Epic: Buck2 as the repository build authority

2026-09-13 — plan of record moved

Live state now lives in the rendered master epic schickling/megarepo-all#120 (rendered from buck2-ledger.json, PR schickling/megarepo-all#121). Decisions of 2026-09-12/13 (oversight seat dev3.direct.omp.t3mkm4sd, q1–q23): BUCK-R15 net complexity is the hard gate (decision 0031); local checks stay devenv verbs over one scoped Buck aggregate (0032); two cache trust tiers (0033); the #1260–#1265 stack lands as #1260 + value-ordered slices; cross-repo composition direction is artifact-default with no registry (release-asset origin + CAS accelerator; PRs #1282, #1287), composed-by-default paused. The sections below are the pre-2026-09-12 state and are retained as history.

Status

The migration is active again. A takeover audit found that the existing draft heads do not yet satisfy the full bounded deterministic JavaScript/TypeScript authority goal, and the stack must be rebased onto current main before its CI evidence is meaningful.

Canonical contract: context/buck2/ (vision, requirements, roadmap, decisions, and operation disposition).

Confirmed remaining work

• restack #1260-#1265 onto current main and restore a linear GitHub stack
• admit the remaining bounded effect-schema-form-aria unit suite
• enable shared-cache uploads for authoritative JavaScript product actions
• replace the last oxc-config Nix/pnpm source build and fixed-output derivation with tracked Buck products, including the StyleX plugin entry
• make regeneration, freshness checking, recomposition, and publication a deterministic task chain
• record the BUCK-R15 complexity ledger and BUCK-R16 exact-head cache/CI benchmarks in the repository
• run the complete local and CI proof before marking the stack ready

Downstream consumer PRs remain blocked on the final merged effect-utils main SHA.

Prior draft-head evidence

The measurements below describe the pre-takeover draft heads. They are useful baselines, but they do not close the current migration or replace fresh evidence from the rebased final heads.

Prior final evidence

• definitive fleet apply passed at megarepo-all 60fb762d9a2028c026ec3657ae8b1790d1fa1468 with effect-utils 24ce88a08c729df567689260300ed5c767f8fb58 (335.10s)
• exact repeat converged without work: 8 foreign mounts AlreadyCurrent, 0 overlays, 0 recoveries, 0 Buck builds, root changedPaths: [] (62.66s)
• complete admitted authority build passed; unchanged warm run 0.377s
• fresh buck-out reconstruction: 112.94s, 1,241/1,241 remote cache hits (100%), 0 local actions
• hermetic static authority passed in 18.35s; megarepo suite passed 33 files / 621 tests
• final aggregate commit: 60fb762d9a2028c026ec3657ae8b1790d1fa1468
• final authority commit: 24ce88a08c729df567689260300ed5c767f8fb58
• immutable megarepo product: buck2-product-v3-megarepo-f86c88640238dd6799ac46ddf4a7d9c836d38397eedc4f93f32a1f2d197d1be2
• product digest: f86c88640238dd6799ac46ddf4a7d9c836d38397eedc4f93f32a1f2d197d1be2
• PR-B check:quick: 66s baseline to 27.361s (-38.639s, -58.5%)

Prior complexity disposition (BUCK-R15)

Final non-doc/non-test ledger: 705 files, +28,714/-12,286 lines, net +16,428. This is a positive implementation phase, not a LOC reduction. The increase is the shared typed Buck/composition/runtime authority plus explicit generated projections carried across eight consumers. The corresponding operational surface is smaller and closed: 43 governed operations, 3 Buck-owned, 40 outside-policy, no pending or dual authority.

Draft PR stack

1. #1260 — TypeScript product authority
2. #1261 — unit-test authority
3. #1262 — TypeScript authority
4. #1263 — static operations
5. #1264 — editor authority
6. #1265 — platform/Rust and final hub authority
7. final JavaScript/TypeScript closure PR — pending
8. overengineeringstudio/diffstream#74
9. livestorejs/livestore#1622
10. schickling/schickling.dev#198
11. schickling/schickling-stiftung#295
12. schickling/dotfiles#2736
13. overengineeringstudio/overeng#216
14. schickling/megarepo-all#118

#1260 is ready; the remaining implementation PRs are draft. No automatic merge or enrollment is enabled.

Review findings closed

• no-op composition republished unchanged overlays
• capability realization could lose Nix store paths to GC during a long apply
• hermetic oxlint wrapper omitted the tsgolint runtime
• the resulting type-aware lint surfaced a deprecated Bun stream helper; the final test captures stderr through a file sink

Tooling follow-up: schickling/dotfiles#2740 and schickling/dotfiles#2741.

Live work has moved to schickling/megarepo-all#120, rendered from buck2-ledger.json; start there and read the canonical contract under context/buck2/. Review the #1260–#1265 stack and the remaining-work list. The migration is done only after rebasing the stack, completing the bounded authority work, recording fresh BUCK-R15/R16 evidence, and passing the full local and CI proof.