feat(buck2): add standalone root and aggregate gates

[schickling-assistant]

Problem

Effect-utils had a tracked standalone Buck root, but ordinary CI still synthesized a self-composition and several aggregate gates retained composition-era prerequisites. A fresh standalone checkout also attempted Genie freshness before publishing the committed-graph dependency views its generator sources require.

Goal

Make the tracked checkout the single normative root for ordinary development and single-repository CI, remove effect-utils CI self-composition, and keep the transition contract explicit in the Buck VRS.

Decisions

• COMP-T01: external consumers and single-repository CI build public members from tracked standalone roots; trust tiers may use separate cache namespaces.
• COMP-R01: the tracked checkout is the normative Buck project root; only explicitly requested cross-repository work may synthesize the paused composed shape.
• COMP-R02: canonical repos/<name> mounts apply only inside that explicit composed shape; the standalone repository maps its canonical cell to ..
• COMP-R06: standalone repositories carry .buckroot; nested member markers do not override an explicitly composed outer root.
• COMP-R07: every supported root shape has a fixed isolation directory; cross-shape action-key parity is not promised and per-invocation isolation is forbidden.
• BUCK-R06: zero re-execution is required between identical same-platform standalone checkouts; the interpretation now cites the second-context key-stability experiment, while standalone-versus-composed comparison remains migration evidence rather than a permanent portability contract.
• Decision 0020 Amendment 4: the one-writable-mount workspace is limited to explicitly requested cross-repository compositions during retirement; effect-utils CI runs from the standalone checkout.
• Decision 0027 Amendment 1: standalone worktrees are again the default for ordinary development, agents, and single-repository CI; composed workspaces are an explicit temporary exception.
• CI no longer runs mr:apply, mr:setup, or composition cleanup. Its checkout is the source root for every lane.
• genie:check runs committed-graph editor bootstrap followed by an editor-generator closure precheck, preserving mutation-free freshness while proving every first-party runtime import is covered.
• TypeScript declaration publication accepts the tracked standalone root only when .buckroot is present and Git reports that same directory as the repository top level. The paused composed compatibility path still requires reciprocal worktree identity; lookalike repos/effect-utils directories fail closed.
• Megarepo command preflight classifies a Git-top-level checkout as standalone only when tracked .buckroot and .buckconfig markers are present and no repos/ mount exists; untracked marker lookalikes and mounted compositions retain the fail-closed legacy checks.
• Editor bootstrap materializes the declared Genie generator import closure (the repository root and OpenTelemetry contract views), emits phase timings, preserves whole-workspace authority checks, and overlaps warm integrity traversals instead of serializing duplicate snapshots.
• CI publishers are scoped to the views each lane consumes, while coalesced Restate, OpenTelemetry-contract, and Playwright publishers retain declared dependency authority.
• CI task traces and cache-reuse diagnostics survive the standalone-root cutover.
• DELTA-002 records the q66-accepted warm-integrity budget: a 62.6-second fresh two-view publication followed by 8.61, 7.18, and 7.76-second warm samples.
• Every standalone Buck entry reconciles the exact BUCK2_NO_REMOTE_CACHE=1 trust posture into a gitignored .buckconfig.local before Buck starts, disabling remote reads and uploads without replacing trusted-context coordinates.
• The second-context experiment resolves and removes DELTA-001: after warming the identical revision, the sandbox reused 1,192 of 1,195 queried actions; the sole local action was the same failing typecheck with an identical digest, proving the earlier 633-local sample was test ordering rather than key instability.
• docs(buck2): resolve second-context reuse delta #1322 is superseded by this PR; its final key-stability record and delta removal are folded into the VRS commit here.
• The paused composed shape and its guards remain only for consumers that still explicitly request cross-repository composition; deletion belongs to L3 cut 2.

Verification

• Pinned Genie --check: all 155 generated files unchanged after folding the bootstrap fix.
• Focused Bun suites for Buck2 projection and GitHub workflow generation: 188 passed, 6 skipped, 0 failed.
• Strict Buck VRS validation reaches only the eight pre-existing metadata errors in decisions 0035/0036.
• Standalone task and materializer guards pass, including tracked-root acceptance, reciprocal composed-root acceptance, lookalike rejection, and a graph with no composition prerequisite on ordinary Buck tasks.
• A pinned Buck run from a detached local-disk checkout passed effect_utils//packages/@overeng/utils:test: 25 files, 233 tests, and 528 local commands with no remote execution or cache use. The daemon connected in about 0.86 seconds and the Buck work completed in about 19 seconds.
• The shallow stage-0 source-input proof passes with the explicit shallow flake reference.
• Standalone cache-posture integration proof: 3 passed, 0 failed; exact 1 disables reads/uploads, non-exact values preserve trusted defaults, and existing local overrides survive managed-block removal.
• The folded second-context record preserves VRS shape and documents identical action digests across normal and sandboxed contexts.
• The folded bootstrap branch passed its clean local-checkout proof before integration; its six commits remain reviewable below the VRS commit.
• q66 closure proof: direct Weaver load passed; the checker accepted all 158 closures and the root-only negative named otel-contract; focused units passed 41 tests/212 expectations; editor integration passed 20 tests/102 expectations.
• The standalone Megarepo classifier passed its local-disk integration suite: 1 file and 7 tests, covering untracked-marker rejection, tracked standalone acceptance, and repos/-mount rejection.
• Pinned Buck passed effect_utils//packages/@overeng/megarepo:test from the local-disk checkout: 34 files, 625 tests, and 612 local commands; a real mr --cwd . status --output json invocation also exits 0 at the standalone root.
• The unbounded OTEL-contract lane passes after excluding the standalone root's buck-out and target trees from the repository seam scan.
• The task-graph checker passes both metadata modes: summarized tasks list --json proves each scoped/full publisher retains executable authority, while the realized registry proves trace identity, canonical package scope, and the whole-workspace unscoped fallback.
• The untouched Restate timing lane passed its one local-disk rerun after the Linux CI flake.
• The nested real-otelite capture passes while a simulated outer CI spool is present: it explicitly isolates nested delivery, then observes the task, compiler-command, and TypeScript phase spans in one trace with child exit 0.
• The downstream bounded aggregate passes from the local-disk checkout: buck2:quick includes the new Weaver targets and Megarepo typecheck, completing 1,212 commands with no remote execution.
• Required GitHub checks completed successfully in run 35798036377.

Complexity

• Ledger: 1,120 lines added, 321 removed, net +799, measured by buck2-ledger.ts at b492ec4b66152fa3193196152b78b48baa33121c.
• Twenty-four commits keep the standalone graph, CI cutover, bootstrap/visibility changes, q66 closure fix and DELTA-002, Megarepo standalone classification and typed probe boundary, repository-scan isolation, cross-mode task metadata checks, nested OTEL capture isolation, and final VRS amendments reviewable in order.

Concerns

• Public pull-request lanes are intentionally cache-cold until dotfiles#2986 deploys the public read-only tier; that public address will replace dev3 for public lanes. The trusted remote-cache proof remains main-only and must not be weakened on this PR.
• Cross-shape action-key parity is deliberately not a long-term promise.

Friction & bottlenecks

• Measurements from /srv/bulk are invalid for Buck budget decisions because dotfiles#2955 makes filesystem walks pathological (roughly 60 seconds for the pruned tree versus 0.17 seconds on local disk). The local-disk control run above proves daemon startup and the representative test target without that host artifact.
• Run 35729756932 exposed four independent clean-CI defects: missing exported-symbol JSDoc, an over-broad standalone-root lookalike, a shallow-flake revCount read, and a Storybook fixture that assumed node on PATH. All four are fixed at the replacement head.
• q66 folds the functional generator-closure fix and public-safe DELTA-002 record into this PR. Run 35798036377 exercises the replacement head b492ec4b66152fa3193196152b78b48baa33121c.

Follow-ups

• Delete the remaining explicitly composed effect-utils shape when downstream edges flip at L3 cut 2.
• Complete the separate public cache tier before expecting the trusted main-only proof lane to pass.

References

• context/buck2/05-composition/requirements.md
• context/buck2/.decisions/0020-one-writable-mount-workspaces.md
• context/buck2/.decisions/0027-composed-default-worktrees.md
• context/buck2/requirements.md
• context/buck2/.experiments/2026-09-19-second-context-key-stability.md

Posted on behalf of @schickling

field	value
agent_identity	dev3.direct.omp.t3mkm4sd
session	unknown
agent_persona	generalist
agent_supervisor	unavailable
agent_tool	OMP
agent_tool_version	18.1.14
agent_runtime	OMP 18.1.14
tooling_profile	dotfiles@5d1e05f

[schickling-assistant]

@codex Please review this pull request for merge readiness, with particular attention to the standalone-root contract, CI cutover, and the documented VRS amendments.

Posted on behalf of @schickling

field	value
agent_identity	dev3.direct.omp.fkazdy6v
session	unknown
agent_persona	generalist
agent_supervisor	unavailable
agent_tool	OMP
agent_tool_version	18.2.6
agent_runtime	OMP 18.2.6
agent_model	openai-codex/gpt-5.6-sol
worktree	2026-09-22-fix-checks/2026-09-22-fix-checks
tooling_profile	dotfiles@965ef7e-dirty

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b492ec4b66

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Route the typecheck job through the aggregate

In the CI typecheck context (.github/workflows/ci.yml.genie.ts:412) and the documented standalone command, buck2:check is invoked directly, but this new implementation only audits two providers. The admitted TypeScript targets moved to buck2:quick, whose dependency points from buck2:quick to buck2:check, so invoking buck2:check never reaches them; a PR containing TypeScript errors can therefore receive a green required “Type check (Buck)” result. Keep the TypeScript build behind this command or update every direct caller to use the aggregate.

AGENTS.md reference: AGENTS.md:L7-L10

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Do not close the reuse delta with a red replay

This evidence does not satisfy the cited BUCK-R06/REUSE-R02 acceptance condition: both //:quick builds exited nonzero, and the second context still executed one action locally, whereas the VRS requires zero locally executed actions for an unchanged warmed target (context/buck2/04-reuse/spec.md:46). Equal digests diagnose the original cache misses, but a failing action cannot populate the successful cache entry needed to test the zero-reexecution contract, so removing DELTA-001 here leaves the requirement without its claimed proof; retain the delta until the same experiment completes green with zero local actions.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-23T00:41:08.310975Z	b492ec4	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.