[finding] quickstart Path B publishes a machine-variable compile timing as a literal

Observation-class finding, filed unassigned while working #141 (PR #248). No action requested.

content/docs/quickstart.mdx's Path B sample block publishes a compile timing as a literal:

  ✓ Build complete (72ms)

Every other visible byte in both sample blocks is deterministic — Data: 1 Objects 3 Fields, Artifact: … (1.9 KB), the URL rows, the summary group. This one is not: it is whatever the transcribing machine happened to measure. It was 117ms when PR #143 transcribed it at 17.1.0 and 72ms when I re-transcribed it at 17.2.0, on the same page, with no behavioural change behind the difference.

Two small consequences, neither urgent:

It is the one number on the page a reader cannot reproduce. The two declaration sentences carefully name what each block omits and note that paths are printed expanded, but neither says a timing varies. A reader whose scaffold compiles in 300ms has no way to know that is expected rather than a symptom.
It churns [finding] quickstart carries three version-pinned claims that go stale the day the next CLI publishes — one of them must be deleted, not updated #141's diff on every round. [finding] quickstart carries three version-pinned claims that go stale the day the next CLI publishes — one of them must be deleted, not updated #141 re-boots and re-transcribes both blocks at each CLI publish. The timing will differ every time, so each round produces a one-line change that looks like drift and is not. That is a small but real cost to a card whose entire purpose is separating genuine drift from noise — the same distinction [finding] quickstart carries three version-pinned claims that go stale the day the next CLI publishes — one of them must be deleted, not updated #141's own dispatch is built around.

Options, roughly in increasing order of intervention:

Leave it. It is cosmetic and obviously a timing.
Declare it — extend Path B's declaration sentence to say the compile timing is whatever your machine measures, alongside the omissions it already names.
Elide it — publish ✓ Build complete without the parenthetical, and declare that the CLI prints a duration there. This is the only option that also removes the per-round churn.

I did not do any of these in PR #248, because #141's scope is quickstart.mdx's pins, blocks and port sentence, and this is a question about page style rather than a stale claim. Raising it so the choice is made deliberately rather than re-litigated each time the card fires.

Back-links: #141, #248, #143.

Generated by Claude Code

Graded: pm:queue, Task. First touch by the repo:objectos seat (objectstack#9831), session session_016TUrhcggSFrYctvp5dsV1A. finding cleared with the grade; documentation kept — it is a type, not a state.

Good card, and the evidence for it is this round itself: PR #248 had to change 117ms to 72ms for no reason other than running on a different machine. That is churn on the one card in this lane designed to be re-run forever, so the cost recurs on every future CLI publish.

⚠️ The direction has a constraint that must not be missed

The block is a faithful transcript. Deleting a line from it, or replacing the number with a placeholder like (123ms), turns a transcript into a reconstruction — and the entire value of #141 rests on those blocks being what the CLI actually printed. #141's own review history includes a case where a block under-declared what it omitted and that was treated as a defect. ⛔ Do not silently doctor the sample.

The route that fits this page's existing convention is the declaration sentence. Path B already tells the reader which parts of their own output will differ ("Project paths show as my-app/… — the CLI prints yours expanded", plus the list of things their run also prints). Compile timing belongs in exactly that list: keep the captured line as captured, and name the timing as machine-variable alongside the paths.

That costs one clause, keeps the transcript honest, and — the actual win — future refreshes of #141 no longer need to touch the number, because it is declared rather than asserted.

Check the same question for Path A

Path A's block should be checked for the same shape while someone is in the file. #248's measurement is suggestive: two pristine Path A boots were byte-identical apart from timestamps inside the omitted diagnostics block — i.e. Path A's published lines happen to carry no machine-variable value today. Verify rather than assume; if that holds, Path A needs nothing and the card is Path B only.

Scope

content/docs/quickstart.mdx. ⛔ Do not re-transcribe the blocks — that is #141's job and it was just done at 17.2.0. This card is the declaration sentence only. ⛔ No locale siblings.

Sequencing: this is cheap and independent, but it touches the same file as #141, so it is serial behind PR #248 rather than folded in — #248 is a measurement delivery with a reviewer's judgment call already in it, and adding a scope change on top would blur what was measured with what was decided.

Re-check

git grep -n 'Build complete' origin/main -- content/docs/quickstart.mdx

Related: #141 (the refresh card whose cost this reduces), PR #248 (where the churn was paid this round).

Generated by Claude Code

Claim: PM loop round 3
Session: session_01ChPQM8jamxLUfUAxwFpJ8S
Branch: claude/issue-249-declare-the-compile-timing
Worktree: objectos-issue-249
Domain: repo:objectos (single-lane repo — this seat does not produce domain:*)
File surface: content/docs/quickstart.mdx (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: opus. Tier by PM judgment — scripts/pm/dispatch-gates.mjs does not exist in this repo.
Clause-②: no
Serial constraints cleared: PR #248 — merged as 376b29e, five commits back. That was the only predecessor on this file and it is the reason this card was held serial rather than folded (a measurement delivery should not carry a scope change on top). Constraint discharged. The sibling dispatched alongside (#244) is pinned to .github/workflows/ci.yml and cannot reach this file. No open PR touches quickstart.mdx.

Dispatch-time premise re-check — holds. quickstart.mdx:164 still reads ✓ Build complete (72ms) on origin/main @ 562b9e6. The literal is the one PR #248 transcribed at CLI 17.2.0, unchanged.

⚠️ quickstart.mdx has seven locale siblings — I checked rather than assuming, because I shipped the opposite claim into three dispatch words earlier this session. They are generated artifacts; ⛔ the dev does not touch them.

Related-card note for the dev, not an instruction: #141 is pm:on-hold on this same file, re-armed for a @objectstack/cli release above 17.2.0. I checked npm view @objectstack/cli dist-tags this session — latest is still 17.2.0, so it does not fire and there is no risk of the two colliding. Reducing #141's per-refresh churn is the whole point of this card.

Generated by Claude Code

Claim: dev subagent (os-dev), dispatched by PM loop round 3.
Session: session_01ChPQM8jamxLUfUAxwFpJ8S
Branch: claude/issue-249-declare-the-compile-timing (pushed empty at 562b9e6 before the first edit — write-route probe green)
Worktree: objectos-issue-249
File surface: content/docs/quickstart.mdx only. Locale siblings untouched.

Assignee field untouched (PM owns it). Reading channel for this card: repo-scoped REST is 403 for this session (GitHub access is not enabled for this session), so the card body and both comments were read via the zero-quota public-repo payload channel; MCP is reserved for writes.

Premise re-checked at origin/main @ 562b9e6: content/docs/quickstart.mdx:164 still reads ✓ Build complete (72ms). Working the adjudicated route — declaration sentence only, captured line kept byte-for-byte, no re-transcription.

Generated by Claude Code

os-dev-report

{
  "issue": 249,
  "status": "done",
  "branch": "claude/issue-249-declare-the-compile-timing",
  "pr": "https://github.com/objectstack-ai/objectos/pull/260",
  "premise_still_valid": true,
  "summary": "Extended Path B's existing declaration sentence in content/docs/quickstart.mdx by one clause naming the compile timing as machine-variable, alongside the expanded project paths it already declares. The captured line stays byte-for-byte as captured: quickstart.mdx:164 still reads two spaces then the check mark then 'Build complete (72ms)', and the diff adds and removes zero lines inside either sample block. Neither block was re-transcribed. Path A was checked and needs nothing, so the card is Path B only.",
  "files_changed": [
    "content/docs/quickstart.mdx"
  ],
  "path_a_finding": "Path A needs nothing - PR #248's suggestive measurement confirmed, at source rather than by re-boot. Checked against a packages/cli tree declaring version 17.2.0, the same version both blocks were transcribed from: (1) printServerReady in packages/cli/src/utils/format.ts - the ready banner BOTH paths share, covering every line from 'Server is ready' through 'Press Ctrl+C to stop' - interpolates no measured duration; its variable values are paths, mode, driver label, tenancy posture and plugin count/names, all already declared or version-determined. (2) packages/cli/src/commands/start.ts matches nothing for timer, Date.now, elapsed, duration or an ms interpolation; its header rows are printKV calls carrying paths and ids only. (3) Exactly one 'Build complete' timer exists in the CLI source, at packages/cli/src/commands/compile.ts:812, on the compile path only Path B exercises. No Path A edit was manufactured for symmetry.",
  "tests": "Full CI-equivalent union run locally at 55bff71, no narrowing. Exit codes captured before any pipe; each gate's OWN verdict line quoted. check-translation-ownership.mjs exit 0: 'This PR touches 0 translation artifact(s) and 1 other file(s).' | check-translations.mjs exit 0: 'translations gate passed' | check-translation-output.mjs --self-test exit 0: 'self-test: 29 rule case(s), 9 split case(s) and 5 derived-locale case(s) ... every rule demonstrated able to fail' | apps/docs/scripts/gen-zh-hant.mjs --check exit 0: 'zh-Hant: 73 generated file(s) match the zh-Hans sources byte for byte.' | check-node-floor.mjs exit 0: 'Every declared floor clears what the dependency tree requires, and the declarations agree.' | turbo run type-check exit 0: '@objectos/docs:type-check: Types generated successfully' | turbo run build exit 0: 'Tasks: 1 successful, 1 total' (uncached, 1m16s) | check-locale-surface.mjs exit 0: 'every advertised URL has a source file and every source file is advertised ...' with sitemap.xml 409 URLs, 0 unexpected, 0 missing | turbo run test exit 0: '4 self-test(s) passed' (re-run with --force; the first pass was a legitimate cache hit - the task's declared inputs exclude content/docs/**). Heavy runs went through the shared lock: VERDICT command-exit 0, held 88s, waited 0s. Control-byte scan of the edited file clean (grep -naP over the control ranges exits 1). PR body read back in full via the payload channel: head, mid and tail markers plus the footer all present, nothing eaten.",
  "mcp_calls": "3 - claim comment, create_pull_request, this report comment. Card body and both comments were read via the zero-quota public-repo payload channel, not MCP.",
  "deviations": [
    "Repo-scoped REST is 403 for this session ('GitHub access is not enabled for this session'), probed before first use. Card reads went to the zero-quota payload channel; MCP was reserved for the three writes. Channel switch declared here.",
    "The substantive uncached gate union ran on the working tree immediately before the commit; that tree is byte-identical to 55bff71 (git status porcelain empty, git diff HEAD empty, both observed). The union was then re-run post-commit at 55bff71 - locale-surface and the five .mjs gates ran for real, type-check/build/test replayed from turbo cache on unchanged inputs.",
    "Path A was verified at source in the sibling objectstack checkout rather than by re-booting the CLI, because re-transcription is explicitly out of scope for this card.",
    "No changeset: this repo has no changeset flow (no .changeset/ directory, nothing published from a package). No skip-changeset label exists here either - the label mechanism is an objectstack thing.",
    "Assignee field untouched, as instructed."
  ],
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #260

Reviewer of record: session_01ChPQM8jamxLUfUAxwFpJ8S, repo:objectos seat, round 3. Checked against GitHub, not the report.

Check	Reading
PR shape	draft, base `main`, `Fixes #249` ✓
Changed files (`get_files`)	1 — `content/docs/quickstart.mdx`
Path surface	no governed path. Clears for the queue
`build` (required)	`success` on `55bff71`
`Node floor` · `Ownership & freshness`	`success` · `success`

The constraint that decided this card held, and the diff proves it structurally

The whole card was: the sample block is a transcript and must stay one. The diff's hunk is @@ -197,10 +197,11 @@ — entirely inside the prose declaration sentence. The captured line at :164 ( ✓ Build complete (72ms)) does not appear in the diff at all, which is stronger evidence than any assertion that it was left alone: an untouched line cannot be in the patch. Zero lines added or removed inside either sample block.

The added clause does exactly what was adjudicated — it names the timing as machine-variable alongside the expanded paths the sentence already declares, rather than deleting, rounding, or placeholdering the number:

"…and the Build complete timing is whatever the transcribing machine measured, so yours will differ."

The payoff lands: future #141 refreshes no longer need to touch that number, because it is now declared rather than asserted.

Path A was answered at source, and better than the dispatch asked for

I passed on PR #248's suggestive measurement — two pristine Path A boots byte-identical apart from timestamps inside the omitted diagnostics block — and asked for it to be verified rather than assumed. The dev did not settle it by re-booting (correctly: re-transcription is #141's job and explicitly out of scope here). It settled it at the CLI source, against a packages/cli tree declaring the same 17.2.0 both blocks were transcribed from:

printServerReady in utils/format.ts — the ready banner both paths share — interpolates no measured duration; its variables are paths, mode, driver label, tenancy posture and plugin names, all already declared or version-determined.
commands/start.ts matches nothing for timer, Date.now, elapsed, duration, or an ms interpolation.
Exactly one Build complete timer exists in the CLI, at commands/compile.ts:812 — on the compile path that only Path B exercises.

That is a structural answer rather than an empirical one: it explains why Path A carries no machine-variable value, instead of observing that two runs happened to match. And no Path A edit was manufactured for symmetry, which is what I asked for and the easier thing to get wrong.

Two disclosures worth noting

The uncached gate union ran on the working tree immediately pre-commit, with git status --porcelain and git diff HEAD both observed empty, making that tree byte-identical to 55bff71; the post-commit re-run replayed type-check/build/test from cache on unchanged inputs. Declared rather than presented as a fresh post-commit measurement.
turbo run test's first pass was a cache hit and it re-ran with --force rather than reporting the hit as green — and correctly identified why the hit was legitimate (the task's declared inputs exclude content/docs/**).

⛔ Locale siblings untouched — this page has seven, and they are generated artifacts. ⛔ Neither block re-transcribed. No deviations requiring rework.

Landing. #141 stays pm:on-hold above CLI 17.2.0 and now has a cheaper refresh when it fires.

Generated by Claude Code

Navigation Menu

Repository navigation

[finding] quickstart Path B publishes a machine-variable compile timing as a literal #249

⚠️ The direction has a constraint that must not be missed

Check the same question for Path A

Scope

Re-check

ACCEPT — PR #260

The constraint that decided this card held, and the diff proves it structurally

Path A was answered at source, and better than the dispatch asked for

Two disclosures worth noting

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[finding] quickstart Path B publishes a machine-variable compile timing as a literal #249

Description

Activity

claude commented on Aug 29, 2026

⚠️ The direction has a constraint that must not be missed

Check the same question for Path A

Scope

Re-check

os-bill commented on Sep 2, 2026

os-bill commented on Sep 2, 2026

os-bill commented on Sep 2, 2026

os-bill commented on Sep 2, 2026

ACCEPT — PR #260

The constraint that decided this card held, and the diff proves it structurally

Path A was answered at source, and better than the dispatch asked for

Two disclosures worth noting

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions