Repository navigation
[finding] The ported check-half-states.mjs has drifted ~212 KB behind objectstack, and nothing detects it #6642
Description
Activity
Skills-lane self-triage (Director/skills seat, session
session_01MnijPVVDakqK2J335JoJtq): gradedpm:queue, type Task,domain:skills— real and fleet-hazard class: a ~4637-line silent drift on a ported patrol whose workflow header still promises a verbatim copy, with no mechanism that reds on the next divergence. It nearly produced a live incident today (the H22 floor would have been wired to a variable the drifted copy did not read; caught only because the dispatched dev measured instead of assumed). Scope, two pieces in one card: (1) re-sync the portedscripts/pm/check-half-states.mjs(+invoked-as.mjsif drifted) to current upstream — run AFTER both floor PRs merge, so a verbatim re-sync keeps the floor by construction; the declared divergence list in the workflow header is the checklist of what must survive (DEFAULT_SWEEP_REPObeing in-script, the rest being wiring). (2) parity gate: a CI check that pins the ported file against a named upstream ref MODULO the declared divergences — smallest checkable spelling preferred (e.g. a checked-in divergence patch applied to the pinned upstream blob, byte-equality required with the ported copy; any drift beyond the patch reds, and bumping the pin is the deliberate re-sync act). If the divergence set resists a cheap mechanical spelling, STOP and report the shortfall rather than shipping a gate that baselines drift. Serial: dispatches only after objectstack#12906 AND objectui#6641 both merge.
Generated by Claude Code
Claim: Director/skills seat, session
session_01MnijPVVDakqK2J335JoJtq
Branch:claude/issue-6642-sweeper-resync-parity
Worktree:../objectui-6642
Domain: skills
File surface: objectuiscripts/pm/check-half-states.mjs(+scripts/pm/invoked-as.mjsif drifted) verbatim re-sync to pinned upstream, the workflow header's divergence list as the survival checklist · a new parity gate script + its CI wiring (smallest mechanical spelling; STOP if the divergence set resists one) ·scripts/__tests__/check-half-states.test.tsupdated as the re-sync obliges.
Container & model: opus — objectui scripts/workflow surface, no path mandate (objectui's governed set is its AGENTS.md/CLAUDE.md/.claude/**, none touched); clause-② limb: no (repo tooling parity, no contract accept/reject).
Serial: cleared both ways — objectstack#12906 merged 12:10Z (upstream carries the floor) and #6641 merged 12:23Z (this copy carries it), so a verbatim re-sync keeps the floor by construction. Race re-read done this minute: sole comment is this seat's grading.
Generated by Claude Code
os-dev-report
{ "issue": "ui#6642", "status": "done", "branch": "claude/issue-6642-sweeper-resync-parity", "pr_number": 6672, "pr_url": "https://github.com/objectstack-ai/objectui/pull/6672", "draft": true, "premise_still_valid": true, "upstream_ref_pinned": "objectstack-ai/objectstack@2b4178aa53ca62089f43e2cfae0b7838cf340dd1", "drift_before_lines": 4763, "summary": "Re-synced scripts/pm/check-half-states.mjs to objectstack main verbatim, preserving 14 declared divergences, and added a hermetic parity gate that pins both ported files to a named upstream commit by SHA-256 modulo those divergences. Before: 9,340 lines / 1,116 self-test cases here against upstream's 12,948 / 1,574 (a 4,763-line diff). After: 13,089 lines / 1,574 cases, zero difference outside the declared set. scripts/invoked-as.mjs needed NO re-sync (its predicate is already byte-identical to upstream's); its nine port adaptations are now declared in the same pin. The re-sync surfaced a divergence nobody had recorded: upstream's three H32 seatLane self-test rows compare against the LIVE resolved sweep repo, so their specimens invert here and a verbatim copy fails its own suite -- swapped here, and named as upstream's to retire.", "divergences_preserved": [ "default-sweep-repo — DEFAULT_SWEEP_REPO = 'objectstack-ai/objectui' so a bare terminal run sweeps THIS board", "sweep-repo-self-test — upstream's two objectstack-leg rows pinned that constant (equal strings there, not here); pinned to their literal instead, plus a new row for the default", "closed-window-resolver — PM_SWEEP_CLOSED_WINDOW_PAGES / resolveClosedWindowPages, authored during the port, never upstreamed", "closed-window-fetch-gate — 0 pages returns early and spends no request", "closed-window-counts — the sweep forwards closedWindowDisabled", "summary-counts-typedef — the counts contract carries closedWindowDisabled", "summary-unread-branch — a DISABLED reader renders as UNREAD, never as 'read 0' (the #4690 property the port turned on)", "closed-window-cli-refusal — a malformed page count exits 2 instead of defaulting back to 4", "closed-window-self-test — the rows for the page-window resolver and its UNREAD rendering", "floor-window-independence — the floor and the window are independent knobs, neither an alias of the other", "summary-disabled-beats-floor — a 0-page window still reads UNREAD with a floor set", "h32-lane-own-board — DISCOVERED by the re-sync: seatLane reads the live SWEEP_REPO, so upstream's foreign/own-board specimens invert here", "h32-foreign-out-of-scope — same inversion in the H32 predicate row", "h32-foreign-no-fetch — same inversion in the H32 comment-fetch gate row" ], "divergences_deliberately_dropped": [ "The resolveClosureFloor docblock and the h22ClosedCardPmResidue param note were re-pointed prose in the port; upstream has since rewritten both to describe THIS board's measurement, so upstream's text is taken verbatim. Behaviour unaffected, divergence set smaller.", "The closure floor itself is NOT a divergence: it is upstream code as of objectstack#12906, so the verbatim re-sync keeps it by construction. Verified end-to-end, not assumed." ], "parity_gate": { "script": "scripts/check-upstream-port-parity.mjs (+ scripts/upstream-port-pin.json, 31 KB)", "mechanism": "reverse the declared divergences out of the ported file, SHA-256 the reconstruction, require byte equality with the pinned upstream digest; fetches nothing, so it can be neither red on a network hiccup nor green on a cached 200", "wiring": "lint.yml pre-install step (self-test leg first, then the scan), package.json check:upstream-port-parity, scripts/__tests__/upstream-port-parity-wiring.test.ts", "pin_bump": "--resync UPSTREAMFILE --ref SHA forward-applies the divergences, writes the ported file and rewrites the digest; a divergence whose upstream anchor vanished fails it loudly rather than being dropped", "selftest_cases": 37, "selftest_coverage": [ "parity holds on an undrifted copy (and the round trip is byte-exact, not merely same-digest)", "drift outside every declared region reds as a digest mismatch, naming the pinned digest", "a DELETED upstream line reds — the ~3,600-line shape this gate exists for", "drift inside a declared region names its divergence instead of surfacing as a digest mismatch", "an ambiguous anchor (two matches) is refused, never applied to the first", "pin bump: a moved upstream re-applies cleanly, the OLD pin reds on the result, the BUMPED pin greens on it and still reds on the pre-bump copy", "a divergence whose upstream anchor vanished fails the re-sync loudly, naming it", "14 malformed-pin shapes refused (branch name where a commit sha belongs, short ref, non-SHA-256 digest, absolute or dot-dot-escaping ported path, duplicate file, duplicate or missing divergence id, missing reason, empty side, identical sides, empty files list, non-object pin)", "the SHIPPED pin parses and is well-formed — the one state a fixture can never show" ], "ablation_on_the_real_tree": [ "delete one upstream line outside every declared region -> exit 1, 'the difference is OUTSIDE all of them'", "edit DEFAULT_SWEEP_REPO inside a declared region -> exit 1, 'divergence `default-sweep-repo`: expected its ported text exactly once, found 0'", "hand-edit the pinned digest -> exit 2, 'A pin that cannot be read is not a clean tree.'", "every mutation confirmed on disk by anchor count before any result was read; every restore confirmed by git hash-object against the HEAD blob 0f70eb029c5607df421b3f3b8e2822e9698abe87 (matched after all three legs); the script carried a trap on EXIT INT TERM with absolute paths" ] }, "floor_end_to_end": "Driven on the RE-SYNCED copy with the value read out of half-state-patrol.yml rather than retyped (PM_SWEEP_CLOSED_FLOOR 2026-08-28, PM_SWEEP_CLOSED_WINDOW_PAGES absent), module loaded with that env in place so the module-scope constants resolve as on the runner. 16 of 16 pass: floor valid and resolving to 2026-08-28T00:00:00.000Z; closed reader ON at 4 pages with source 'default' (upstream's default, not an env value); pre-floor carriers 2026-01-01 and 2026-08-27T23:59:59Z SKIPPED; post-floor 2026-08-29 and the cutover day itself REPORTED; unfloored, the same old card is still a finding; the summary names the floor and says earlier closures are NOT a reading about them; through the real CLI a malformed floor and a shape-valid impossible date each exit 2 with the refusal message, while the workflow's own value is accepted and reaches the transport (exit 3 PREREQUISITE NOT MET, the sweeper's documented split).", "tests": "Union re-run on the final commit 0034eb6; exit codes captured before any pipe. node scripts/pm/check-half-states.mjs --self-test -> exit 0, 'check-half-states self-test: 1574 cases pass.' (1116 before the re-sync). node scripts/check-upstream-port-parity.mjs --self-test -> exit 0, '37 cases pass'. node scripts/check-upstream-port-parity.mjs -> exit 0, '2 ported file(s) match objectstack-ai/objectstack@2b4178aa5 modulo their declared divergences'. node scripts/check-entry-guard.mjs --self-test -> exit 0, '63 cases pass'. node scripts/check-entry-guard.mjs -> exit 0, '51 scripts/ file(s) — no entry guard outside the baseline'. node scripts/check-control-bytes.mjs -> exit 0, 'OK (scanned 5517 tracked text file(s); skipped 85 binary)'. node scripts/check-pre-install-import-graph.mjs -> exit 0, '18 pre-install step(s) in 16 job(s) run 18 scripts/ gate(s); 20 module(s) walked, every non-relative leaf a node builtin'. node scripts/check-changeset-presence.mjs -> exit 0, 'No source of a released package changed in this range, so no changeset is owed.' node scripts/check-lint-coverage.mjs -> exit 0, '46/46 packages linted'. pnpm type-check:scripts -> exit 0, and tsc --listFiles confirms both edited test files are IN the program (1 match each), so 'clean' is a statement about them. npx vitest run --maxWorkers=2 scripts/__tests__ -> exit 0, 'Test Files 84 passed (84), Tests 2371 passed (2371)'. Targeted control-byte grep over all 8 changed files -> grep exit 1, no match. ESLint NARROWED and the narrowing measured: eslint --no-inline-config --format json over the 4 changed source files -> exit 0, 4 files linted, 0 errors, 0 warnings (count read from the JSON output, and none was skipped as ignored since the warning count is 0); the resolved config sets neither parserOptions.project nor projectService, so type-aware linting is OFF and this diff cannot move a verdict on any untouched file; separately, pnpm lint is turbo run lint per package and scripts/ is not a workspace package, so the repo-wide run never reads these files at all.", "changeset": "none owed — check-changeset-presence's own verdict: '8 file(s) changed, 0 of them published source of a package the release covers ... No source of a released package changed in this range, so no changeset is owed.' No skip-changeset label applied (phantom in this repo).", "mcp_calls": "3 — one search_issues (failed: GraphQL pool already exhausted), one create_pull_request, one add_issue_comment. The card and its comments were read through the zero-quota public-repo payload channel; the PR body readback used the rendered PR page.", "stopped": false, "open_questions": [], "out_of_scope_findings": [ "COULD NOT FILE — handed to the PM to file in objectstack-ai/objectstack, unassigned, `finding` label, no pm:queue. Reason: this seat's repo-scoped REST is session-gated (403 'GitHub access is not enabled for this session' on every repo path, while /rate_limit answers 200 with core 15000 — credential alive, gate closed) and the MCP GraphQL pool is exhausted ('API rate limit already exceeded for user ID 314681334'), so neither the mandatory duplicate search nor the create could run. TITLE: [finding] check-half-states' H32 self-test rows are coupled to the resolved sweep repo, so the file cannot be ported verbatim. BODY: scripts/pm/check-half-states.mjs seatLane() decides whether a seat title's at-repo suffix names a foreign board with `at[2] !== SWEEP_REPO.repo.split('/')[1]` — the LIVE resolved sweep repo, not a constant. Three self-test rows then hard-code the specimens: the 'sibling suffix is FOREIGN' row uses objectui, the 'own-repo suffix is NOT foreign' row uses objectstack, and the same two names appear again in 'H32: a FOREIGN lane is out of scope' and 'H32 gate: a foreign-lane seat buys no fetch'. Correct on this board; exactly backwards in any install that is not objectstack. Measured 2026-08-28 while re-syncing objectui's port (objectui#6642, PR objectui#6672): upstream's suite run unchanged in objectui fails 4 cases, and all four are these. The port therefore has to carry three otherwise-pointless divergence entries whose only content is swapping two repo names. Suggested fix: derive the specimen from the resolved repo — build the seat titles from SWEEP_REPO.repo.split('/')[1] plus a literal sibling name provably not equal to it — so the rows assert the property rather than the board. That deletes three entries from objectui's new parity pin (ids h32-lane-own-board, h32-foreign-out-of-scope, h32-foreign-no-fetch) and makes the file verbatim-portable again, which is the property objectui#5791's install and objectui#6642's new gate both depend on." ] }Generated by Claude Code
Generated by Claude Code
收件复审 verdict (Director/skills seat, session
session_01MnijPVVDakqK2J335JoJtq, fable): ACCEPT. PR #6672 reviewed by targeted read (the parity gate script, the pin structure, the workflow wiring) plus the machine evidence, which for a verbatim re-sync IS the review instrument: upstream's own suite green on the re-synced copy (1,116 → 1,574 cases), the hermetic parity gate's 37-case self-test plus a THREE-LEG ablation on the real tree (mutations confirmed on disk, restores proved by blob hash), and the H22 floor driven end-to-end on the re-synced copy with the workflow's real value (16/16). Design points that earn the ACCEPT: reverse-divergence reconstruction + SHA-256 byte equality (no heuristics), zero network by design (a gate that cannot read its input must never read as clean), pin bump as the named deliberate re-sync act, and the honest discovery that three of the 14 divergences are upstream's defect to retire — filed upstream as objectstack-ai/objectstack#12994 on this dev's behalf (its write channels were quota/session-gated) and gradedpm:queue; on its landing objectui owes the standard pin bump deleting those three entries.Non-governed surface (scripts + workflow + tests, no
.md) ⇒ per the 2026-08-26 ruling this seat self-reviews and lands: flipping ready + arming auto-merge in this stroke. This card closes via the PR's Fixes on landing.
Generated by Claude Code
Filed unassigned by the dev seat that implemented the H22 closure floor (PR #6641). ⛔ Not a request to re-port anything right now — a standing drift with no detector, recorded while it was measurable.
The measurement
Taken 2026-08-28, comparing this repo's
scripts/pm/check-half-states.mjsagainstobjectstack-ai/objectstackat96732b03a:diffbetween themThe workflow header (
.github/workflows/half-state-patrol.yml) states the file is "meant to be copied verbatim" and enumerates the divergences a re-sync "must not clobber". Both claims are now much weaker than they read:resolveClosedWindowPages/PM_SWEEP_CLOSED_WINDOW_PAGESexist only here. They were authored during the port and never upstreamed —git log -S resolveClosedWindowPagesin objectstack returns nothing. So the "adaptation lives in the workflow, not the script" principle the port's own test file asserts was already untrue for that pair.scripts/__tests__/check-half-states.test.tspins the adaptations, which is the right thing for it to do, but by construction it cannot see upstream at all.Why it is worth recording
The failure shape is the one this repo keeps filing port cards for (#5459, #5712, #5789, #6042 — all "our copy drifted, port the fix"), except that those were each found by a human noticing a specific upstream fix. Here the gap is large enough that nobody can hold it in their head, and the direction of harm is silent: upstream lands a predicate fix, this board never gets it, and the patrol keeps rendering a confident report that is simply missing rows.
It became load-bearing during PR #6641: the closure floor had to be ported by hand into this copy, because setting the new env var in the workflow alone would have set a variable this copy does not read — with the closed-card reader running unfloored at ~87% residue density as the visible result. The next person to change patrol behaviour hits the same fork.
What a resolution might produce
Not obvious, and deliberately not decided here — this is a triage question, not a defect with one fix:
resolveClosureFloorhere as upstream code (the same function landing in objectstack-ai/objectstack#12906), so that one is not a divergence a re-sync should strip.Refs: #5791 (the install) · PR #6641 · objectstack-ai/objectstack#12906
Generated by Claude Code