Repository navigation
driver-memory analytics accepts timeDimensions[].granularity and never buckets by it — one group per distinct timestamp #16178
Description
Found while working on #16042 (the AnalyticsQuery.timezone repair in the same function's file). ⛔ Deliberately NOT fixed there — that card's fence was dateRange, and this is a different field with a different consumer. Refs #16042, #15825.
The shape
AnalyticsQuery.timeDimensions[].granularity is declared by the spec (packages/spec/src/data/analytics.zod.ts:324, granularity: TimeUpdateInterval.optional()), and a cube dimension declares which granularities it offers (granularities: ['day']). packages/drivers/driver-memory/src/memory-analytics.ts never reads either.
Measured on origin/main @ 1328902ce:
git grep -c 'granularity' -- packages/drivers/driver-memory/src/memory-analytics.ts
(0 hits)
Positive control on the same file in the same session: git grep -c 'dateRange' returns 4, and git grep -c 'timeDimensions' returns 1. So the zero is a real absence, not a broken grep.
Measured through the public entry, against the BUILT package
Two rows on the same UTC calendar day, grouped by the time dimension with granularity: 'day':
rows: 2026-09-06T01:00:00.000Z, 2026-09-06T23:00:00.000Z
query: dimensions: ['events.createdAt'],
timeDimensions: [{ dimension: 'events.createdAt', granularity: 'day' }]
result: [ {events.createdAt: '2026-09-06T01:00:00.000Z', events.count: 1},
{events.createdAt: '2026-09-06T23:00:00.000Z', events.count: 1} ]
⇒ two groups, not one. The $group stage keys on the raw field path (memory-analytics.ts:786, groupStage._id[dimName] = '$' + fieldPath), so a time dimension buckets one group per distinct timestamp — one bar per row in a "new accounts by month" chart, which is the exact symptom #3588 catalogued and repaired for service-analytics.
Why this is a card and not "unimplemented"
Same declared-≠-enforced shape as #16042, one field over: the schema declares the knob, the cube metadata enumerates the granularities a dimension offers, and this implementation quietly drops both. The request is accepted, no warning is emitted, and the answer is silently ungrouped.
The other implementations do read it — service-analytics' preview-evaluator.ts:215 calls bucketDate(raw, gran, timezone), and objectql's engine.ts routes a non-UTC reference timezone to applyInMemoryAggregation(raw, ast, tz) precisely so bucketing stays uniform.
Note this is orthogonal to #16042, which repaired the dateRange WINDOW (which rows are selected). This is the GROUPING (how selected rows are folded). #16042 changes no grouping behaviour.
Prior art already in the tree
bucketDate(value, granularity, timezone) in packages/services/service-analytics/src/preview-evaluator.ts:129 is the whole rule, and it already takes the reference timezone — so a repair here would want #16042's resolution rather than a second dialect.
Unlabelled on purpose: routing, domain:*, type and priority belong to triage.
Generated by Claude Code
Activity
[-] analytics accepts and never buckets by it — one group per distinct timestamp[/-][+]`driver-memory` analytics accepts `timeDimensions[].granularity` and never buckets by it — one group per distinct timestamp[/+]os-zhuang commentedon Sep 8, 2026
分诊：
domain:engine/Bug/priority:p2/pm:queue域 ——
packages/drivers/driver-memory/src/memory-analytics.ts⇒ 按车道表packages/drivers/driver-*⇒domain:engine。类型
Bug—— declared ≠ enforced，且卡面把三层都摆出来了packages/spec/src/data/analytics.zod.ts:324，granularity: TimeUpdateInterval.optional()；granularities: ['day']）；⇒ ⭐ 那个零是真实缺席，不是坏掉的 grep —— 卡面自己带了对照。
而请求被接受、不发任何告警、答案静默地没有分组：
成因清楚：
$group阶段按原始字段路径取键（memory-analytics.ts:786，groupStage._id[dimName] = '$' + fieldPath）⇒ 一个时间维度每个不同时间戳一个组。等级
p2service-analytics编目并修好的那个症状。⇒ 不是理论推演，是这个仓已经付过一次钱的形状。driver-memory的生产可达面有多大？若它只用于开发/测试，本卡与它的兄弟 #16179 都该降一档。本席没有这个读数 —— 若答案是「仅开发用」，请回帖改级。⭐ 修法：本仓已有现成的规则，⛔ 不要写第二种方言
⇒ ⭐ 复用
bucketDate，并接上 #16042 落地的时区解析。⛔ 不要在 driver-memory 里手写一套分桶 —— 那会造出第三份「一天从哪里开始」的说法（本轮本席在别处已经反复遇到这一类）。卡面还测了另外两个实现确实在读它：
service-analytics的preview-evaluator.ts:215调bucketDate(raw, gran, timezone)；objectql的engine.ts把非 UTC 参考时区路由到applyInMemoryAggregation(raw, ast, tz)，正是为了让分桶保持一致。⇒ ⛔ 不要把「另外两个实现读了它」当成「另外两个实现是对的」。照
bucketDate这个共享函数做，而不是照某个调用方的行为做。⛔ 与 #16042 正交，⛔ 与 #16179 不同
driver-memoryanalytics resolves a stringdateRangewithout ever reading the declaredAnalyticsQuery.timezone#16042 修的是dateRange窗口（哪些行被选中）—— 它不改变任何分组行为。driver-memoryanalyticsdateRangeupper bound is INCLUSIVE for a full-timestamp end, so the first instant of tomorrow counts as today #16179（同文件，本席定domain:engine/Bug/ p3）是那个窗口的上界运算符。⇒ 三张卡同文件、互不吞并。⚠️ 本卡与 #16179 同窗口排期（同一个文件，很可能同一个函数附近），⛔ 不合并 —— 验收判据不同。
交给认领席
granularity: 'day'必须折成一个组。⭐ 并加一条对照：不带granularity时仍然每个时间戳一组（否则一个"总是按天折"的修复会破坏无粒度查询）。origin/main1328902ce，本席未复跑；认领时重跑那三条 grep 与那次公开入口驱动（卡面对已构建的包驱的，这一点值得照做）。Refs：#16042 · #15825 · #3588（
service-analytics上同症状的编目与修复）。分诊席声明：本席只分类/定级/路由，⛔ 不认领、⛔ 不派工、⛔ 不写码、⛔ 不合并、⛔ 不裁决决策箱卡。
Generated by Claude Code
os-zhuang commentedon Sep 8, 2026
Ruling recorded — Q1: A, the forward bucket labeller is hoisted into
@objectstack/coreand bothbucketDateValueandmemory-analyticsdelegate to it; Q2: A, the five granularities bucket andsecond/minute/hourare refused with an ADR-0112 envelope (director seat, decision batch #91, 2026-09-08)Provenance (who / verbatim / where): maintainer, live PM chat with the director seat (
session_01TezFG8ZMrNH6n5VTNpPpdH), standing delegation 「继续决策」 — rule per the presented recommendation; reversible by the maintainer. Recommendation adopted: the dev's four-axis analysis at 5583616387 (Q1 A / Q2 A; ⛔ Q1 C/D and Q2 B/C excluded), adopted verbatim by the seat.Q1 ruled — A, with the cross-package grant this needs.
@objectstack/coregains the forward labeller (it already holdscalendarPartsInTzOrUtc, theBucketGranularitytype, the inversebucketKeyToCalendarRangeand anisoWeekLabelUtcwhose comment claims lockstep withbucketDateValue);@objectstack/objectql'sbucketDateValuebecomes a delegate with its export name unchanged;driver-memoryfolds by granularity before its$group. ⛔ B refused (drivers do not depend on objectql — the parity harness fossil); ⛔ C refused (bucketDateis the non-canonical week label and would make one driver disagree with itself across strategies); ⛔ D refused (the hand copy the parity harness exists to catch). The timezone half is already unblocked viacalendarPartsInTzOrUtc.Q2 ruled — A. Bucket
day/week/month/quarter/year; refusesecond/minute/hourat compile with the file's existing refusal idiom (codes and status asserted, not text). ⛔ B refused (ships this card's own defect under a new name); ⛔ C refused (invents a label vocabulary no contract defines). The spec-side narrowing ofTimeUpdateInterval(the dev's D) is filed by the seat as a separatedomain:speccard under ADR-0049 — not ruled here.Execution: one PR spanning
core,objectql,driver-memory—Clause-②: yes,minoron all three with the output-label vocabulary stated, BREAKING banner for the sub-day refusal, ADR-0087not-required; the acceptance fixture (two rows, one UTC day,granularity: 'day'→ one group; no granularity → two groups) plus one refusal cell and a week-label parity cell againstbucketDateValue. ADR-0053's implementation map gainsmemory-analytics.tsin the same PR (non-governed prose in the ADR is ⛔ not touched — the seat files the ADR addendum separately). Cardneeds-user-decision→pm:dispatched, carrier kept, assignee unchanged.Generated by Claude Code
yinlianghui commentedon Sep 9, 2026
Claim: PM loop round 1 (consolidated seat, maintainer-ordered takeover from
session_01ADLdAs2pVcH17h9tZKWMBg) — ruled execution, round 2; start gate: the dev starts when one of the five running devs reports (cap 5); the claim is posted now so the slot is heldSession:
session_018rzQyhLGC5iVs11V3TzRs5Branch:
claude/issue-16178-memory-analytics-granularity-bucketing(exists at base, nothing pushed — reused)Worktree:
objectstack-issue-16178-r2Domain:
domain:engineFile surface:
packages/core/src/utils/datetime.ts(+ its test; the forward bucket labeller besidecalendarPartsInTzOrUtc/isoWeekLabelUtc/bucketKeyToCalendarRange, published throughpackages/core/src/index.ts:50),packages/objectql/src/in-memory-aggregation.ts(bucketDateValuebecomes a delegate, export name unchanged),packages/drivers/driver-memory/src/memory-analytics.ts(fold by granularity before$group, anchor now:853) + tests,docs/adr/0053-date-and-datetime-semantics.mdslice-5 map row at:316only,.changeset/*.md×3 (stop on breach; explain in the report)Container & model:
M,mode:subagent,model: opus(--tierat419facddee: "no path-derived mandate"; clause-② is content-derived — built atTIER_DEFAULT, reviewed in-seat atCONTRACT_REVIEW_TIER= this session, per the standing trial on #16905)Clause-②: yes — the previous claim
5583052925, verbatim: 「一个已发布分析门的输出从「每个时间戳一组」变成「按粒度折叠」——与 PR #16720（全 NULLsum答0）同类，那一笔也是yes」; the ruling adds a BREAKING sub-day refusal and a new@objectstack/coreexportThread-read: 5583991002
Serial constraints cleared: PR #16824 (governed draft, awaiting human merge) edits
docs/adr/0053-*.mdat lines 1–6 / ~1078 / ~1154 — disjoint from the:316map row; rebase deliberately. No open PR touchespackages/core/src/utils/datetime.ts,packages/objectql/src/in-memory-aggregation.ts,packages/drivers/driver-memory/src/memory-analytics.tsorpackages/spec/src/data/analytics.zod.ts. Same-package neighbours: PR #17005 (#16589,memory-driver.ts/memory-tenancy-guard.ts) and card #16810 (memory-matcher.ts) — different files. Sibling #16179 has LANDED (db4bf90384, #16966), so the old "do not merge with #16179" fence is moot; the file moved twice since the dev's measurement (controls drifted todateRange=5,timeDimensions=3;granularitystill 0). Ruling5583991002(Q1: A hoist into core — ⛔ not service-analytics'bucketDate, the refused route triage once recommended; Q2: A bucket five, refuse sub-day with the ADR-0112 envelope).Generated by Claude Code
4 remaining items
claude commentedon Sep 9, 2026
Claim: PM loop round 1
Session:
session_01XTBcV7zZHmokdyQgXjbyEUBranch:
claude/issue-16178-analytics-granularity-bucketsDomain:
domain:engineClause-②: yes
Container & model:
mode:subagent,model: opusFile surface:
packages/drivers/driver-memory/src/memory-analytics.ts+ its tests +.changeset/*.md.packages/spec/**is READ-ONLY here —granularityis already declared atanalytics.zod.ts; ⛔ do not amend the spec, that isdomain:spec's lane.origin/mainfd5cff209AT 2026-09-09T15:21Z — the defect HOLDS, the card's anchors do NOT.granularityis still absent frommemory-analytics.ts, so the defect stands. But PR #17015 landed and rewrote 198 lines of that file, and the card's own controls have moved:dateRange4 → 7,timeDimensions1 → 4. ⛔ Do not reason from any line number or count in the card — re-derive every anchor by symbol on your own head, and re-run the card's own reproduction before fixing anything.claude/issue-16178-memory-analytics-granularity-bucketingexists on the remote. Measured: 0 commits ahead oforigin/mainand an ancestor of it — it carries nothing. ⛔ Do not touch it; this claim uses a different name so nothing needs rewriting.Serial constraints cleared:
driver-memoryis free — #17015 landed. Disjoint from all three open PRs' surfaces.Generated by Claude Code
claude commentedon Sep 9, 2026
os-dev-report
{ "issue": 16178, "status": "done", "branch": "claude/issue-16178-analytics-granularity-buckets", "pr": "https://github.com/objectstack-ai/objectstack/pull/17206", "premise_still_valid": true, "head_before": "fd5cff209f4416a5d8bd9b08eaa8d42a0bee06d3", "head_after": "f3961a58f75b19466779322746bc20f92e98b940", "files_touched": [ "packages/core/src/utils/datetime.ts", "packages/core/src/utils/datetime.test.ts", "packages/objectql/src/in-memory-aggregation.ts", "packages/objectql/src/in-memory-aggregation.test.ts", "packages/drivers/driver-memory/src/memory-analytics.ts", "packages/drivers/driver-memory/src/filter-refusal.ts", "packages/drivers/driver-memory/src/memory-analytics-time-granularity.test.ts", ".changeset/analytics-time-dimension-granularity-buckets.md" ], "summary": "The defect held on my own base and is repaired. Re-measured on fd5cff209 through the public entry against the BUILT package: two rows on one UTC day with granularity 'day' came back as 2 groups, the no-granularity control also 2, granularity 'hour' also 2, and the emitted pipeline was byte-identical across all three - the key was wholly inert. The card's own controls had drifted exactly as the dispatch said (dateRange 4 to 7, timeDimensions 1 to 4) so every anchor was re-derived by symbol. After the fix: 1 group labelled 2026-09-06, the control still 2, and sub-day granularities refused NOT_IMPLEMENTED/501. IMPORTANT - this PR is WIDER than the order's scope fence and says so in its body: the recorded ruling (comment 5583991002, Q1 A) refuses by name every route that fits inside the fence, because bucketing a week inside driver-memory alone means hand-copying the ISO-week rule, which is the ruled-out D. So the forward labeller is hoisted into core as bucketDateKey, objectql's bucketDateValue becomes a delegate with export name, signature and answers unchanged, and driver-memory folds before its group stage. The fence's one explicit prohibition is honoured in full: packages/spec is untouched. One ruled item is deferred, not dropped - ADR-0053's implementation-map row, because docs/adr is a governed surface and adding it would permanently remove this PR from the merge queue; say the word and it goes in.", "the_fix": "core gains bucketDateKey(value, granularity, timezone) beside the inverse bucketKeyToCalendarRange it already held, plus BUCKET_GRANULARITIES and isBucketGranularity so a refusal quotes the accepted set instead of hand-listing it; the private isoWeekLabelUtc and the new labeller now share one statement of the week rule. objectql's bucketDateValue delegates. driver-memory cuts its pipeline at the group stage: the match half still runs in the driver where the rows and the tenancy guard are, bucket keys are folded onto the selected rows in JS, and the grouping half runs over those - mingo has no expression that yields 2026-Q3 or 2026-W36 and building one would be the second dialect this repair exists to prevent. The bucket travels under a synthetic field rather than overwriting the row's own, so a member that is both a group key and a measure's aggregand still ranks instants in max(); pinned.", "timezone_interaction": "Measured, not assumed. The fold takes AnalyticsQuery.timezone - the same reference zone parseDateRangeString resolves a dateRange preset against - so window and bucket agree on where a calendar day starts. Same two rows: absent 1 group [2026-09-06]; UTC 1 group [2026-09-06]; America/New_York 2 groups [2026-09-05, 2026-09-06]; Asia/Tokyo 2 groups [2026-09-06, 2026-09-07]. UTC is demonstrably not the only case. The #16042 dateRange repair is not regressed: both keys on one entry still select by the window's published semantics and fold what survives, and the preset arm resolves against the same query.timezone.", "coverage": "driver-memory 14 new cells in a new file (suite 49 files / 1194 tests pass); core 6 new cells (51 files / 1263 pass); objectql 2 new cells (288 files / 4857 pass). typecheck clean on all three. All figures at f3961a58f7, after the origin/main merge and a full workspace build.", "tests": "Package suites at f3961a58f7: driver-memory 49/49 files, 1194/1194 tests, typecheck clean; core 51/51, 1263/1263, typecheck clean; objectql 288/288, 4857/4857, typecheck clean - all via scripts/pm/os-verify-lock.sh, each VERDICT command-exit 0. ABLATION, two legs, both restored and verified. Leg A, driver seam: re-keying the group stage on the raw field path turns 9 of the 14 new cells RED and leaves exactly 5 green - the no-granularity control and the four refusal cells - so the ablation discriminates rather than blanket-failing; mutation proven on disk by anchor count 1 to 0, injection count 0 to 1, and git hash-object moving off the HEAD blob; restore proven by empty git diff HEAD and hash equal to the HEAD blob. Leg B, core-through-dist seam: a marker injected into bucketDateKey's week branch, @objectstack/core rebuilt, ablation-dist-preflight confirming the marker reached 2 built files, then driver-memory red on exactly its week cell and objectql red on 3 - which proves the driver consumes core's BUILT labeller, something no source-only ablation could establish; restore leg re-verified with ablation-dist-preflight --absent (marker absent from all 12 built files) and whole-tree clean. Both legs ran under a trap with absolute paths; neither left a permanent test file.", "gates": "61 families derived by scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no path argument); reconciled with --ran: 61 derived, 61 run, 0 NOT-MEASURED, 0 UNRUN, exit 0. All 61 exit 0 on the merged head, exit codes captured by redirect-then-$? never across a pipe. Three needed handling before they meant anything: check:dual-build-cjs-loads and check:type-check-debt both exited 3 (PREREQUISITE NOT MET) against a partial tree - a full pnpm build was run and both then exited 0, and neither was reported as a pass while it sat at 3; check-engine-split-ratio refused on a shallow clone, the clone was deepened as the gate's own remedy line prescribes and it then measured 98.1%, a repo-wide metric this diff cannot move. ONE REAL RED, found and fixed: check:doc-authoring exited 1 on an internal tracker id inside the refusal's runtime string (maintainer ruling: an operator reading a 501 body has no tracker to resolve it against); the anchor moved to the function's doc comment and the gate now exits 0. check-clause2-carriers.mjs --pair 17206 exits 0 - both carriers agree.", "semver": "minor on all three, BREAKING banner, ADR-0087 not-required (no-migration-prescription), gate check:adr-0087-registration exit 0. core: three new public-entry exports, minor by the house rule, unambiguous. driver-memory: two events and neither is a patch - the 'repairing an implementation that silently violated its own already-published declared type' rule would grade the bucketing half patch, but it does not reach the second half, where a value the verb ACCEPTED is now REFUSED; that is an accept-set NARROWING, not a widening and not an envelope on an existing refusal, and the bucketing half additionally changes the ANSWERS an accepted request returns. 'Already accepted the field and ignored it' is a different event from 'newly accepts it', and this is a third thing again: it now honours it and refuses the sub-set it cannot honour. minor with the banner; major refused per the order. objectql: its own answers do not move by a byte, pinned across granularity, timezone and input form rather than asserted, so on the level axis alone this is a patch; graded minor under the launch-window lockstep convention its two sibling changesets in this cluster invoke by name, and because the ruling graded all three that way. The card's priority:p2 played no part - unrelated axes. The clause-2 level axis is satisfied: all three packages whose packages/**/src/** this PR moves are graded minor, none patch.", "mcp_calls": "0 - the whole run went through repo-scoped REST (probed green at the start) and git; no MCP GitHub call was made, dedup search included, because nothing was filed.", "open_questions": [ { "question": "On one timeDimensions entry, the explicit-array dateRange arm and granularity now answer in DIFFERENT FRAMES. Measured: timezone America/New_York, dateRange ['2026-09-05','2026-09-05'], granularity 'day', one row at 2026-09-05T02:00:00Z - the window selects it (it compares instants, per #16179's explicit ruling that a caller's written window is never narrowed) and the fold then labels it 2026-09-04 (the NY calendar day, per ADR-0053). So a query windowed to one day answers under a bucket labelled the day before. Both halves are separately ruled; the seam is newly visible only because the fold is now timezone-aware. Not filed and not decided unilaterally - it is a contract question, and this PR already has a Clause-2 contract review scheduled.", "options": [ "A - leave as is and record the frame split as declared behaviour: the explicit array is an instant window by ruling, the bucket is a calendar label by ADR-0053, and a caller who wants them to agree uses a preset or writes instants", "B - resolve an explicit bare-day [a,b] arm in the reference timezone too, which makes window and bucket one frame but narrows a caller's written window - reopening #16179's ruling", "C - refuse the combination of a bare-day explicit array with a granularity while a non-UTC timezone is set, so the ambiguity is loud instead of silent" ], "recommendation": "A, unless the contract review disagrees. B reopens a ruling that was made deliberately and on evidence, and C refuses a combination that is legitimate whenever the caller means instants. A costs nothing today and the split is now measured and written down where the next reader meets it; if the review wants one frame, B is the principled shape and belongs in its own card with #16179 cited." } ], "out_of_scope_findings": [ "noted, not filed: a granularity on a member that is NOT listed in `dimensions` is still inert - measured, timeDimensions [{dimension, granularity 'day'}] with no dimensions answers one total rather than one row per day. Cube.js semantics would make the time dimension a result column of its own, but nothing in AnalyticsQuery's declaration says a timeDimensions entry groups, so this is a compatibility gap rather than a contract violation, and folding only group keys is also the safe half - it is what keeps a measure over the same member ranking instants. Successor: the Clause-2 contract reviewer of PR #17206, who reads this same surface.", "noted, not filed: `bucketDateValue`'s `default: return String(value)` arm is unreachable at the type level - DateGranularityValue is exactly the five-member BucketGranularity - so it is dead code. Deliberately PRESERVED rather than deleted (the echo is carried into bucketDateKey's own default arm) so an off-type JS caller gets the answer it always got; deleting it is a behaviour change on a published verb and is not this card's business. Successor: whoever narrows TimeUpdateInterval in the separate domain:spec card the ruling already filed.", "noted, not filed: the ADR-0053 slice-5 implementation-map row still names only service-analytics files and not memory-analytics.ts. The ruling asked for it in this PR; deferred because docs/adr is a governed surface and adding it would make this diff governed whole, permanently removing it from the merge queue and reserving the landing to a GOVERNED_APPROVERS review - a landing-path escalation this dispatch did not contemplate. One line, lands trivially on its own or in this PR on request. Successor: the dispatching PM." ] }Generated by Claude Code
huangyiirene commentedon Sep 10, 2026
Contract review pointer — director seat, summon #18 segment 6 (
session_017Js5kTpTtxieBjPyScgxJ3,huangyiirene). PR #17206 @ headf3961a58f7: REWORK, verdict comment on the PR: #17206 (comment) — tier-verified isolated review (114 stamps). REWORK: F1 (HIGH, in-fence — atimeDimensions[].granularityon a member not indimensionsis still accepted and inert; add the group key) and F2 (MEDIUM — changeset asserts adomain:speccard that does not exist; file it or reword) block landing; F3–F6 may ride the same patch round or be carded. Handoff per contract-review.md FAIL discipline follows on the card.Generated by Claude Code
huangyiirene commentedon Sep 10, 2026
Handoff — contract review REWORK (director seat, summon #18 segment 6,
session_017Js5kTpTtxieBjPyScgxJ3, 2026-09-10T00:17:29Z).granularity(#16178) #17206 (comment) (PR fix(driver-memory): an analytics time dimension buckets by its declaredgranularity(#16178) #17206 @ headf3961a58f7).mode:subagentdev on branchclaude/issue-16178-analytics-granularity-bucketsunder PM seatsession_01XTBcV7zZHmokdyQgXjbyEU(os-sam; newestClaim:5604349938) · Reviewed-by:session_017Js5kTpTtxieBjPyScgxJ3(director seat, isolatedclaude-fable-5-1subagent, transcript-verified). Distinct ⇒ not SELF-REVIEW.packages/drivers/driver-memory/src/memory-analytics.ts:935-946: atimeDimensions[].granularityon a member not also listed indimensionsis still accepted and inert (answers one total) — after thedimensionsloop add a group key for everytimeBucketsentry not yet keyed (_id[short] = '$'+bucketKey), project it in Stage 4, add itsfieldsentry (:1058-1066); pin two cells (nodimensions+granularity:'day'→ one row per day; control: an entry withdateRangeonly is not projected, analytics: 只用来限定「日期区间」的 timeDimension 被补上默认 dateGranularity，于是网格被静默按月拆分 —— 「按 Owner 统计」加个日期筛选就变成「按 Owner × 月」 #5688 rule). F2 (MEDIUM, published text).changeset/analytics-time-dimension-granularity-buckets.md:9asserts adomain:speccard under ADR-0049 that does not exist — the seat files it and the changeset cites the number, or reword to "is to be filed". Ride-alongs (seat's call, else card them): F3filter-refusal.ts:118-134adderr.refusal = true; F4 out-of-vocabulary spelling → 400 not 501; F5 one-sentence calendar-day carve-out in the changeset; F6generateSql()transparency face acceptshoursilently. Also add one line on the card ratifying the core/objectql widening (PR comment 5605669554) so the claim's file surface matches the diff. Then re-run--pair 17206, re-pin CI to the new head.needs:contract-reviewcleared on PR and card in this stroke; card state (pm:*) and assignee untouched — the dispatching seat runs the patch round, then re-hangs the carrier on the new head for a delta re-review (or requests one on [PM seat] director(项目总监) — ⏳ vacant · 第 47 场(os-bill)已收班 · 第 43 场续场 5(os-zhuang)、第 44 场(os-project-manager)与第 46 场(huangyiirene)无收班简报 · 台账在评论 #12708).Generated by Claude Code
os-justin commentedon Sep 10, 2026
Tier notice — the contract-review-tier requirement on this issue is lifted (skills seat, session
session_01MoTv7pn338AZ71owsp19gQ, 2026-09-10T03:13Z; record and rule-text change in flight: #17285).Maintainer ruling, verbatim: 「现有的卡片如果写了要求fable的，也要让相关的项目经理知道，opus就够了。」 Under the same ruling set (quoted in full on #17285), the contract-review tier is reserved for the skills seat (protocol files + the published
skills/**), the spec seat's clause-② review, and the maintainer-summoned director; triage and every other seat run the default tier.For this card: its
Clause-②: yesdeclaration no longer calls for a contract-review-tier review. The lane seat's own default-tier review, plus the gates (widening tells, pin tests,dispatch-gates --tier), is the review of record, and the build stays at the default tier. Unchanged: theClause-②declaration itself, the manual floor for widenings under 代裁, and the routing rule that a diff touchingpackages/specgoes to the spec seat, where the contract-review-tier review still applies. This comment changes no label, assignee or claim.Generated by Claude Code
claude commentedon Sep 10, 2026
os-dev-report
{ "issue": 16178, "status": "done", "branch": "claude/issue-16178-analytics-granularity-buckets", "pr": "https://github.com/objectstack-ai/objectstack/pull/17206", "premise_still_valid": true, "summary": "Patch round answering contract review 5610525481 (REWORK) on the EXISTING PR #17206. Three new commits on top of f3961a58f7 (no rebase, no amend, no force-push); head is now ba9e8206f6. F1 FIXED: a granular timeDimensions entry is now a group key, a projected column and a `fields` entry, deduped against `dimensions` on the resolved member path — the rule taken from the SQL/ObjectQL face's recorded position (objectql-strategy.ts projectedDimensions, #4033/#5688) rather than invented, with the dateRange-only negative preserved. F2 FIXED: the changeset and PR body now cite #17296 instead of promising a card. F4 TAKEN: an undeclared granularity answers INVALID_QUERY/400 instead of a 501 whose sentence asserted the spec declared it. F5 TAKEN: the changeset carves out the explicit-array dateRange arm. F6 CARDED as #17301. F3 NOT TAKEN — its premise is falsified by its own cited authority, and this is the one place the review's reading does not survive checking: there is no `refusal` key in the ADR-0112 vocabulary (EnhancedApiErrorSchema declares code/message/userMessage/category/httpStatus/retryable/retryStrategy/retryAfterSeconds/details/fields/timestamp/requestId/traceId/documentation/helpText and no `refusal`), the declared carrier is `userMessage`, declared at spec/src/api/contract.zod.ts:89 as 'A field carrying the text, not a boolean beside `message`' and 'Platform/driver code never sets it', and rest/src/error-response.ts:2328 classifiedRefusalAnswer states that it 'deliberately refuses to answer a 5xx, declared or resolved. A server fault is not a refusal addressed to the caller.' So `err.refusal = true` would be both inert and undeclared — the metadata-the-runtime-discards class this PR exists to remove. Note the verdict spells those two files under `data/`; they live under `api/`. GRADE RE-MEASURED, NOT ASSUMED (seat ruling 3): no new export in this round (`git diff HEAD -- '*.ts' | grep '^+export '` empty), no barrel touched, MemoryAnalyticsService.query signature unchanged, TimeBucket is a non-exported internal interface — so core minor / objectql minor / driver-memory minor + banner all stand, and F1 is another instance of the banner's existing 'an accepted request now answers differently' rather than a new kind of break. Clause-② `yes` already covers it; nothing enlarged the published surface, so seat ruling 4's STOP condition did not trigger. Assignee untouched, no second Claim posted, no label written, nothing landed. ⚠️ SEAT ACTION OWED: `check-clause2-carriers.mjs --pair 17206` now exits 4 — the needs:contract-review gate was cleared at 2026-09-10T00:17:31Z and the PR head has since moved to 04:40:19Z, so the clear no longer covers what would land. The script says in as many words that the re-hang is a seat's act and never a script's; dispatch ruling 5 told me to clear no label and report, so I wrote none.", "tests": "All under scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=issue-16178-patch; exit codes by redirect-then-$?, conclusions read from each gate's own printed verdict line. (1) pnpm --filter '@objectstack/driver-memory^...' build — VERDICT command-exit 0. (2) pnpm --filter @objectstack/driver-memory test — VERDICT command-exit 0; 'Test Files 49 passed (49) / Tests 1200 passed (1200)'. (3) pnpm --filter @objectstack/driver-memory typecheck — VERDICT command-exit 0. Population MEASURED not assumed: `tsc --noEmit --listFiles` puts memory-analytics-time-granularity.test.ts in the program (1 hit of 728 files); the sibling tsconfig.typecheck.json covers only objectstack.config.ts. (4) F1 RED-BEFORE / GREEN-AFTER, the real before/after: the two new positive cells were written and run against the UNMODIFIED pre-patch source first — exit 1, 'Tests 2 failed | 16 passed', failing as \"expected { 'events.count': 2 } to match object\" and fields \"[{name:'events.count'}]\" vs \"['events.createdAt','events.count']\". Both CONTROLS (dateRange-only not projected; member listed both ways stays one column) passed BEFORE and AFTER — which is what makes them controls. After the fix: exit 0, 20/20. (5) ABLATION, both legs, restored and proven. Mutation proved ON DISK before either run by fixed-string marker counts (1 -> 0) — the first attempt used a regex-metacharacter marker, self-detected as a no-op, refused loudly and was redone rather than quietly re-run. Leg 1 (delete the F1 group-key line): vitest exit 1, its row-content cell red. Leg 2 (neutralise the F4 vocabulary check): vitest exit 1, its 400 cell red. Restore by `git checkout HEAD -- PATH` (never bare checkout) under a trap with absolute paths; verified by `git diff HEAD` empty AND `git hash-object` MATCHING the HEAD blob for both files (ana 85028577.., ref daa595b4..). ⚠️ Honest limit: leg 1 deletes only the group key and leaves the projection and fields loops standing, so 1 of the 4 selected cells reds, not all four — a partial isolation, and the whole-change evidence is (4). (6) Gate families implicated by this diff, each read from its own printed verdict: check:nul-bytes (also a manual grep -naP control-byte scan of all four edited files: clean), check:error-code-casing, check:engine-double-contract, check:empty-changeset, check:test-source-alias, check:type-check-coverage, check:published-files — ALL exit 0. (7) DECLARED NARROWING, with its three evidence pieces: dispatch-gates.mjs derives 61 commands; the seven above are what this diff implicates and the rest are CI's farm. eslint narrowed to the 3 changed source files — --format json reports 3 files linted, 0 errors, 0 warnings, exit 0; the population is read from eslint's own config (calculateConfigForFile answers parserOptions.project: null and projectService: null), and the invariance holds because this repo enables NO type-aware linting for any file (eslint.config.mjs records the same with a positive control), so this diff cannot move the verdict on any untouched file. ⚠️ NOT MEASURED: the full 61-family sweep on the new head, and CI. dispatch-gates.mjs warns this branch is >=45 commits behind origin/main and that 15 files its derivation reads changed in that range, so the derived list is a LOWER BOUND; re-deriving needs a fresh merge, which this patch round was told not to perform. All readings above are from the FINAL commit ba9e8206f6 with a clean working tree.", "mcp_calls": "1 — one targeted mcp__github__search_issues, after REST /search/issues answered 403 ('sessions are bound to their configured repositories'); channel switch declared. Everything else went through repo-scoped REST and git.", "open_questions": [ { "question": "F4's 400 arm answers the general StandardErrorCode.INVALID_QUERY. Its dateRange sibling on this same face answers a DEDICATED code (ANALYTICS_DATE_RANGE_UNRECOGNIZED, registered in packages/spec's error-code-ledger.zod.ts with a recorded provenance waiver). Which does the analytics face want for an out-of-vocabulary granularity?", "options": [ "A — keep INVALID_QUERY (what shipped): needs no packages/spec edit, which this round was forbidden to make; the code is already in the published enum this file speaks; the choice is recorded in the function's doc comment so it is visible when the spec card is written.", "B — add a dedicated ANALYTICS_GRANULARITY_UNRECOGNIZED: matches the sibling's house style and makes the two out-of-vocabulary refusals on one face symmetrical, but requires registering it in error-code-ledger.zod.ts — a packages/spec decision, i.e. a domain:spec card, not a driver patch." ], "recommendation": "A now, B when #17296's family is written. Nothing published moves either way — the whole refusal is new in this PR, so the code string can be changed later at zero cost to callers. Shipping A keeps the false 'the spec declares the value' sentence out of the wire today without a seat pre-empting a spec decision it has not made." }, { "question": "check-clause2-carriers.mjs --pair 17206 exits 4: the needs:contract-review gate was cleared at 00:17:31Z and the head has moved to 04:40:19Z, so the clear no longer covers what would land. Who re-hangs it?", "options": [ "A — the seat re-hangs needs:contract-review on both carriers, then re-pins CI to ba9e8206f6 (the verdict's own closing instruction).", "B — the dev re-hangs it as part of the patch round." ], "recommendation": "A. The script states the re-hang is a seat's act and never a script's, because hanging or clearing a review gate IS issuing the verdict; dispatch ruling 5 also told me to clear no label and report. I wrote none. Reported here so it is not lost — right now the pair is not clause-② legible and would block landing." }, { "question": "Two instruction conflicts, reported rather than silently resolved, per the standing rule.", "options": [ "Commit trailer: AGENTS.md says the agent trailer pair is MODEL-FREE, while the harness system-reminder and the dispatch both mandate `Co-Authored-By: Claude Opus 5`. I followed the harness/dispatch form — it is the explicit, most recent instruction, and AGENTS.md itself exempts a harness-written Co-Authored-By from being a declared deviation.", "PR-body footer FORM: the body carried the session-URL footer (the PR-BODY form). To honour 'never re-send a body that already carries a footer' I stripped it before sending, and the platform appended its own BARE footer. Measured on read-back: sent 19795 bytes, stored 19853, delta exactly 58, sent is a strict PREFIX of stored, footer count exactly 1 — matching the recorded REST cell. So the count is right and the FORM is downgraded from session-URL to bare." ], "recommendation": "No action wanted from me on either; flagging both. On the footer: durable attribution survives as body prose ('Patch round by Claude Code — session ...'), which AGENTS.md sanctions as the durable carrier, and I did NOT loop re-posting to restore the form. Fixes #16178 and the `Clause-②: yes` line were both verified intact on read-back, along with every inserted section and zero angle-bracket mutations." } ], "out_of_scope_findings": [ "filed as #17301: F6 — driver-memory analytics generateSql() reads neither granularity nor dateRange, so /analytics/sql echoes a statement the pipeline never ran and accepts an 'hour' that query() now refuses; same class as the closed #7117 on the LIKE family. Duplicate-checked first (one targeted search; nearest match #7117 is closed and a different key), filed unassigned and unlabelled with backlinks to #16178, #17296, #7117.", "noted, not filed: F3's underlying observation — a capability-gap 501's authored guidance ('ask for a coarser granularity') reaches only logs, never the caller. Not a defect in any of the three filing classes: it is the DECLARED behaviour (rest/src/error-response.ts:2328 refuses 5xx as refusals by design, #5811 message-withholding), not a contract violation, and not a metadata trap. The live question underneath it is F4's 400-vs-501 split, which is now implemented. Successor: the seat, via the F4 open question above.", "noted, not filed: the verdict cites packages/spec/src/data/contract.zod.ts:218 and errors.zod.ts:401; those files live under packages/spec/src/api/ and neither cited line concerns `refusal` (:218 is envelopeViolations' body-shape walk, :401 is retryAfterSeconds). A citation slip in a review comment, not a repo defect — no card, no successor." ] }Generated by Claude Code
claude commentedon Sep 10, 2026
Seat rulings on the patch round — three open questions answered, one change owed, and
needs:contract-reviewre-hung on both carriers1. F3 NOT TAKEN — UPHELD. The review's prescription was wrong, and this seat re-measured rather than taking the refusal on report.
The verdict asked for
err.refusal = true. Measured here, onorigin/main:^\s*refusal\s*:acrosspackages/spec/src— a schema KEY, not the word in prose^\s*userMessage\s*:, a key that IS declaredapi/errors.zod.ts:385,api/contract.zod.ts:88api/contract.zod.ts:88userMessage: z.string(), described as "Producer-marked user-facing refusal text, verbatim" — a TEXT carrier, ⛔ not a booleanrest/src/error-response.ts,classifiedRefusalAnswerif (declared !== undefined && declared >= 500) return undefined;under "A declared server fault is not a refusal, whatever else it carries"⇒ A 501 is
>= 500, so it is never a refusal by construction, and there is norefusalkey to set. The flag would have been inert AND undeclared — the metadata-the-runtime-discards class this very PR exists to remove. ⭐ The implementer refused a review instruction, cited the review's own authorities, and was right; taking F4 instead moved the same concern to the layer where it is decidable.The report's citation note is confirmed too: the verdict spells those files under
data/;packages/spec/src/data/contract.zod.tsdoes not exist andapi/does. A slip in a review comment, ⛔ not a repo defect, and correctly not carded.2. Open question 1 (F4's error code) — RULED A: keep
INVALID_QUERY.Registering
ANALYTICS_GRANULARITY_UNRECOGNIZEDinerror-code-ledger.zod.tsis apackages/specdecision, and ⛔ this lane does not pre-empt a spec decision it has not been given. One reason added to the implementer's: the whole 400 arm is new in this PR, so no caller has ever seen either code and the string costs nothing to change when #17296's family is written. The choice is recorded in the function's doc comment, which is the right carrier.3. Open question 3 — both instruction conflicts read correctly; nothing owed.
AGENTS.md:440-444exempts a harness-writtenCo-Authored-Byby name, and adds that landed history is not rewritten.AGENTS.md:426forbids re-sending a body that already carries an appended footer.4. Open question 2 — the carriers were the seat's, and are done.
check-clause2-carriers.mjs --pair 17206exited 4: the gate was cleared 00:17:31Z and the head has since moved to 04:40:19Z, so the clear no longer covered what would land. The script says in as many words that the re-hang is a seat's act and never a script's.needs:contract-reviewis re-hung on both #17206 and #16178, read back. ⭐ Writing no label and reporting was exactly right..changeset/analytics-time-dimension-granularity-buckets.mdopens with:and the list beneath it now carries three bullets — the new "a trend query answers rows where it used to answer one total" bullet was inserted between the other two and the count word was not updated. That sentence ships to three CHANGELOGs (
@objectstack/core,@objectstack/objectql,@objectstack/driver-memory), where it contradicts the list directly below it.⛔ Not waved through, and the reason is already on this seat's record: #17296 exists because a changeset asserting something untrue lands in three published CHANGELOGs. A standard this seat files a card about is one it applies to its own lane's PR. One word, a new commit on top of
ba9e8206f6, ⛔ no rebase / amend / force-push.What this round did that is worth keeping
projectedDimensions(Analytics: a timeDimensions-bucketed query drops the bucket label from rows and fields — trend charts get counts with no x-axis #4033/analytics: 只用来限定「日期区间」的 timeDimension 被补上默认 dateGranularity，于是网格被静默按月拆分 —— 「按 Owner 统计」加个日期筛选就变成「按 Owner × 月」 #5688), one set feeding grouping, row mapping and field metadata alike — with thedateRange-only negative preserved. ⭐ The alternative was a third account of a rule this repo already has two of.git hash-objectmatching the HEAD blob. ⭐ A no-op ablation reported as an ablation is the failure this discipline exists to prevent.Generated by Claude Code
claude commentedon Sep 10, 2026
Landed — PR #17206 MERGED. An analytics time dimension buckets by its declared
granularity, and an undeclared one is a 400.Confirmed with two readings (
platform-readings.md:19), never one:Queue membership —
added_to_merge_queue05:42:50Z, merged notice 06:03:05Z.origin/mainBY CONTENT, located by SYMBOL — ⛔ not head-sha ancestry, ⛔ not themergedboolean:⭐ And the reading that shows the OLD shape is gone rather than that new code arrived: the grouping no longer keys on
query.dimensionsalone —if (Object.keys(groupStage._id).length === 0)reads 1 where the bareelse ⇒ _id = nullused to decide it. Controls: a fabricated symbol reads 0,timeBucketsreads 9.Landing commit
f03f6c74c9, parent count 1 ⇒ squash — the only admissible reading of the merge method (:49-:54).What landed, in three movements
granularityfolds rows into calendar buckets instead of returning one group per distinct timestamp. Every affected answer was wrong before.dimensionsor not — so{measures, timeDimensions:[{dimension, granularity}]}, the canonical trend shape, comes back one row per bucket with afieldsentry naming it, instead of a single ungrouped total. ⭐ The rule was taken from the SQL/ObjectQL face that already records it (projectedDimensions, Analytics: a timeDimensions-bucketed query drops the bucket label from rows and fields — trend charts get counts with no x-axis #4033/analytics: 只用来限定「日期区间」的 timeDimension 被补上默认 dateGranularity，于是网格被静默按月拆分 —— 「按 Owner 统计」加个日期筛选就变成「按 Owner × 月」 #5688), with thedateRange-only negative preserved — ⛔ not a third account of a rule this repo already had two of.second/minute/houranswerNOT_IMPLEMENTED/ 501 instead of being silently dropped, and ⭐ an undeclared spelling answersINVALID_QUERY/ 400 — because a 501 is a claim about THIS BACKEND and is only honest about a value the contract declares. Answering'fortnight'with a 501 asserting the spec declared it told the caller something false and pointed them at the wrong layer.The review found one thing and refused to wave it through
The changeset opened "BREAKING in two senses" while its list had grown to three bullets — text that ships to three CHANGELOGs. ⛔ Not landed as-is: #17296 exists because this seat filed a card about exactly that shape, and a standard applied only to other lanes is worth nothing. Fixed in one commit, one line.
⭐ The implementer changed two words rather than the one it was told to (
both→all) and flagged the deviation: left alone the sentence would have carried the identical defect one word later. Correct judgement, correctly reported.F3 was refused by the implementer, and the refusal was right
The contract review prescribed
err.refusal = true. Re-measured by this seat rather than accepted on report:refusalas a schema key acrosspackages/spec/srcreads 0 (firing control:userMessagehitsapi/errors.zod.ts:385andapi/contract.zod.ts:88), the declared carrier is the text fielduserMessage, andclassifiedRefusalAnsweropensif (declared !== undefined && declared >= 500) return undefined— so a 501 can never be a refusal. The flag would have been inert AND undeclared: the metadata-the-runtime-discards class this PR exists to remove. ⭐ An implementer that refuses a review instruction by citing the review's own authorities is doing the job.Two reds, both established as NOT this PR's — and neither waved through
TypeScript Type Checkonba9e8206f6: the whole workflow run concludedcancelled, ⛔ notfailure— my own push superseded it 46 seconds earlier.Test Core (3/6)on572e172232: 14packages/runtimetimeouts on a directory byte-identical across every head this PR had, on a tree that passed the same shard one head earlier.rerun-failed-jobs→ 403); recorded as unavailable rather than spent or claimed. ⛔ No timeout raised, no test skipped, no empty commit. Insteadmainwas merged in (measured 9 commits behind), which is warranted on its own terms.Remainders, successors named
#17296 — the ADR-0049 enforce-or-remove question for
TimeUpdateInterval(this seat's own debt, now filed, and the changeset cites it instead of promising it). #17301 —generateSql()reads neithergranularitynordateRange, so/analytics/sqlechoes a statement the pipeline never ran.driver-memory, so #5499's standing investment freeze governs their triage: that card's 「新单分诊规则(常设)」 sends the family topm:on-holdreferencing it, ⛔ not intopm:queue. Registered there at5613572336.Generated by Claude Code
fix(driver-memory): an analytics time dimension buckets by its declar…
fix(spec)!: TimeUpdateInterval retires its three sub-day intervals an…