The docs deploy pipeline is failing

[github-actions]

Blocked-by: #261

Triaged 2026-09-04T14:10Z: this is the #269 mechanism working, reporting the known deliberate state on #261. The live site is unaffected — a rejected upload creates no version, so 69c79ee3-… (2026-08-25) keeps serving. ⛔ Do not close it to quiet it; the next push would just open another. See the triage comment.

Automated report from ci.yml -> deploy-docs.yml.

run	https://github.com/objectstack-ai/objectos/actions/runs/33881450166
commit	d6f5ddab72f1881def890ec5716ddce676987c70
deploy	failure
smoke	success
rollback	skipped
version serving before	69c79ee3-ea73-4a2f-839a-9fad49e20d5a
version published	none

A deploy failure with no published version usually means Cloudflare
refused the upload — while #261 is open that is the expected steady
state and the live site is unaffected. A smoke failure means the
live site did not render; the run log names which rule fired.

[os-project-manager]

This card is the mechanism from #269 working, not a new problem. Triaged by the repo:objectos seat (objectstack#9831) 2026-09-04T14:10Z, minutes after it filed itself.

It reports the known, deliberate state tracked on #261: main builds a docs Worker over Cloudflare's 64 MiB limit, so the upload is refused at version creation. A rejected upload creates no version, so the live site keeps serving 69c79ee3-… (2026-08-25) and is unaffected.

pm:blocked, Blocked-by: #261 — this cannot be resolved on its own terms and is not separate work.

Why it exists at all, and why that is the point

Before today, this exact failure produced 36 consecutive red runs across two PM tenures and not one card. A red run in a list nobody reads is not a signal. This card is.

It also demonstrates the dedupe semantics in both directions, which is worth recording because a beacon that spams is a beacon people mute:

• The docs deploy pipeline is failing #271 was filed by the same mechanism during verification. While it was open, a second failing run commented on it rather than opening a new card.
• The docs deploy pipeline is failing #271 was then closed. This run — the first real failure after the layer landed — correctly filed a new card rather than reopening the closed one.

That is the right reading of "I fixed it, tell me if it comes back": an open card accumulates, a closed one means the next occurrence deserves a fresh report.

What it will do from here

Every push to main that reaches the deploy will add a comment here while #261 is open. Do not close it to quiet it — closing only means the next push opens another. It goes away when the bundle fits, which is #261's job.

⚠️ If this card ever reports smoke: failure rather than deploy: failure, that is a different and much more serious event — it means the live site stopped rendering, and the rollback job should already have fired. deploy: failure with version published: none is the boring expected case.

Related: #261 (the underlying defect), #269 / PR #270 (the layer that files this), #262 (the bundle budget that would have caught it before the cliff), #266 (the human gate, still awaiting the maintainer).

---

Generated by Claude Code

[os-project-manager]

Closing — the condition this beacon reported is gone

Reading taken 2026-09-04T15:50Z. The triage note above said ⛔ do not close this to quiet it, because while #261 was open the next push would just open another card. That premise no longer holds: the deploy pipeline publishes.

Run 33891143864 on 0e26657f:

deploy	success
smoke	success
rollback	skipped
version serving before	69c79ee3-ea73-4a2f-839a-9fad49e20d5a
version published	2170b929-5879-4b3f-b7a2-9eda750158dd

Upstream #261 is closed with the full readings. Closing this as completed rather than not-planned: the card did its job — it is the artifact of the #269 mechanism correctly reporting a real, deliberate failure state for the six hours that state lasted, and it is the reason that state was never silent.

The mechanism stays armed. If a future deploy or smoke check fails, deploy-docs.yml opens or updates a fresh card under the same deploy-failure label; this closed one is a record, not a switch.

---

Generated by Claude Code