docs(quickstart): re-transcribe both boot samples against CLI 17.3.0

[os-project-manager]

Part of #141

Deliberately Part of rather than a closing keyword. This card is a watcher that is re-armed every round, and last round's PR used a closing keyword on it — GitHub closed the card on merge and the PM had to reopen it by hand. The keyword is kept away from the card number here on purpose: GitHub's parser matches it even inside backticks and even under a negation, so the only safe way to not close a card is to not put the word near its number.

What this round measured

Both paths were booted on @objectstack/cli 17.3.0 as independent pty captures — Path B's banner was never assembled from Path A's — and each raw capture was diffed against its published block mechanically, line by line. Every difference was then classified against a 17.2.0 control whose scaffold dependencies were pinned to exact versions, so the control could not resolve forward into a second copy of the experiment.

That separation is load-bearing again this round: of the four differences found, two are genuine 17.3.0 changes, one is a declared machine-variable, and one is a pre-existing under-declaration that has been there since at least 17.2.0. Without the control all four would have read as 17.3.0 drift, and the fourth would have been written into the docs as a change this release did not make.

#	Item	Outcome at 17.3.0
1	Re-transcribe both blocks	Both were stale, in different ways. Path A lost its Console: header row. Path B gained a fifth compile counter and its artifact shrank to 1.8 KB.
2	Port behaviour	No drift. Unchanged and re-measured with a live probe.
3	Both pins	17.2.0 to 17.3.0.
4	Declaration sentences	Two corrections — one new (the MCP trailer moved), one pre-existing (the SQL driver's schema-sync notes were never named).

Path A — the Console: header row is gone

The banner's console row is absent from two independent 17.3.0 boots and present in the pinned 17.2.0 control. That is what makes it a 17.3.0 removal rather than a capture artefact. The console URL is still reachable in the block through the Console: line in the ready list, and through the "What's running" table below it, so nothing a reader needs disappeared — only the duplicate.

Path B — a fifth compile counter, published whole

17.3.0 adds Runtime: 0 plugins to the compile summary. It is published into the block rather than declared-and-omitted, following the group-completeness rule this page already sets: a summary group is published whole (Path A publishes all of Mode/Driver/Tenancy/Plugins), and only long continuations get declared and omitted. Publishing four members of a five-line group would show the reader a hole in the middle of something they see intact on their own screen.

The scaffold's artifact also moved from 1.9 KB to 1.8 KB — deterministic from scaffold content, present as 1.9 KB in the control, so this is real.

The Build complete timing is re-transcribed, not frozen

#260 converted this number from an asserted literal into a declared machine-variable, and that declaration is doing its job — the timing differing is not drift. But the block is a faithful transcript, and once the pin says 17.3.0 the transcript has to be of a 17.3.0 boot, so the number is re-transcribed from this capture rather than carried over. It is not placeholdered or averaged.

Item 4 — the two declaration corrections

Both found by diffing the raw capture against the published block, not by reasoning about what was probably left out.

• The SQL driver's schema-sync notes were never named by either sentence. They appear on both paths, and the control shows them at 17.2.0 too — in far greater volume (338 lines on Path A, 356 on Path B, against 5 each now). So this is a pre-existing under-declaration, not a 17.3.0 change, and it is described as such. Both sentences now name it.
• Path B's MCP trailer moved. At 17.2.0 it printed after Press Ctrl+C to stop; at 17.3.0 it prints between the dev-admin lines and the Config: banner. The sentence said "below the banner", which is now false, and reads "above the banner".

Port behaviour, controlled both ways

Verified as a live probe rather than a bare grep that would pass either way:

• Uncontended: binds 3000, and the rebind message appears zero times.
• Contended (a real os start holding 3000, confirmed serving HTTP 200): binds 3001, and the published literal server bound to port 3001 (requested 3000) appears exactly once, byte-exact.

Untouched on purpose

• The seeded dev credential and its dev-only caveat are byte-identical at 17.2.0, at 17.3.0, and on the page. Transcribed faithfully, not redacted or altered.
• The expired Path B Warning: admonition stays discharged: zero hits, with a live control confirming one other blockquote is still present in the file, so the zero is a probe that works.
• reference/cli.mdx and all seven locale siblings are untouched. The siblings are stale by design and are the translation pass's input.

Verification

Gate union run on the final commit 4f41c03, with git status --porcelain empty so the tree equals the commit. Exit codes captured before any pipe; verdicts quoted from each gate's own printed line. Turbo run with --force throughout — every reading is 0 cached, not a shared-cache hit.

Gate	Exit	Verdict line
turbo run build --force	0	Tasks: 1 successful, 1 total · Cached: 0 cached, 1 total
turbo run type-check test --force	0	Tasks: 2 successful, 2 total · Cached: 0 cached, 2 total
check-locale-surface.mjs	0	every advertised URL has a source file and every source file is advertised
check-node-floor.mjs	0	Every declared floor clears what the dependency tree requires
check-node-floor.mjs --self-test	0	17 rule cases, every rule demonstrated able to fail
check-half-states.mjs --self-test	0	1551 cases pass
gen-zh-hant.mjs --check	0	73 generated files match the zh-Hans sources byte for byte
check-translations.mjs	0	translations gate passed
check-translation-output.mjs --self-test	0	29 rule cases, every rule demonstrated able to fail
check-translation-ownership.mjs	0	0 translation artifacts and 1 other file
check-translation-output.mjs --files	0	translation output gate passed (145 pre-existing findings reported)

check-locale-surface.mjs first exited 1 with "no built artifact ... run pnpm turbo run build first". That is a prerequisite not met, not a red gate — the gate reads build output on purpose. It is re-run above after the build and passes.

Measurement evidence, which is the actual deliverable here:

• Path A captured twice from separate pristine homes; the two are byte-identical after normalizing the home path and timestamps.
• Every published line verified present in the raw 17.3.0 capture after editing: count=0 published-but-absent on Path A against both boots, and on Path B.
• Order preserved: a separate check confirms the published lines appear in the capture in the same sequence, since presence alone does not prove ordering.
• Control genuinely at 17.2.0: spec, runtime, objectql and cli all resolve to 17.2.0 in the control tree and 17.3.0 in the experiment tree.
• Every server killed by explicit PID resolved through /proc; ports 3000 and 3001 confirmed released afterwards by curl. Note ss returns nothing at all in this container, so port probes are curl-based.
• Control-character scan on the edited file: no hits.

Reviewer's choice

Publishing Runtime: 0 plugins into the block is the group-completeness reading and is what this PR does. Declaring it instead is a one-line switch if you prefer the tighter block, but it would leave the counter group with a hole.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01GkauAsZBEemRbco2rEX9Lx

---

Generated by Claude Code