-09-25 00:46Z to 03:58Z (137 runs), the cli-pipe-regressions job ran 41 times, median 197 s, mean 382 s, 261 runner-minutes total, median queue 14 s. Admission's 'Run early CLI binary smoke checks' step went from median 9 s (n=54, same window) to 11 s (n=15, runs created 03:58Z to 04:47Z). The lane finished last among non-status jobs in 3 of 41 runs."},"claims":[{"id":"cli-lane-runner-time","dimension":"compute","metric":"macOS runner time of the removed cli-pipe-regressions job","direction":"reduction","value":3.28,"unit":"runner-minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"lane jobs n=41, 2026-09-25 00:46Z to 03:58Z; smoke step n=54 before, n=15 after (03:58Z to 04:47Z)","population":"ci.yml pull_request runs that routed to the CLI lane (41 of 137 in the window); 18 on glaeda minis, 23 on Blacksmith macOS","note":"Median lane job 197 s, gross, on runs that routed to the CLI lane. The mean lane job was 382 s, so the median is the conservative figure. Averaged over all 137 PR runs in the window this is about 59 s per run. This removal also ends the cli-pipe-resolve-time claim in cmux-14257.","headline":true,"additive":false,"overlapGroup":"cmux-cli-lane"},{"id":"admission-smoke-step-added","dimension":"critical-path","metric":"admission 'Run early CLI binary smoke checks' step, median","direction":"shift","value":2,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=54 before (2026-09-25 00:46Z to 03:58Z), n=15 after (03:58Z to 04:47Z)","population":"every ci.yml PR compile admission","note":"9 s to 11 s median: the cost of running the two moved tests in admission, paid by every admission, including runs that never used the CLI lane. Admission is on the critical path, so this is a small increase there.","headline":false,"additive":false,"overlapGroup":"cmux-cli-lane"},{"id":"cli-lane-window-baseline","dimension":"compute","metric":"total cli-pipe-regressions runner time before removal","direction":"baseline","value":261,"unit":"runner-minutes/window","recurrence":"none","confidence":"measured","measurementWindow":"ci.yml PR runs created 2026-09-25 00:46Z to 03:58Z (137 runs, 41 lane jobs)","population":"cli-pipe-regressions jobs on PR runs","note":"Step totals: CLI build 88 min, Restore Swift packages 85 min, resolve 32 min, tests about 23 min, the rest setup and checkout. Critical-path effect is small: the lane was the last job to finish in only 3 of 41 runs, so no time claim is made.","headline":false,"additive":false,"overlapGroup":"cmux-cli-lane"}],"sourcePath":"work/impact/records/cmux-14360.md"},{"id":"cmux-14364","repository":"manaflow-ai/cmux","kind":"pull-request","number":14364,"title":"Build cmuxTests without emitting its Swift module","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["ci","build-performance"],"summary":"After cmux#14349, most of a one-test-file cmuxTests rebuild was the target's own emit-module job, a serial type-check of every test file that produced a module nothing imports. CI now builds cmuxTests with the standalone Swift driver and -no-emit-module-separately, so no module is planned, and the same edit rebuilds cmuxTests in about 4 s with the same tests in the bundle.","evidence":{"label":"CMUX PR 14364","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14364","basis":"Canary cmux#14352, run 36089490735 attempt 2 (job 107930437510, blacksmith-12vcpu-macos-26, Xcode 26.6), cmux-unit rebuilds over the admission's DerivedData without stats flags, read from the job summary and the canary-emit-module-probe-2 artifact. Two one-test-file body edits each way: main settings, cmuxTests spans 35.3 s and 31.6 s (SwiftEmitModule 29.8 s and 26.4 s; whole rebuilds 55 s and 47 s); no-module settings, 4.2 s and 4.1 s (whole rebuilds 19 s and 19 s). Both bundles enumerated 11,758 tests with the same __swift5_tests size (0x44140) and 332 ObjC test classes. The profiling run 36087174643 used fine-grained timers and was not used for the numbers."},"claims":[{"id":"cmuxtests-test-edit-wall-time","dimension":"critical-path","metric":"cmuxTests build time after a one-test-file body edit","direction":"reduction","value":27.4,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"run 36089490735, 2026-09-25 03:25Z to 03:40Z: 35.3 s and 31.6 s with the module emitted, 4.2 s and 4.1 s without","population":"Incremental cmux-unit rebuilds over admission DerivedData on blacksmith-12vcpu-macos-26, Xcode 26.6","note":"Uses the smaller before (31.6 s) minus the larger after (4.2 s). The whole cmux-unit rebuild went from 47 to 55 s to 19 s in the same series, which contains this saving, so it is not claimed separately. A standalone-driver series that still emitted the module took 36.4 s, so the gain comes from dropping the module. Only pays off when an admission rebuilds on adopted DerivedData, and the first admissions after the merge rebuild cmuxTests once because its arguments changed. Too few post-merge admissions existed at recording time (29 minutes after the merge) to measure it there. cmux#14378 (merged 04:40Z) applies the same settings to the local scripts/test-unit.sh.","headline":true,"additive":false,"overlapGroup":"cmux-cmuxtests-rebuild"}],"sourcePath":"work/impact/records/cmux-14364.md"},{"id":"glaeda-1190","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1190,"title":"worktree reclaim: delete local branches whose work has landed (--branches)","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["developer-loop"],"summary":"Removing a worktree kept its branch, so every agent task left a local branch behind. The hourly reclaim now also deletes local branches that are not checked out, are idle, and whose work landed (a merged PR containing the tip, or git alone), with a compare-and-swap delete and a receipt to restore each one.","evidence":{"label":"Glaeda PR 1190","url":"https://redirect.github.com/teamleaderleo/glaeda/pull/1190#issuecomment-5826944682","basis":"PR comment summarizing the hourly LaunchAgent's receipts on Air Blue (~/Library/Logs/glaeda-worktree-reclaim.jsonl): the four runs with a branch pass after merge, ending 2026-09-25 00:15 EDT, deleted 64, 64, 64 and 63 branches (255 total), each hitting the 64-per-run cap; 1 branch was kept because it changed at recheck; circuit breaker never tripped."},"claims":[{"id":"landed-branches-deleted","dimension":"developer-loop","metric":"landed local branches deleted by the hourly reclaim on Air Blue","direction":"reduction","value":255,"unit":"branches","recurrence":"one-time","confidence":"measured","measurementWindow":"first four branch-pass runs after merge, 2026-09-24 20:03 to 2026-09-25 00:15 EDT","population":"Local branches in the Git repositories under ~/Projects on Air Blue","note":"Backlog clearance, capped at 64 per run, and eligible branches remained, so the total keeps rising. Before, these were only removed by a manual git branch -D. Big Red runs the same job; its receipts were not read here.","headline":true,"additive":false,"overlapGroup":"glaeda-worktree-reclaim"}],"sourcePath":"work/impact/records/glaeda-1190.md"},{"id":"glaeda-1192","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1192,"title":"glaeda-fleet-cas: ProcessType Interactive for the node and store agents","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["fleet","caching","reliability"],"summary":"The fleet cache node and store daemons ran at background priority, and under load they failed a large share of store calls. They now run as Interactive.","evidence":{"label":"Glaeda PR 1192","url":"https://github.com/teamleaderleo/glaeda/pull/1192#issuecomment-5824669691","basis":"PR body and comment: daemon priority 4 to 31. A fill through the signing writer on cmux7s went from 187 failed and 543 skipped store calls plus 751 failed writes to 0, 0 and 0; the fixed run pushed 7,706 entries and 25,032 objects (2.3 GB) and published a verified marker."},"claims":[{"id":"failed-cache-writes","dimension":"reliability","metric":"failed fleet cache writes in one fill","direction":"reliability","value":751,"unit":"failures/run","recurrence":"per-run","confidence":"observed","measurementWindow":"one fill before (2026-09-24) and one after (2026-09-25) on cmux7s","population":"Fleet compilation cache fills","note":"751 to 0. Wall time went 684 s to 2,643 s, likely from a 66 GB VM image pull on the same host (unconfirmed); this is a reliability win, not a speed win.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/glaeda-1192.md"},{"id":"glaeda-1209","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1209,"title":"fleet-cas: tailnet addressing, launchd wrapper, upstream error logging","status":"merged","happenedAt":"2026-09-25","recordedAt":"2026-09-25","areas":["fleet","caching","reliability"],"summary":"Cross-machine reads through the deployed fleet cache services never worked: macOS Local Network privacy blocks launchd-started third-party binaries on the headless minis from other hosts' LAN addresses. The store and nodes now use tailnet addresses, and a fresh full cmux app build on cmux11s hit every cacheable task through its deployed node.","evidence":{"label":"Glaeda PR 1209","url":"https://redirect.github.com/teamleaderleo/glaeda/pull/1209#issuecomment-5826438514","basis":"PR body: before, the launchd-started node on cmux8s got \"No route to host\" for the store's LAN address on cmux7s while the same binary from ssh connected. Pre-merge comment: fresh full-app read on cmux11s through its deployed node LaunchAgent, over the tailnet, from the store on cmux7s, commit 2ae26d1c, Xcode 26.6: 4,192 of 4,192 cacheable tasks hit, all 7,706 signed index entries fetched and verified (kv_sig_fail 0, up_errors 0), BUILD SUCCEEDED in 326 s."},"claims":[{"id":"deployed-cross-machine-hits","dimension":"reliability","metric":"cacheable tasks hit through a deployed node reading another mini's store","direction":"reliability","value":4192,"unit":"tasks/build","recurrence":"per-run","confidence":"observed","measurementWindow":"one fresh full-app build on cmux11s before merge, 2026-09-25","population":"Fresh full cmux app builds on Manaflow minis using the deployed fleet-cas node (Xcode 26.6)","note":"Before, the deployed (launchd) node could not reach the store at all, so deployed cross-machine reads served nothing; lab reads in glaeda-1162 and glaeda-1180 ran from ssh, which the block does not affect. The 326 s build time is not claimed as a reduction: there is no same-host, same-Xcode cold baseline for it (756 s in glaeda-1162 was Xcode 26.3).","headline":true,"additive":false,"overlapGroup":"glaeda-fleet-cas-fresh-build"}],"sourcePath":"work/impact/records/glaeda-1209.md"},{"id":"cmux-14089","repository":"manaflow-ai/cmux","kind":"pull-request","number":14089,"title":"Prune expired R2 cache archives, never one a latest pointer names","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","caching","storage"],"summary":"Nothing deleted CI cache archives from R2, so the bucket grew about 35 GiB a day, mostly per-commit Xcode compilation archives. A daily prune now deletes archives past their family's retention (1 day for per-commit families, 30 days for content-keyed ones) and never deletes one a latest pointer names.","evidence":{"label":"CMUX PR 14089","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14089","basis":"Before: the PR body measured 143.4 GiB in the bucket after about five days, 141.4 GiB of it compilation archives. First deleting run 35972452822 deleted 114 archives, 117.2 GiB."},"claims":[{"id":"r2-storage-freed-first-run","dimension":"storage","metric":"R2 cache bytes deleted by the first prune","direction":"reduction","value":117.2,"unit":"GiB","recurrence":"one-time","confidence":"measured","measurementWindow":"run 35972452822, 2026-09-24","population":"manaflow-ai/cmux CI cache bucket","note":"One-time reclaim. The steady-state effect is capping growth at about one day of per-commit archives, not measured here.","headline":true,"additive":false,"overlapGroup":"cmux-r2-cache-storage"},{"id":"r2-storage-growth-baseline","dimension":"storage","metric":"R2 cache growth before pruning","direction":"baseline","value":35,"unit":"GiB/day","recurrence":"daily","confidence":"observed","measurementWindow":"about five days before 2026-09-24","population":"manaflow-ai/cmux CI cache bucket","note":"The PR's stated growth rate (143.4 GiB held after about five days; the PR puts steady growth at about 35 GiB a day, mostly compilation archives).","headline":false,"additive":false,"overlapGroup":"cmux-r2-cache-storage"}],"sourcePath":"work/impact/records/cmux-14089.md"},{"id":"cmux-14174","repository":"manaflow-ai/cmux","kind":"pull-request","number":14174,"title":"Order R2 cache pointers by run ID, so seeder saves take effect","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","caching","reliability"],"summary":"The latest-seed pointer could name an older save than the newest one, so seeders kept adopting a stale seed. Pointers are now ordered by run ID.","evidence":{"label":"CMUX PR 14174 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14174#issuecomment-5825462298","basis":"Measured after merge (read-only), posted on the PR: seeder changed inputs median 451 (n=5, all adopting stale f2abe643) before; 221, 129 and 64 on the next three runs on the same 6 vCPU pool."},"claims":[{"id":"seeder-changed-inputs","dimension":"reliability","metric":"changed inputs when the seeder adopts a seed","direction":"reliability","value":322,"unit":"inputs/run","recurrence":"per-run","confidence":"observed","measurementWindow":"5 runs before, 3 runs after on the same pool, 2026-09-24","population":"seed-derived-data.yml runs on main","note":"451 to 129 median on the same pool. Later runs reached a median of 28, but those also had #14188's faster seed cadence and #14190. Build time on the same pool moved only 845 s to 737 s.","headline":false,"additive":false,"overlapGroup":"cmux-seed-freshness"}],"sourcePath":"work/impact/records/cmux-14174.md"},{"id":"cmux-14184","repository":"manaflow-ai/cmux","kind":"pull-request","number":14184,"title":"Download the admission DerivedData seed while packages resolve","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path","caching"],"summary":"Compile admission downloaded the DerivedData seed after resolving packages. The download now runs during the resolve.","evidence":{"label":"CMUX PR 14184 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14184#issuecomment-5825461750","basis":"Measured after merge from GitHub Actions job and step timings (read-only), posted on the PR: seeded admissions on 6 vCPU macOS 26: package-cache start to compile start, median 148 s (n=63, 06:00Z to 09:10Z) before, 110 s (n=37, 09:15Z to 14:00Z) after; adopt step 46 s to 8 s. Examples jobs 107514602188, 107569667264."},"claims":[{"id":"pre-compile-window","dimension":"critical-path","metric":"package cache start to compile start in seeded admissions","direction":"reduction","value":38,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=63 before, n=37 after, 2026-09-24","population":"Seeded compile admissions on 6 vCPU macOS 26","note":"148 s to 110 s median. From job start the gain is 23 s, because checkout got about 12 s slower in the same period.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14184.md"},{"id":"cmux-14186","repository":"manaflow-ai/cmux","kind":"pull-request","number":14186,"title":"Balance the guard legs the macOS gate waits on","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path"],"summary":"The macOS admission gate waits on the slowest guard leg. This PR rebalanced the guard legs, and #14176 started them beside Fast static checks.","evidence":{"label":"CMUX PR 14186 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14186#issuecomment-5825461500","basis":"Measured after merge from GitHub Actions job and step timings (read-only), posted on the PR: changes-done to admission gate created, median 100 s (n=54, 06:00Z to 08:13Z) before, 79 s (n=39, 09:00Z to 12:00Z) after; slowest guard leg 81 s to 59 s. Covers #14176 too."},"claims":[{"id":"gate-open-time","dimension":"critical-path","metric":"changes done to macOS admission gate created","direction":"reduction","value":21,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=54 before, n=39 after, 2026-09-24","population":"pull_request ci.yml runs, first attempts","note":"100 s to 79 s median shortly after merge; over the whole day 89 s, and nothing measurable from run start because change detection itself got slower. #14176 accounts for about 5 s of it.","headline":false,"additive":false,"overlapGroup":"cmux-admission-start"}],"sourcePath":"work/impact/records/cmux-14186.md"},{"id":"cmux-14187","repository":"manaflow-ai/cmux","kind":"pull-request","number":14187,"title":"Stop counting queue time on cancelled jobs as runner minutes","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","observability"],"summary":"The CI health report counted the queue time of cancelled jobs as runner minutes, inflating the apparent cost of cancelled work. It now counts only minutes a runner actually ran.","evidence":{"label":"CMUX PR 14187","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14187","basis":"PR body, health report window 2026-09-23 20:37 to 09-24 02:37Z: cancelled jobs had 345 real runner minutes against 4,732 queue-only minutes (iroh-v2 client 0 vs 334, cli-pipe-regressions 21 vs 1,022, Claude wrapper regressions 3 vs 728, compile admission 321 vs 2,648)."},"claims":[{"id":"miscounted-queue-minutes","dimension":"observability","metric":"queue-only minutes reported as runner minutes","direction":"reliability","value":4732,"unit":"minutes/window","recurrence":"none","confidence":"measured","measurementWindow":"health report window, 2026-09-23 20:37 to 09-24 02:37Z","population":"Cancelled iroh-v2 client, cli-pipe-regressions, Claude wrapper regressions and compile admission jobs","note":"A reporting correction, not a compute saving: queue-only minutes were about 14 times the real runner minutes (4,732 vs 345), so the old report showed about 15 times the real cost.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14187.md"},{"id":"cmux-14188","repository":"manaflow-ai/cmux","kind":"pull-request","number":14188,"title":"Seed DerivedData on the 12 vCPU macOS 26 pool","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","caching","critical-path"],"summary":"The main-branch seeder queued behind PR jobs on the 6 vCPU pool, so seeds landed long after each push. It now runs on the 12 vCPU pool.","evidence":{"label":"CMUX PR 14188 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14188#issuecomment-5825462059","basis":"Measured after merge from GitHub Actions job and step timings (read-only), posted on the PR: seed job queue wait median 1,342 s (n=6, runs created 06:00Z to 09:01Z) before, 16 s (n=16, 09:12Z to 14:00Z) after; push to seed saved 2,628 s to 740 s. Examples 35971831008 (2,179 s wait), 35979832359 (14 s)."},"claims":[{"id":"seed-queue-wait","dimension":"critical-path","metric":"seeder job queue wait","direction":"reduction","value":1326,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=6 before, n=16 after, 2026-09-24","population":"seed-derived-data.yml runs on main","note":"1,342 s to 16 s median. After 14:00Z the 12 vCPU pool queued too (621 s and 663 s), once PR admissions began preferring it.","headline":true,"additive":false,"overlapGroup":"cmux-seed-freshness"},{"id":"push-to-seed","dimension":"critical-path","metric":"main push to seed saved","direction":"reduction","value":1888,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same runs","population":"seed-derived-data.yml runs on main","note":"2,628 s to 740 s median, including the faster build and #14174 and #14190.","headline":false,"additive":false,"overlapGroup":"cmux-seed-freshness"}],"sourcePath":"work/impact/records/cmux-14188.md"},{"id":"cmux-14199","repository":"manaflow-ai/cmux","kind":"pull-request","number":14199,"title":"One git archive for the trusted router; delete duplicate CI guard tests","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","testing"],"summary":"The trusted CI router copied its 82 base files one git call at a time. It now extracts them with one git archive; the change also deleted duplicate guard tests.","evidence":{"label":"CMUX PR 14199","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14199","basis":"PR body: copying the 82 base files at 6d203e81b6c took 14.6 s before and 0.43 s after, with identical output (diff -r). Measured locally on Big Red's full clone at load average 40 to 100."},"claims":[{"id":"router-base-copy-time","dimension":"critical-path","metric":"trusted-router base file copy","direction":"reduction","value":14.2,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"one local run each on Big Red, 2026-09-24","population":"Trusted CI router","note":"14.6 s to 0.43 s, measured locally on a loaded host with a full clone; a CI shallow checkout will differ and was not measured.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14199.md"},{"id":"cmux-14212","repository":"manaflow-ai/cmux","kind":"pull-request","number":14212,"title":"Build only the CLI product for CLI-only changes","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path"],"summary":"A CLI-routed change built all four app schemes. It now builds only the CLI product and its tests.","evidence":{"label":"CMUX PR 14212 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14212#issuecomment-5825517561","basis":"Measured after merge (read-only), posted on the PR with a correction: the one like-for-like pair, #14325 on 6 vCPU macOS 26 with a seed hit, compiled in 120 s before (job 107891367186, four schemes) and 40 s after (job 107897749996, cli profile); 288 s to 253 s for the whole job."},"claims":[{"id":"cli-profile-compile","dimension":"critical-path","metric":"compile step for CLI-profile admissions","direction":"reduction","value":80,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"one pair, 2026-09-24/25","population":"Admissions routed to the CLI product profile","note":"120 s to 40 s compile on the same PR. The broader 388 s to 36 s medians (n=4 each) compare first admissions against reruns and are not like for like. No true CLI-only PR has run yet.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14212.md"},{"id":"cmux-14226","repository":"manaflow-ai/cmux","kind":"pull-request","number":14226,"title":"Build focused runs from compile admission's DerivedData seed","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","testing","caching"],"summary":"Focused E2E runs compiled the app cold. They now adopt compile admission's DerivedData seed.","evidence":{"label":"CMUX PR 14226 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14226#issuecomment-5825517035","basis":"Measured after merge from GitHub Actions job and step timings (read-only), posted on the PR with a correction: test-e2e.yml build step over all builds, 6 vCPU macOS 26 median 888 s (n=27) before, 435 s (n=16, all seed hits) after; 12 vCPU 553 s (n=15) to 518.5 s (n=22, 14 seed hits at 367.5 s, 8 misses at 675 s). macOS 15 got no seed."},"claims":[{"id":"e2e-build-6vcpu","dimension":"critical-path","metric":"E2E app build step on 6 vCPU macOS 26","direction":"reduction","value":453,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=27 before, n=16 after, 2026-09-24/25","population":"All focused test-e2e.yml builds on 6 vCPU macOS 26","note":"888 s to 435 s median. The before side includes 7 builds that had already adopted main's DerivedData or an early admission seed.","headline":true,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"},{"id":"e2e-build-12vcpu","dimension":"critical-path","metric":"E2E app build step on 12 vCPU macOS 26","direction":"reduction","value":34.5,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"n=15 before, n=22 after, 2026-09-24/25","population":"All focused test-e2e.yml builds on 12 vCPU macOS 26","note":"553 s to 518.5 s median over all builds. A seed hit builds in 367.5 s, but 8 of 22 builds missed with no-seed, cause unknown. macOS 15 has no seed until #14315.","headline":false,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"}],"sourcePath":"work/impact/records/cmux-14226.md"},{"id":"cmux-14229","repository":"manaflow-ai/cmux","kind":"pull-request","number":14229,"title":"Run focused cmuxTests against products CI already compiled","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","testing","developer-loop"],"summary":"Focused E2E runs of cmuxTests always compiled the whole app first. The dispatcher now reuses products PR CI already built, or waits for an in-flight CI compile of the same commit, and recompiles only cmuxTests.","evidence":{"label":"CMUX PR 14229","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14229","basis":"PR body: 82 of 100 sampled test-e2e runs on 2026-09-24 selected only cmuxTests, with 12 to 27 minute build jobs against about 2 minutes of tests. Of 25 distinct app-host-only refs that day, 15 paid for a second compile of code CI had built or was building."},"claims":[{"id":"duplicate-compiles-baseline","dimension":"compute","metric":"app-host-only E2E refs that recompiled code CI already built","direction":"baseline","value":15,"unit":"refs/day","recurrence":"daily","confidence":"measured","measurementWindow":"2026-09-24, 25 distinct refs","population":"Focused cmuxTests E2E dispatches","note":"The 82 sampled cmuxTests-only runs had 12 to 27 minute build jobs; the duplicate refs' own compile times were not listed. No after-count was taken.","headline":false,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"}],"sourcePath":"work/impact/records/cmux-14229.md"},{"id":"cmux-14236","repository":"manaflow-ai/cmux","kind":"pull-request","number":14236,"title":"Compile only the cmuxTests sources a rerun's suites reach","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","testing","developer-loop"],"summary":"App-host test reruns still compiled all 1,091 test-target sources (1,069 cmuxTests files plus 22 app and CLI files) to run one suite. The rerun now prunes the test sources to those the selected suites can reach (1 to 24 of 1,069 files on that day's traffic), with a full-compile fallback.","evidence":{"label":"CMUX PR 14236","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14236","basis":"PR body 'Result' table, same suite and same products: run 36009423131 took 583 s before pruning; run 36016462173, on the PR branch head 2d7207ae with pruning and the cache and fetch fixes, took 371 s. Run 36009423131 also spent 220 s compiling all test sources for a suite whose tests took 0.01 s."},"claims":[{"id":"rerun-wall-time","dimension":"critical-path","metric":"app-host-test-rerun.yml wall time","direction":"reduction","value":212,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"runs 36009423131 (before) and 36016462173 (pre-merge PR branch run), 2026-09-24","population":"Focused cmuxTests reruns","note":"583 s to 371 s. One run each, so indicative only.","headline":false,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"}],"sourcePath":"work/impact/records/cmux-14236.md"},{"id":"cmux-14240","repository":"manaflow-ai/cmux","kind":"pull-request","number":14240,"title":"E2E: wait for an earlier dispatch's compile of the same revision","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","testing"],"summary":"Two test-e2e dispatches of one commit with different selectors ran side by side, each compiling the same product for 12 to 27 minutes. A sibling job now waits for the earlier run's build and adopts its product through the existing lookup.","evidence":{"label":"CMUX PR 14240","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14240","basis":"PR body: from 2026-09-23 to 09-24, 25 of 197 test-e2e runs repeated a commit and pool an earlier run had built, and at least 12 started while the identical compile was still running."},"claims":[{"id":"concurrent-duplicate-compiles-baseline","dimension":"compute","metric":"test-e2e runs that started while an identical compile was running","direction":"baseline","value":12,"unit":"runs/window","recurrence":"none","confidence":"measured","measurementWindow":"2026-09-23 to 2026-09-24, 197 runs","population":"test-e2e.yml dispatches","note":"Each would have cost 12 to 27 minutes of macOS compile. No after-count was taken.","headline":false,"additive":false,"overlapGroup":"cmux-e2e-duplicate-compile"}],"sourcePath":"work/impact/records/cmux-14240.md"},{"id":"cmux-14241","repository":"manaflow-ai/cmux","kind":"pull-request","number":14241,"title":"Replay directory times when adopting a DerivedData seed","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"Seed adoption restored file times but not directory times, so Xcode saw Assets.xcassets as changed on every adoption and reran the asset catalog. Directory times are now recorded with a digest of their entry names and restored when the names match.","evidence":{"label":"CMUX PR 14241","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14241","basis":"PR comment: before, 29 of 29 seeded PR admissions from 11:00Z to 15:00Z on 2026-09-24 reran CompileAssetCatalog and GenerateAssetSymbols. After (seeds that also carry cmux-14262), 0 in 3 of 3 seeded admissions at distance 0 (runs 36041488043, 36040706124, 36040299378). Canary 36017663612: touching only the directory cost a 241 s rebuild against 29 s null."},"claims":[{"id":"asset-catalog-reruns","dimension":"reliability","metric":"seeded admissions that reran the asset catalog","direction":"reliability","value":29,"unit":"runs/window","recurrence":"per-run","confidence":"measured","measurementWindow":"29 admissions before, 3 after, 2026-09-24","population":"Seeded macOS compile admissions","note":"29 of 29 to 0 of 3. The after seeds also carried cmux-14262's checksum-only mode, so the 0 of 3 is both changes combined. The cmux-unit phase did not get faster: 12 vCPU 129 s median to 112 s (n=1); 6 vCPU 199 s median to 207 s and 242 s. cmuxTests still recompiles fully after adoption (cmux#14320).","headline":false,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"}],"sourcePath":"work/impact/records/cmux-14241.md"},{"id":"cmux-14254","repository":"manaflow-ai/cmux","kind":"pull-request","number":14254,"title":"Clone the canonical build root instead of rsyncing it","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path"],"summary":"Compile admission copied the checkout into the canonical build root with rsync. It now uses an APFS clone (cp -cpR).","evidence":{"label":"CMUX PR 14254","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14254#issuecomment-5818649514","basis":"PR comment: canonical-root copy median 10.0 s over 8 PR admissions since 13:00Z on 2026-09-24 (range 8.72 to 22.31 s; runs 36029137471, 36023014704, 36022937624, 36028782656, 36028762508, 36022099083, 36025040765, 36027687674); 2.54 s on run 36029504166 after."},"claims":[{"id":"canonical-root-copy-time","dimension":"critical-path","metric":"admission canonical build root copy","direction":"reduction","value":7.5,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"8-run median before, 1 run after, 2026-09-24","population":"macOS compile admissions","note":"10.0 s median to 2.54 s. The before runs span three pools; the after run was 6vcpu macOS 26.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14254.md"},{"id":"cmux-14255","repository":"manaflow-ai/cmux","kind":"pull-request","number":14255,"title":"Start macOS checkouts from main's git objects","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path"],"summary":"Every macOS job cloned the tested commit at depth 1 into an empty repository, downloading about 129 MB. Main now publishes a git seed, and macOS jobs restore it before checkout so the fetch moves only the delta.","evidence":{"label":"CMUX PR 14255","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14255","basis":"PR body: checkout medians across 80 ci.yml PR runs on 2026-09-24 were 48 s (admission), 34 s (app-host shards), 44 s (swift-package-tests), 35 s (cli-pipe-regressions). PR comment after merge: 3 to 9 s checkout plus 2 to 8 s restore on five jobs, admission 4 s checkout plus 8 s restore (job 107802311019). The comment sums it as about 48 s to 12 s."},"claims":[{"id":"admission-checkout-time","dimension":"critical-path","metric":"compile admission checkout plus git seed restore","direction":"reduction","value":36,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"80-run median before; 1 admission job after","population":"macOS compile admission","note":"48 s median to 12 s (4 s checkout plus 8 s restore). One after-sample per lane.","headline":true,"additive":false,"overlapGroup":"cmux-macos-checkout"},{"id":"checkout-time-baseline","dimension":"compute","metric":"checkout time across four macOS PR lanes","direction":"baseline","value":88,"unit":"minutes/window","recurrence":"none","confidence":"measured","measurementWindow":"80 ci.yml PR runs, 11:00Z onward on 2026-09-24","population":"admission, app-host shards, swift-package-tests, cli-pipe-regressions","note":"35 + 21 + 17 + 15 minutes of checkout in the window.","headline":false,"additive":false,"overlapGroup":"cmux-macos-checkout"}],"sourcePath":"work/impact/records/cmux-14255.md"},{"id":"cmux-14257","repository":"manaflow-ai/cmux","kind":"pull-request","number":14257,"title":"Restore SwiftPM's manifest cache for package resolves","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path","caching"],"summary":"Every admission re-evaluated about 90 Package.swift manifests. SwiftPM keys its manifest cache on the whole process environment, so a plain restore never hit. The resolve now runs under a fixed environment and restores a manifest cache main seeds.","evidence":{"label":"CMUX PR 14257","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14257","basis":"PR comment: admission 'Resolve Swift packages' median 63 s (n=21) before, 17 s (n=12) after on macOS 26 pools; cli-pipe-regressions 48 s (n=13) to 15.5 s (n=8). The macOS 15 pool missed until cmux#14272 seeded it."},"claims":[{"id":"admission-resolve-time","dimension":"critical-path","metric":"admission Resolve Swift packages step","direction":"reduction","value":46,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"before 11:00Z to 16:40Z (n=21), after 17:34Z to 18:40Z (n=12), 2026-09-24","population":"macOS 26 compile admissions","note":"63 s to 17 s median.","headline":true,"additive":false,"overlapGroup":"cmux-swiftpm-resolve"},{"id":"cli-pipe-resolve-time","dimension":"critical-path","metric":"cli-pipe-regressions package resolve step","direction":"reduction","value":32.5,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same windows (n=13 before, n=8 after)","population":"macOS 26 cli-pipe-regressions jobs","note":"48 s to 15.5 s median.","headline":false,"additive":false,"overlapGroup":"cmux-swiftpm-resolve"}],"sourcePath":"work/impact/records/cmux-14257.md"},{"id":"cmux-14261","repository":"manaflow-ai/cmux","kind":"pull-request","number":14261,"title":"Skip a main seed build only when the nearest seed has the same inputs","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","caching","reliability"],"summary":"The seeder skipped a main commit whenever its parent's inputs matched, even when the parent had no seed, leaving main on a stale seed for 80 minutes in the latest case. It now compares against the nearest seeded ancestor.","evidence":{"label":"CMUX PR 14261","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14261","basis":"PR body: of 22 skips in the last 100 seeder runs (2026-09-23 17:00Z to 09-24 17:10Z), 8 left main on a stale seed 2 to 6 commits back. Replaying the new rule over the same 22 skips builds all 8, keeps the 13 correct skips, and builds 1 extra (a05c7e6)."},"claims":[{"id":"stale-seed-skips","dimension":"reliability","metric":"seeder skips that left main on a stale seed","direction":"reliability","value":8,"unit":"runs/window","recurrence":"none","confidence":"derived","measurementWindow":"22 skips in 100 seeder runs, 2026-09-23 17:00Z to 09-24 17:10Z","population":"seed-derived-data.yml runs on main","note":"8 of 22 to 0 of 22 by replaying the new rule on the same skips, not by live runs after merge. One extra, harmless build.","headline":false,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"}],"sourcePath":"work/impact/records/cmux-14261.md"},{"id":"cmux-14262","repository":"manaflow-ai/cmux","kind":"pull-request","number":14262,"title":"Compare build inputs by content so an adopted seed rebuilds only real changes","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"Seeded PR compile admissions reran every C compile and re-emitted every package module even with an unchanged source tree, because llbuild compared mtimes and 1,616 Xcode SDK and toolchain files carry different mtimes on each runner image. Building with Swift Build's FileSystemMode=checksum-only, and folding the mode into the seed fingerprint, made the C and resource work cache-hit.","evidence":{"label":"CMUX PR 14262","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14262","basis":"XCBuild rerun-reason trace from canary run 36022099083 (seed distance 0: 94 SwiftDriver, 64 SwiftEmitModule). CompileC counts from the before/after comment on cmux#14275: run 36001051652 had 129 CompileC, run 36040706124 had 0."},"claims":[{"id":"seeded-admission-c-compiles","dimension":"compute","metric":"CompileC tasks per seeded test-only admission","direction":"reduction","value":129,"unit":"tasks/admission","recurrence":"per-run","confidence":"observed","measurementWindow":"one admission before (run 36001051652) and one after (run 36040706124)","population":"Seeded macOS compile admissions for pull requests","note":"129 to 0. One run each: the before run was 12 vCPU with a same-width seed, the after run 6 vCPU with a j12 seed, and the after seed also carried cmux-14241's directory-time replay. The Swift module re-emits remained until cmux-14275.","headline":false,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"}],"sourcePath":"work/impact/records/cmux-14262.md"},{"id":"cmux-14275","repository":"manaflow-ai/cmux","kind":"pull-request","number":14275,"title":"Seed each Swift job width admission compiles at","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","build-performance","caching"],"summary":"Swift Build passes -j<CPU count> to every swift-driver invocation, and no build setting overrides it. A seed built on 12 vCPU therefore changed every Swift driver task on a 6 vCPU runner, and all 62 modules re-emitted. Seeds are now keyed by width and built for each Blacksmith macOS 26 pool, and admission adopts a seed of its own width first.","evidence":{"label":"CMUX PR 14275","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14275#issuecomment-5824983225","basis":"Before/after posted on the PR: test-only canary #14269. Run 36040706124 (6 vCPU, j12 seed): 62 SwiftEmitModule, 94 SwiftDriver, cmux scheme 145 s. Run 36077356582 (6 vCPU, j6 seed, distance 0): 0 SwiftEmitModule, 0 SwiftDriver, 0 CompileC, cmux scheme 59 s."},"claims":[{"id":"seeded-admission-module-reemits","dimension":"compute","metric":"SwiftEmitModule tasks in the cmux scheme of a test-only seeded admission","direction":"reduction","value":62,"unit":"tasks/admission","recurrence":"per-run","confidence":"observed","measurementWindow":"one admission before (run 36040706124) and one after (run 36077356582), both 6 vCPU","population":"Seeded macOS compile admissions on Blacksmith macOS 26 pools","note":"62 to 0, one run each. Holds when a seed of the runner's width exists. Owned 14-CPU minis have none yet (cmuxterm-hq#590). cmuxTests still recompiles fully after adoption (cmux#14320).","headline":true,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"},{"id":"cmux-scheme-wall-time","dimension":"critical-path","metric":"cmux scheme build-for-testing wall time in admission","direction":"reduction","value":86,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"same two 6 vCPU runs","population":"Seeded macOS compile admissions on 6 vCPU macOS 26","note":"145 s to 59 s. One run each, so indicative only.","headline":false,"additive":false,"overlapGroup":"cmux-seeded-admission-rebuild"}],"sourcePath":"work/impact/records/cmux-14275.md"},{"id":"cmux-14282","repository":"manaflow-ai/cmux","kind":"pull-request","number":14282,"title":"Keep the runner account out of the SwiftPM manifest cache key","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","critical-path","caching"],"summary":"The SwiftPM manifest cache from cmux-14257 never hit on the owned Mac minis, because the runner account's HOME was part of the key. The resolve now runs with a fixed HOME, so the minis hit the same cache as Blacksmith.","evidence":{"label":"CMUX PR 14282","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14282#issuecomment-5823586250","basis":"PR comment: admission 'Resolve Swift packages' on the glaeda minis took 52 to 56 s before (PR runs 19:07 to 20:40Z, e.g. job 107819765887) and 13 to 16 s after (n=4, runs created 21:55 to 22:55Z, after seed run 36059263610). Blacksmith macOS 26 in the same window: median 15 s (n=11)."},"claims":[{"id":"mini-admission-resolve-time","dimension":"critical-path","metric":"admission Resolve Swift packages step on owned minis","direction":"reduction","value":36,"unit":"seconds/run","recurrence":"per-run","confidence":"observed","measurementWindow":"before 19:07 to 20:40Z; after 21:55 to 22:55Z (n=4), 2026-09-24","population":"Compile admissions on owned glaeda minis","note":"52 to 56 s before, 13 to 16 s after; 36 s is the conservative gap (52 minus 16). cli-pipe-regressions on a mini still missed (41 s, n=1) because its workspace path differs.","headline":true,"additive":false,"overlapGroup":"cmux-swiftpm-resolve"}],"sourcePath":"work/impact/records/cmux-14282.md"},{"id":"cmux-14285","repository":"manaflow-ai/cmux","kind":"pull-request","number":14285,"title":"Keep compile admission's build state on an owned Mac between jobs","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["ci","fleet","caching"],"summary":"Owned Mac minis now keep compile admission's DerivedData and package checkouts between jobs, instead of restoring caches and a seed every time.","evidence":{"label":"CMUX PR 14285 after-merge measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/14285#issuecomment-5825517334","basis":"Measured after merge from GitHub Actions job timings (read-only), posted on the PR with a correction: warm owned admissions median 10.7 min (n=4; jobs 107886727411, 107903167882, 107904138254, 107908956658) against cold owned 15.4 min (n=2, app-profile jobs 107883200927 and 107909027255) in the same window; seeded Blacksmith macOS 26 10.3 min (n=20). The first owned run (36048804178, PR body) took 25 min."},"claims":[{"id":"owned-admission-time","dimension":"critical-path","metric":"owned-mini compile admission job time, warm kept state against cold","direction":"reduction","value":4.7,"unit":"minutes/run","recurrence":"per-run","confidence":"observed","measurementWindow":"2 cold and 4 warm owned runs, 2026-09-24/25","population":"Compile admissions on owned glaeda minis","note":"15.4 to 10.7 min. It comes from skipping the seed and cache restore: the kept DerivedData still recompiled about 2,600 tasks, because the kept tree is whatever PR the mini built last (cmuxterm-hq#590). Warm owned is still 0.4 min behind the Blacksmith median. One warm job (107903167882) failed after compiling.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-14285.md"},{"id":"glaeda-1146","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1146,"title":"glaeda-cmux-runner: one-command GitHub Actions runner for a cmux Mac mini","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["fleet","ci"],"summary":"One command turns a cmux Mac mini into a GitHub Actions runner with fleet labels. The first owned-mini runner (cmux15-glaeda) ran a job green on the fork.","evidence":{"label":"Glaeda PR 1146","url":"https://github.com/teamleaderleo/glaeda/pull/1146#issuecomment-5812849555","basis":"PR comment: teamleaderleo/cmux run 35990802447 ran on cmux15-glaeda and succeeded in 29 s. Later the same day, glaeda#1175's body lists 13 glaeda pool runners registered on manaflow-ai/cmux (std 11, light 2), before the owned pool was switched on."},"claims":[{"id":"owned-runners","dimension":"compute","metric":"owned Mac mini runners with a verified job","direction":"shift","value":1,"unit":"runners","recurrence":"none","confidence":"observed","measurementWindow":"2026-09-24","population":"Owned Manaflow Mac mini runners (first job on the teamleaderleo/cmux fork)","note":"0 to 1, job-verified on the fork. By glaeda#1175, 13 runners were registered on manaflow-ai/cmux but not yet taking jobs; glaeda#1200 plans 36. Neither is job-verified here.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/glaeda-1146.md"},{"id":"glaeda-1162","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1162,"title":"Measure the fleet compilation cache mini to mini on the full cmux app","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["build-performance","caching","fleet"],"summary":"Measured Xcode's compilation cache served from a shared fleet store on the Manaflow M4 Pro minis: a fresh full cmux app build that reads a filled store instead of compiling.","evidence":{"label":"Glaeda PR 1162","url":"https://github.com/teamleaderleo/glaeda/pull/1162","basis":"PR body and docs/experiments/fleet-compilation-cache-minis-2026-09-24.md: full cmux app, Xcode 26.3, cold build 756 s; fresh build reading the store 132 s with 4,191 of 4,191 cacheable tasks hit and byte-identical outputs. The PR attributed about 52 s of the 132 s to fetching; glaeda-1180 later found most of that was a loaded store host, with the network adding about 20 to 30 s."},"claims":[{"id":"fresh-full-app-build","dimension":"developer-loop","metric":"fresh full cmux app build with a filled fleet cache","direction":"reduction","value":624,"unit":"seconds/event","recurrence":"per-event","confidence":"observed","measurementWindow":"one cold and one cached build, 2026-09-24","population":"Fresh builds on a Manaflow M4 Pro mini","note":"756 s to 132 s. Lab measurement, not CI. For the full app the store and reader were on the same mini; cross-machine was shown on the CmuxSettingsUI chain, and for the full app in glaeda-1180.","headline":true,"additive":false,"overlapGroup":"glaeda-fleet-cas-fresh-build"}],"sourcePath":"work/impact/records/glaeda-1162.md"},{"id":"glaeda-1180","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1180,"title":"Record the cross-machine full-app read of the fleet compilation cache","status":"merged","happenedAt":"2026-09-24","recordedAt":"2026-09-25","areas":["build-performance","caching","fleet"],"summary":"Repeated glaeda-1162's full-app measurement with the store filled on one mini and read from another over the LAN.","evidence":{"label":"Glaeda PR 1180","url":"https://github.com/teamleaderleo/glaeda/pull/1180","basis":"PR body: cmux8s filled, cmux7s read, Xcode 26.3: 4,191 of 4,191 hits in 127 s (warm node 77 s; same-host reference 108 s). 7,631 of 7,634 outputs byte-identical; the 3 that differ are two Go WireGuard library copies and the debug dylib."},"claims":[{"id":"cross-machine-full-app-read","dimension":"developer-loop","metric":"fresh full cmux app build reading another mini's store","direction":"reduction","value":629,"unit":"seconds/event","recurrence":"per-event","confidence":"observed","measurementWindow":"one build, 2026-09-24, against glaeda-1162's 756 s cold baseline","population":"Fresh builds on Manaflow M4 Pro minis","note":"756 s to 127 s. Members share entries only on the same Xcode build. The network adds about 20 to 30 s.","headline":false,"additive":false,"overlapGroup":"glaeda-fleet-cas-fresh-build"}],"sourcePath":"work/impact/records/glaeda-1180.md"},{"id":"cmux-13935","repository":"manaflow-ai/cmux","kind":"pull-request","number":13935,"title":"Run the CmuxMobileShell package tests serially","status":"merged","happenedAt":"2026-09-23T07:12:30Z","recordedAt":"2026-09-23","areas":["ci","testing","swift","reliability","performance"],"summary":"Diagnosed a package lane that timed out without reporting a result as contention from parallel Swift Testing execution; serial execution finished the same 1,060 tests in about a minute and exposed the smaller set of real failures.","evidence":{"label":"CMUX PR 13935","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13935","basis":"The PR body records parallel and serial runs of the same 1,060 tests, their wall time, issue counts, and a representative test family."},"claims":[{"id":"serial-test-time-lower-bound","dimension":"critical-path","metric":"package-test-wall-time-reduction-lower-bound","direction":"reduction","value":655,"unit":"seconds/run","recurrence":"per-run","confidence":"derived","measurementWindow":"same-machine parallel/serial comparison","population":"1,060 CmuxMobileShell package tests","note":"Parallel execution exceeded 12 minutes without finishing; serial execution completed in 65 seconds. The stored 655-second reduction is therefore a lower bound.","headline":true,"additive":false,"overlapGroup":null},{"id":"contention-failures-removed","dimension":"reliability","metric":"contention-failures-removed-in-comparison","direction":"reliability","value":144,"unit":"failures/run","recurrence":"one-time","confidence":"derived","measurementWindow":"same-machine parallel/serial comparison","population":"1,060 CmuxMobileShell package tests","note":"The parallel run produced about 161 issues while the serial run reported 17 real failures; the difference is approximately 144 contention artifacts.","headline":false,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-13935.md"},{"id":"cmux-13902","repository":"manaflow-ai/cmux","kind":"pull-request","number":13902,"title":"Move focused E2E dispatches to the low-queue macOS pool","status":"merged","happenedAt":"2026-09-23T06:05:47Z","recordedAt":"2026-09-23","areas":["ci","latency","macos","runner-routing","testing"],"summary":"Compared focused E2E dispatches across the macOS 15 and macOS 26 pools, then changed the automatic default to the pool with dramatically lower queue tails while preserving an explicit macOS 15 override.","evidence":{"label":"CMUX PR 13902","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13902","basis":"The PR body records 60 dispatches, queue median/p90/max, execution median, total elapsed median, and the limitation that the two runner groups did not execute identical tests."},"claims":[{"id":"e2e-queue-p90-gap","dimension":"critical-path","metric":"focused-e2e-queue-p90-reduction","direction":"reduction","value":82.3,"unit":"minutes","recurrence":"none","confidence":"observed","measurementWindow":"60 dispatches from 2026-09-22T21:19Z to 2026-09-23T05:18Z","population":"33 macOS 15 and 25 macOS 26 focused E2E dispatches","note":"Observed queue p90 was 83.3 minutes on macOS 15 and 1.0 minute on macOS 26. Execution median was about four minutes slower on macOS 26, and the test mixes differed, so the comparison is operational rather than causal.","headline":true,"additive":false,"overlapGroup":"cmux-macos-queue-2026-09"},{"id":"e2e-execution-tradeoff","dimension":"compute","metric":"focused-e2e-execution-median-increase","direction":"shift","value":3.8,"unit":"minutes","recurrence":"none","confidence":"observed","measurementWindow":"same 60-dispatch comparison","population":"focused E2E dispatches","note":"Execution median was 18.7 minutes on macOS 15 and 22.5 minutes on macOS 26 while total elapsed median still favored macOS 26 after queueing.","headline":false,"additive":false,"overlapGroup":"cmux-macos-queue-2026-09"}],"sourcePath":"work/impact/records/cmux-13902.md"},{"id":"cmux-13749","repository":"manaflow-ai/cmux","kind":"pull-request","number":13749,"title":"Read app-host artifacts over parallel range requests","status":"merged","happenedAt":"2026-09-23T03:59:13Z","recordedAt":"2026-09-23","areas":["ci","performance","artifact-transport","macos"],"summary":"Traced app-host product restore time to the single-stream artifact transport rather than hashing or extraction, then added a verified parallel range-request path with the canonical GitHub download retained as fallback.","evidence":{"label":"CMUX PR 13749","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13749","basis":"The PR body records per-phase timings, artifact size and throughput, the seven-consumer cost, integrity checks, and fallback behavior."},"claims":[{"id":"artifact-transfer-baseline","dimension":"compute","metric":"artifact-transfer-job-minutes-per-full-suite","direction":"baseline","value":60,"unit":"job-minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"two referenced CI runs","population":"Seven app-host product consumers in a full-suite run","note":"The baseline was about eight minutes per consumer; the PR reports the transfer step as roughly 98 percent transport time and moves the path to concurrent ranges with verification and fallback.","headline":true,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-13749.md"},{"id":"cmux-13646","repository":"manaflow-ai/cmux","kind":"pull-request","number":13646,"title":"Throttle push-triggered Nightly builds by published commit age","status":"merged","happenedAt":"2026-09-23T01:25:39Z","recordedAt":"2026-09-23","areas":["ci","cost","macos","nightly","capacity"],"summary":"Measured seven days of push-triggered Nightly activity and added a two-hour publication interval so busy main-branch periods stop running the expensive macOS publication pipeline back to back.","evidence":{"label":"CMUX PR 13646","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13646","basis":"The PR body records seven days of push runs, publish frequency, per-publish macOS minutes, a replay of alternative intervals, and the delivery-latency trade-off."},"claims":[{"id":"nightly-push-savings-low-edge","dimension":"compute","metric":"projected-macos-runner-minutes-avoided-per-day","direction":"reduction","value":450,"unit":"runner-minutes/day","recurrence":"daily","confidence":"projected","measurementWindow":"seven-day replay","population":"push-triggered Nightly publication work","note":"The PR projects roughly 450–600 macOS minutes/day avoided at the two-hour default. This record stores the lower edge of that stated range; the PR also gives a roughly 280-minute/day lower-bound scenario.","headline":true,"additive":false,"overlapGroup":"cmux-nightly-capacity-2026-09"},{"id":"nightly-push-baseline","dimension":"compute","metric":"push-triggered-nightly-macos-minutes-per-day","direction":"baseline","value":1030,"unit":"runner-minutes/day","recurrence":"daily","confidence":"measured","measurementWindow":"seven days","population":"push-to-main Nightly macOS jobs","note":"The seven-day sample recorded 7,197 push-to-main macOS minutes, about 1,030 minutes/day including three hung builds.","headline":false,"additive":false,"overlapGroup":"cmux-nightly-capacity-2026-09"}],"sourcePath":"work/impact/records/cmux-13646.md"},{"id":"cmux-13810","repository":"manaflow-ai/cmux","kind":"pull-request","number":13810,"title":"Measure CI waste on a schedule","status":"merged","happenedAt":"2026-09-23T00:09:55Z","recordedAt":"2026-09-23","areas":["ci","observability","capacity","cost","reliability"],"summary":"Turned repeated manual Actions investigations into a scheduled health report covering runner minutes, queue percentiles, cancelled work, repeated unchanged heads, mostly-skipped workflows, fork cache behavior, and previous-window comparison.","evidence":{"label":"CMUX PR 13810","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13810","basis":"The PR body contains the first real report, including partial-window caveats, queue percentiles, sampled runner-minute tables, cancellation patterns, skipped workflows, and unchanged-head repetition."},"claims":[{"id":"unchanged-head-repeat-runs","dimension":"observability","metric":"runs-beyond-first-on-unchanged-head","direction":"baseline","value":9763,"unit":"runs/window","recurrence":"one-time","confidence":"observed","measurementWindow":"first real report, approximately 23.1 hours of a 24-hour window","population":"2,424 workflow/head pairs in the fetched run set","note":"The report found 9,763 runs beyond the first for a head that did not change. This is an observation to investigate, not a claim that every repeated run is removable.","headline":true,"additive":false,"overlapGroup":"cmux-control-plane-runs-2026-09"},{"id":"macos15-queue-p90-observed","dimension":"critical-path","metric":"macos15-queue-p90","direction":"baseline","value":119.8,"unit":"minutes","recurrence":"one-time","confidence":"observed","measurementWindow":"same first health report","population":"sampled blacksmith-6vcpu-macos-15 jobs","note":"The previous window's corresponding p90 was 0.9 minutes, which made the new 119.8-minute queue visible as a capacity regression.","headline":false,"additive":false,"overlapGroup":"cmux-macos-queue-2026-09"}],"sourcePath":"work/impact/records/cmux-13810.md"},{"id":"cmux-13703","repository":"manaflow-ai/cmux","kind":"pull-request","number":13703,"title":"Remove full-history checkout and unconditional Zig setup from iOS uploads","status":"merged","happenedAt":"2026-09-23T00:00:11Z","recordedAt":"2026-09-23","areas":["ci","ios","release","macos","cost"],"summary":"Replaced a full recursive checkout and unconditional Zig installation in both iOS upload workflows with a shallow checkout, prebuilt-first GhosttyKit path, and bounded history fetch only when release notes actually need it.","evidence":{"label":"CMUX PR 13703","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13703","basis":"The PR body records a 12-minute checkout and 6-minute Zig install in a real upload, local shallow/full checkout timings, the new fallback semantics, and the expected recurring savings."},"claims":[{"id":"ios-upload-setup-savings","dimension":"compute","metric":"macos-minutes-avoided-per-upload","direction":"reduction","value":16,"unit":"runner-minutes/event","recurrence":"per-event","confidence":"derived","measurementWindow":"real upload baseline plus local shallow-checkout timing","population":"iOS TestFlight/App Store upload runs when the pinned GhosttyKit prebuilt exists","note":"The PR states about 16–17 macOS minutes saved per upload. This record stores the lower edge of that range.","headline":true,"additive":false,"overlapGroup":"cmux-ios-upload-capacity-2026-09"},{"id":"ios-hourly-lane-ceiling","dimension":"compute","metric":"maximum-hourly-lane-macos-minutes-avoided-per-day","direction":"reduction","value":390,"unit":"runner-minutes/day","recurrence":"daily","confidence":"projected","measurementWindow":"24 hourly upload opportunities","population":"official cmux.app hourly upload lane","note":"The PR gives an upper ceiling of about 6.5 macOS-hours/day when all 24 hourly opportunities upload. Actual savings depend on upload frequency.","headline":false,"additive":false,"overlapGroup":"cmux-ios-upload-capacity-2026-09"}],"sourcePath":"work/impact/records/cmux-13703.md"},{"id":"cmux-13652","repository":"manaflow-ai/cmux","kind":"issue","number":13652,"title":"Measure CMUX macOS CI capacity and replacement-cost exposure","status":"measured","happenedAt":"2026-09-23","recordedAt":"2026-09-23","areas":["ci","cost","macos","capacity","economics"],"summary":"A GitHub Actions 90-day usage export records 1,966,401 self-hosted macOS runner-minutes, averaging 21,848.9 minutes/day. Pricing every minute at only WarpBuild's cheapest documented macOS rate yields a $52,437.36 per 30-day-month replacement-cost floor.","evidence":{"label":"Scrapbook 90-day Actions usage source","url":"https://github.com/teamleaderleo/scrapbook/blob/main/work/impact/sources/cmux-actions-90d-2026-09-23.json","basis":"Derived directly from the GitHub Actions 90-day usage CSV downloaded on 2026-09-23; the checked-in source summary preserves aggregate minutes and the leading macOS workflows."},"claims":[{"id":"actions-self-hosted-macos-90d","dimension":"compute","metric":"self-hosted-macos-runner-minutes-90d","direction":"baseline","value":1966401,"unit":"runner-minutes/90-days","recurrence":"none","confidence":"measured","measurementWindow":"GitHub Actions 90-day usage export downloaded 2026-09-23","population":"All rows in the export where runner type is self-hosted and runtime OS is macOS","note":"The export contains 1,966,401 self-hosted macOS minutes. ci.yml alone accounts for 1,345,657 minutes, or about 68.4 percent of this total.","headline":false,"additive":false,"overlapGroup":"cmux-macos-capacity-90d-2026-09"},{"id":"actions-self-hosted-macos-daily-average","dimension":"compute","metric":"self-hosted-macos-runner-minutes-per-day-90d-average","direction":"baseline","value":21848.9,"unit":"runner-minutes/day","recurrence":"daily","confidence":"derived","measurementWindow":"Same 90-day export divided by 90 days","population":"Self-hosted macOS Actions workload","note":"This 90-day average is about 16.8 percent above the separate 18,704-minute one-day accounting from September 22.","headline":false,"additive":false,"overlapGroup":"cmux-macos-capacity-90d-2026-09"},{"id":"warp-replacement-monthly-floor-90d","dimension":"replacement-cost","metric":"warp-list-price-equivalent-floor-per-30-day-month","direction":"baseline","value":52437.36,"unit":"USD/30-day-month","recurrence":"none","confidence":"derived","measurementWindow":"90-day Actions export; Warp price checked 2026-09-23","population":"All 1,966,401 self-hosted macOS minutes priced at WarpBuild's cheapest documented 6-vCPU macOS rate","note":"Floor valuation only: 655,467 equivalent minutes per 30 days at $0.08/min. No 12-vCPU premium is applied. This is replacement-cost exposure, not a claim that every minute was actually billed to Warp or saved by one contributor.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-90d-2026-09"},{"id":"warp-replacement-annual-floor-90d","dimension":"replacement-cost","metric":"warp-list-price-equivalent-floor-annualized","direction":"baseline","value":637987.88,"unit":"USD/year","recurrence":"none","confidence":"derived","measurementWindow":"Same 90-day Actions export annualized over 365 days","population":"Same self-hosted macOS workload priced at $0.08/min","note":"Annualized replacement-cost floor at the 90-day average. Any 12-vCPU Warp usage would raise the equivalent value.","headline":false,"additive":false,"overlapGroup":"cmux-macos-capacity-90d-2026-09"},{"id":"point-in-time-macos-runner-occupancy","dimension":"compute","metric":"macos-runner-minutes-per-day-point-in-time","direction":"baseline","value":18704,"unit":"runner-minutes/day","recurrence":"none","confidence":"measured","measurementWindow":"24 hours ending approximately 2026-09-22T17:00Z","population":"Nine CMUX workflows scheduling macOS jobs; only time that actually held a runner","note":"This independent issue-level accounting remains a useful corroborating snapshot but is no longer the primary economic baseline.","headline":false,"additive":false,"overlapGroup":"cmux-macos-capacity-2026-09"}],"sourcePath":"work/impact/records/cmux-13652.md"},{"id":"cmux-13908","repository":"manaflow-ai/cmux","kind":"pull-request","number":13908,"title":"Compile the E2E test product once and reuse it on reruns","status":"merged","happenedAt":"2026-09-23","recordedAt":"2026-09-23","areas":["ci","testing","macos","build-reuse"],"summary":"Split the focused E2E dispatch into filter, build, and test jobs so invalid selectors fail on Linux before a Mac starts and reruns download the retained test product instead of recompiling it.","evidence":{"label":"CMUX PR 13908","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13908","basis":"The PR body measures 60 dispatches, isolates the approximately 15.3-minute compile portion, and states that reruns now download the product and compile nothing."},"claims":[{"id":"rerun-compile-avoided","dimension":"critical-path","metric":"compile-minutes-avoided-per-failed-dispatch-rerun","direction":"reduction","value":15.3,"unit":"minutes/event","recurrence":"per-event","confidence":"derived","measurementWindow":"60 dispatches over about 8 hours","population":"Reruns of a failed focused E2E dispatch while the retained product remains available","note":"Fresh dispatches do not claim lower total runner minutes; this figure is the measured compile portion removed from a rerun by the new job split.","headline":true,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-13908.md"},{"id":"cmux-13962","repository":"manaflow-ai/cmux","kind":"pull-request","number":13962,"title":"Report recorded failures even when app-host results are incomplete","status":"merged","happenedAt":"2026-09-23","recordedAt":"2026-09-23","areas":["ci","testing","reliability","diagnostics"],"summary":"Fixed an early-return path where one missing terminal test result caused the shard to suppress every failure it had successfully recorded, leaving red full-suite runs without the names of the tests that regressed.","evidence":{"label":"CMUX PR 13962","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13962","basis":"The PR body compares CI output with the same run's logs and records 60 XCTest assertion failures plus 88 Swift Testing failures that produced zero ratchet lines."},"claims":[{"id":"hidden-recorded-failures","dimension":"reliability","metric":"recorded-failures-hidden-in-observed-run","direction":"reliability","value":148,"unit":"failures/run","recurrence":"one-time","confidence":"measured","measurementWindow":"one referenced full-suite run","population":"Recorded XCTest and Swift Testing failures in the incomplete app-host shards","note":"The fix changes diagnostics, not the fail-closed verdict: incomplete runs remain failed while recorded failures are now named.","headline":true,"additive":false,"overlapGroup":null}],"sourcePath":"work/impact/records/cmux-13962.md"},{"id":"glaeda-1133","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1133,"title":"Reclaim eligible linked worktrees with --apply","status":"merged","happenedAt":"2026-09-23","recordedAt":"2026-09-25","areas":["developer-loop","storage"],"summary":"glaeda-worktree-reclaim gained --apply: it re-plans each eligible linked worktree, pins an orphan HEAD, rechecks for movement and removes it with git's own guards, recording commit and branch so it can be recreated. The first runs cleared a backlog of 248 idle worktrees across Air Blue and Big Red. The first apply on Air Blue ran from the PR build shortly before merge.","evidence":{"label":"Glaeda PR 1133","url":"https://redirect.github.com/teamleaderleo/glaeda/pull/1133#issuecomment-5826944522","basis":"PR comment summarizing the apply receipts from 2026-09-23 (24 h idle window): first apply on Air Blue, 2 repos, 57 linked, 8 reclaimed; Air Blue backlog, 21 repos, 211 linked, 148 reclaimed; Big Red backlog, 23 repos, 161 linked, 92 reclaimed. No other outcomes, circuit breaker never tripped. Per-repository free space rose by 11.4 GiB in total."},"claims":[{"id":"worktrees-reclaimed-backlog","dimension":"storage","metric":"idle linked worktrees removed in the first apply runs","direction":"reduction","value":248,"unit":"worktrees","recurrence":"one-time","confidence":"measured","measurementWindow":"three apply runs, 2026-09-23 03:00 to 03:15 EDT","population":"Git repositories under ~/Projects on Air Blue and Big Red","note":"Exact count from receipts; each reclaim is recoverable from the recorded commit and branch. Free space rose by 11.4 GiB (sum of per-repository before and after readings, so other host activity can move it).","headline":true,"additive":false,"overlapGroup":"glaeda-worktree-reclaim"}],"sourcePath":"work/impact/records/glaeda-1133.md"},{"id":"glaeda-1140","repository":"teamleaderleo/glaeda","kind":"pull-request","number":1140,"title":"Add glaeda-disk: where the disk went, and reclaim idle build output","status":"merged","happenedAt":"2026-09-23","recordedAt":"2026-09-25","areas":["developer-loop","storage"],"summary":"glaeda-disk reports where the disk went by family and deletes idle, reproducible build output (Xcode DerivedData, temp-dir scratch, idle session scratchpads) with process, git and idle-window vetoes. On Air Blue it deleted 52.8 GiB of idle build output in its first day and a half, 35.5 GiB of it after merge.","evidence":{"label":"Glaeda PR 1140","url":"https://redirect.github.com/teamleaderleo/glaeda/pull/1140#issuecomment-5826944345","basis":"PR comment summarizing Air Blue's receipt file (~/Projects/recovery/disk-reclaim/receipts.jsonl): 32 deletions, all outcome reclaimed, all idle 6 h or more, 2026-09-23 11:57 to 2026-09-24 18:50 EDT. 17.3 GiB in 18 deletions before merge (PR build), 35.5 GiB in 14 after. By family: DerivedData 36.1 GiB, temp scratch 12.7, session scratch 4.0."},"claims":[{"id":"air-blue-idle-output-deleted","dimension":"storage","metric":"idle build output deleted by glaeda-disk on Air Blue","direction":"reduction","value":52.8,"unit":"GiB","recurrence":"one-time","confidence":"measured","measurementWindow":"32 receipt lines, 2026-09-23 11:57 to 2026-09-24 18:50 EDT","population":"Air Blue (Leo's MacBook Air)","note":"Gross bytes deleted, not sustained free space: DerivedData and scratch regrow when the next build or session needs them. 35.5 GiB of it came after merge; the other 17.3 GiB was deleted by the PR's own build before merge. The minis run the same pressure job but have not been under pressure (194 to 270 GiB free against a 69.1 GiB threshold), so there are no mini receipts.","headline":true,"additive":false,"overlapGroup":"air-blue-disk-reclaim"}],"sourcePath":"work/impact/records/glaeda-1140.md"},{"id":"swift-92545","repository":"swiftlang/swift","kind":"issue","number":92545,"title":"Cached-diagnostics replay crash with mixed prefix-mapped inputs","status":"filed","happenedAt":"2026-09-23","recordedAt":"2026-09-23","areas":["swift","compiler","build-system","compilation-cache","investigation"],"summary":"Reduced a real compilation-cache crash to three Swift source files and one resource, traced the replay failure through CachedDiagnostics and FileSpecificDiagnosticConsumer, decoded the CAS diagnostics blob, and connected the missing-file condition to an earlier prefix-map fix.","evidence":{"label":"Swift issue 92545","url":"https://redirect.github.com/swiftlang/swift/issues/92545","basis":"The issue contains the deterministic reproducer, controls, decoded cache evidence, stack trace, suspected failure path, workarounds, and related prior fixes."},"claims":[],"sourcePath":"work/impact/records/swift-92545.md"},{"id":"cmux-13797","repository":"manaflow-ai/cmux","kind":"pull-request","number":13797,"title":"Seed the test compilation cache from one clean build","status":"merged","happenedAt":"2026-09-22T23:51:24Z","recordedAt":"2026-09-23","areas":["ci","cache","swift","macos","performance"],"summary":"Stopped the test compilation-cache seeder from restoring its previous seed by prefix and recursively accumulating old generations until it crossed the save bound and froze.","evidence":{"label":"CMUX PR 13797","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13797","basis":"The PR body records cache growth across scheduled seeds, the 5 GiB save bound, one clean-build result, producer timings, and the per-pull-request download consequence."},"claims":[{"id":"compilation-seed-download-reduction","dimension":"cache-transfer","metric":"compilation-seed-download-reduction-per-pr","direction":"reduction","value":1.5,"unit":"GiB/run","recurrence":"per-run","confidence":"measured","measurementWindow":"scheduled seed comparison","population":"pull requests restoring the shared test compilation-cache seed","note":"The clean seed is about 3.5 GiB instead of roughly 5.0 GiB, so every pull request restores about 1.5 GiB less.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"}],"sourcePath":"work/impact/records/cmux-13797.md"},{"id":"cmux-13644","repository":"manaflow-ai/cmux","kind":"pull-request","number":13644,"title":"Debounce macOS admission on superseded pull-request pushes","status":"merged","happenedAt":"2026-09-22T14:00:30Z","recordedAt":"2026-09-23","areas":["ci","cost","macos","queueing"],"summary":"Inserted a bounded Linux admission delay before expensive macOS work so rapid follow-up pushes cancel stale runs before a Mac starts, then replayed a day of job timings to estimate the avoided work and latency trade-off.","evidence":{"label":"CMUX PR 13644","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13644","basis":"The PR body contains the 24-hour run sample, cancelled-minute breakdown, replay method, and delay trade-off table."},"claims":[{"id":"cancelled-macos-minutes-avoided","dimension":"compute","metric":"runner-minutes-avoided-per-day","direction":"reduction","value":590,"unit":"runner-minutes/day","recurrence":"daily","confidence":"derived","measurementWindow":"24-hour replay","population":"Superseded pull-request runs that would otherwise start macOS work","note":"Replay estimate at the default 180-second delay; the same analysis measured a 1.4-minute median increase to the green signal.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-2026-09"}],"sourcePath":"work/impact/records/cmux-13644.md"},{"id":"cmux-13658","repository":"manaflow-ai/cmux","kind":"pull-request","number":13658,"title":"Route pull-request macOS jobs to sponsored capacity","status":"merged","happenedAt":"2026-09-22T13:46:48Z","recordedAt":"2026-09-23","areas":["ci","cost","macos","runner-routing"],"summary":"Measured a day of macOS workload by triggering event, then routed the pull-request burst lane from paid overflow onto sponsored Blacksmith capacity while leaving other events on the existing repository variable.","evidence":{"label":"CMUX PR 13658","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13658","basis":"The PR body records the 24-hour workload measurement, event split, routing change, and rollback control."},"claims":[{"id":"paid-capacity-shift","dimension":"paid-compute","metric":"runner-minutes-shifted-from-paid-capacity-per-day","direction":"shift","value":5900,"unit":"runner-minutes/day","recurrence":"daily","confidence":"measured","measurementWindow":"24 hours","population":"Pull-request macOS overflow work in the sampled CMUX CI day","note":"This moves workload onto sponsored capacity; it does not claim the compute disappeared.","headline":true,"additive":false,"overlapGroup":"cmux-macos-capacity-2026-09"}],"sourcePath":"work/impact/records/cmux-13658.md"},{"id":"cmux-13645","repository":"manaflow-ai/cmux","kind":"pull-request","number":13645,"title":"Trigger the persistent Mac router once per CI run","status":"merged","happenedAt":"2026-09-22T13:44:27Z","recordedAt":"2026-09-23","areas":["ci","control-plane","macos","capacity","routing"],"summary":"Changed the persistent-Mac router from a workflow activity emitted for every CI job to the once-per-run requested event, eliminating thousands of duplicate router invocations while preserving the trust boundary and hosted fallback.","evidence":{"label":"CMUX PR 13645","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13645","basis":"The PR body records the 24-hour router-run counts, explains why the old activity fired per job, and documents the re-run coverage trade-off."},"claims":[{"id":"router-runs-avoided","dimension":"control-plane","metric":"router-runs-avoided-per-day","direction":"reduction","value":5611,"unit":"runs/day","recurrence":"daily","confidence":"derived","measurementWindow":"24-hour measured baseline","population":"persistent-macos-router workflow invocations","note":"The measured baseline was about 6,864 router runs/day; the once-per-CI-run trigger was expected to produce about 1,253/day, an 82% reduction.","headline":true,"additive":false,"overlapGroup":"cmux-control-plane-runs-2026-09"}],"sourcePath":"work/impact/records/cmux-13645.md"},{"id":"cmux-13754","repository":"manaflow-ai/cmux","kind":"pull-request","number":13754,"title":"Seed compilation cache on the runner pull requests restore from","status":"merged","happenedAt":"2026-09-22","recordedAt":"2026-09-23","areas":["ci","swift","compilation-cache","runner-routing"],"summary":"Found that the nightly compilation-cache seed and pull-request consumer ran on different runner layouts even though the cache fingerprint included the workspace path, making the seed impossible to restore by construction.","evidence":{"label":"CMUX PR 13754","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13754","basis":"The PR body records the exact cache miss, runner labels, fingerprint mechanism, and a measured 19.9-minute cold compile step."},"claims":[{"id":"cold-pr-compile-baseline","dimension":"critical-path","metric":"cold-compile-minutes-per-affected-pr","direction":"baseline","value":19.9,"unit":"minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"referenced pull-request run","population":"Pull-request admission before the cache producer and consumer runner layouts matched","note":"The PR explicitly left warm compile time unmeasured until the next scheduled seed.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"}],"sourcePath":"work/impact/records/cmux-13754.md"},{"id":"cmux-13132","repository":"manaflow-ai/cmux","kind":"pull-request","number":13132,"title":"Move Computer Use runtime out of the 3,700-file app module","status":"merged","happenedAt":"2026-09-20T11:18:22Z","recordedAt":"2026-09-23","areas":["swift","build-system","developer-experience","modularity","latency"],"summary":"Moved the Computer Use runtime into a local package and a shared process-identity type into CmuxFoundation so edits in those files stop paying the app target's whole-module planning and compilation cost.","evidence":{"label":"CMUX PR 13132","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13132","basis":"The PR body records warm incremental comment-only edits before and after the move, app-target file counts, package tests, and the limit imposed by public-interface changes."},"claims":[{"id":"process-identity-incremental-build-reduction","dimension":"developer-loop","metric":"warm-incremental-build-time-reduction","direction":"reduction","value":25.6,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same-machine warm incremental builds","population":"comment-only edit to AgentPIDProcessIdentity.swift before and after moving it from the app target to CmuxFoundation","note":"The measured build fell from 38.9 seconds to 13.3 seconds and compiled zero app-target Swift files after the move.","headline":true,"additive":false,"overlapGroup":"cmux-local-build-loop-2026-09"}],"sourcePath":"work/impact/records/cmux-13132.md"},{"id":"cmux-13131","repository":"manaflow-ai/cmux","kind":"pull-request","number":13131,"title":"Reuse one warm DerivedData across development tags","status":"merged","happenedAt":"2026-09-20T10:15:05Z","recordedAt":"2026-09-23","areas":["developer-experience","xcode","swift","build-cache","latency"],"summary":"Changed tagged development builds from one fresh DerivedData tree per tag to a reusable warm tree, while keeping build serialization, CLI lookup, path validation, and cleanup ownership explicit.","evidence":{"label":"CMUX PR 13131","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13131","basis":"The PR records a same-machine matrix covering fresh DerivedData, same-commit tag reuse, branch/tag switches, small edits, larger app edits, and interface-invalidating changes."},"claims":[{"id":"same-commit-warm-derived-data","dimension":"developer-loop","metric":"same-commit-tag-build-time-reduction","direction":"reduction","value":917.3,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same-machine build matrix","population":"new development tag at the same commit after a warm build","note":"Fresh DerivedData took 953 seconds; a new tag using the already-warm shared DerivedData took 35.7 seconds.","headline":true,"additive":false,"overlapGroup":"cmux-local-build-loop-2026-09"}],"sourcePath":"work/impact/records/cmux-13131.md"},{"id":"cmux-13160","repository":"manaflow-ai/cmux","kind":"pull-request","number":13160,"title":"Stop pull requests from evicting shared cache seeds","status":"merged","happenedAt":"2026-09-20T09:06:05Z","recordedAt":"2026-09-23","areas":["ci","cache","swift","cost","reliability"],"summary":"Found that pull-request jobs were writing gigabyte-scale, PR-scoped cache entries that other pull requests could never read, rapidly evicting the main-branch seeds and forcing expensive macOS builds cold.","evidence":{"label":"CMUX PR 13160","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13160","basis":"The PR body records the cache inventory, entry scope, cold build timings, cache-store limits, and the change to restore-only pull-request consumers."},"claims":[{"id":"github-cache-churn-baseline","dimension":"cache","metric":"github-cache-occupancy-observed","direction":"baseline","value":37.5,"unit":"GB","recurrence":"one-time","confidence":"measured","measurementWindow":"cache inventory at 2026-09-20 05:55 UTC","population":"27 GitHub cache entries visible to the repository","note":"The oldest macOS entry was only five minutes old and the shared Debug compilation seed was already missing. Pull-request jobs had 14 save+restore cache steps and wrote 1–3 GB PR-scoped entries.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"}],"sourcePath":"work/impact/records/cmux-13160.md"},{"id":"cmux-13060","repository":"manaflow-ai/cmux","kind":"pull-request","number":13060,"title":"Seed the Debug compilation cache from main","status":"merged","happenedAt":"2026-09-20T02:42:11Z","recordedAt":"2026-09-23","areas":["ci","swift","compilation-cache","macos"],"summary":"Added a scheduled main-branch seed for the Debug compilation cache so pull-request compile admission could restore shared package work instead of always beginning from an empty cache.","evidence":{"label":"CMUX PR 13060","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13060","basis":"The PR body records the cold admission baseline, module timing breakdown, cache policy, local experiments, and the distinction between measured and expected CI behavior."},"claims":[{"id":"cold-admission-baseline","dimension":"critical-path","metric":"cold-compile-seconds-per-admission-run","direction":"baseline","value":800,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"referenced compile-admission run","population":"macOS compile admission from an empty compilation cache","note":"The PR deliberately did not quote an unmeasured CI savings number before the scheduled seed had run.","headline":true,"additive":false,"overlapGroup":"cmux-compilation-cache-2026-09"}],"sourcePath":"work/impact/records/cmux-13060.md"},{"id":"cmux-13097","repository":"manaflow-ai/cmux","kind":"pull-request","number":13097,"title":"Parallelize slow and history-dependent CI guards","status":"merged","happenedAt":"2026-09-20T01:11:39Z","recordedAt":"2026-09-23","areas":["ci","latency","linux","critical-path"],"summary":"Split one roughly 390-second guard job into shallow, history-only, CLI-script, and source-lint jobs so independent policy checks run concurrently before macOS admission.","evidence":{"label":"CMUX PR 13378 follow-up measurement","url":"https://redirect.github.com/manaflow-ai/cmux/pull/13378","basis":"The follow-up reusable-workflow PR records the accepted before/after measurement for PR 13097: 385.5 seconds to 124.3 seconds on the guard critical path, with Linux runner-minutes increasing from 6.43 to 7.49."},"claims":[{"id":"guard-critical-path-reduction","dimension":"critical-path","metric":"linux-guard-critical-path-reduction","direction":"reduction","value":261.2,"unit":"seconds/run","recurrence":"per-run","confidence":"measured","measurementWindow":"accepted before/after workflow runs","population":"Linux guard path gating macOS admission","note":"Measured guard wall time fell from 385.5 seconds to 124.3 seconds, about a 68% reduction.","headline":true,"additive":false,"overlapGroup":"cmux-linux-guard-latency-2026-09"},{"id":"guard-runner-minute-tradeoff","dimension":"compute","metric":"linux-runner-minutes-added-by-parallelism","direction":"shift","value":1.06,"unit":"runner-minutes/run","recurrence":"per-run","confidence":"measured","measurementWindow":"same before/after runs","population":"Linux guard jobs","note":"Runner consumption rose from 6.43 to 7.49 minutes because parallel jobs duplicate setup work. The change intentionally spends cheap compute to return the macOS admission answer sooner.","headline":false,"additive":false,"overlapGroup":"cmux-linux-guard-latency-2026-09"}],"sourcePath":"work/impact/records/cmux-13097.md"}],"links":{"publicPage":"/work/impact","repositoryRecord":"https://github.com/teamleaderleo/scrapbook/tree/main/work/impact","compactIndex":"/data/impact-index-v1.json","compactSummary":"/data/impact-summary-v1.json","candidates":"/api/work-impact/candidates","compactCandidates":"/data/impact-pr-candidates-v1.json"}}