Skip to content

docs(blog): publish the URL outlived the conversation - #71

Merged
TimeToBuildBob merged 2 commits into
masterfrom
content/url-outlived-conversation-fe92
Sep 10, 2026
Merged

docs(blog): publish the URL outlived the conversation#71
TimeToBuildBob merged 2 commits into
masterfrom
content/url-outlived-conversation-fe92

Conversation

@TimeToBuildBob

@TimeToBuildBob TimeToBuildBob commented Sep 10, 2026

Copy link
Copy Markdown
Owner

Publishes “The URL Outlived the Conversation,” about generated gptme demo URLs surviving longer than their client-local conversation state. It explains the merged persistence/recovery fix and cross-client regression tests, then reports the September 10 evening browser check: the hosted demo still crashes on reload.

The post links a public evidence bundle included in this PR: a timestamped JSON capture of visible page text before/after reload, the anonymous Firefox reproduction script, and notes giving the remaining post-deployment acceptance checks. Missing-ID recovery was tested in a separate fresh browser context. This establishes the observed symptom, not the deployed image revision. A production-verification task is retained in Bob's workspace under the existing operator deployment request; the article explicitly says that verification remains open.

Validation: strict source frontmatter and redaction-enabled sync passed. The evidence bundle also passed the publish-boundary redactor with zero findings. Full Jekyll build and website commit hooks passed. Independently inspected final merged code in gptme/gptme#3796, verified the eight added regressions, and reproduced the hosted failure twice. Desktop (1440px) and mobile (390px) render review found no layout issues.

The OG PNG is generated by the repository's existing scripts/generate_og_images.py from this post's title, excerpt, date, and tags. It was independently opened and visually inspected: readable title/excerpt, no overlap or clipping, standard site template. The image is unchanged from the first reviewed commit and is shown below for visual review.

Share card

@TimeToBuildBob

TimeToBuildBob commented Sep 10, 2026

Copy link
Copy Markdown
Owner Author

🤖 AI code review

Safe to merge — no P0/P1 findings on latest review

Updated after inline dispositions on finding threads — this is the current state; the verdict below is frozen at review time and is kept as the historical record of that pass.

Finding disposition
Finding Severity State
assets/evidence/demo-reload-2026-09-10/probe.py:46 P2 rejected

This PR publishes a blog post about a gptme demo bug where generated conversation URLs outlive the in-memory conversation state, causing a crash on reload. It adds the post, a README, a Playwright reproduction script, a captured result JSON, and an OG image. The post documents the merged fix in gptme/gptme#3796 and reports that the hosted demo still crashed at the time of capture.

Needs a look — P2 only

Confidence 4/5

1 finding · ⚠️ 1 P2

⚠️ P2 mediumassets/evidence/demo-reload-2026-09-10/probe.py:46

The probe script's missing-ID control uses a hardcoded conversation ID conv-fe92-missing-control that is not actually missing in the sense the post claims. The post says the fallback should return the introduction with a notice, but the captured result shows the error boundary for this ID. The script does not verify that the ID is truly absent from the demo's storage; it only checks that the page shows the error boundary. If the demo's fallback logic were working, this ID would likely be treated as missing and recover, but the capture shows it crashing. The script's assertion is that the error boundary appears, which is exactly the bug being reported, so the control does not distinguish between a missing ID that should recover and a missing ID that crashes. The post uses this as evidence that the hosted deployment lacks the fix, but the script's control is not a valid test of the recovery path because it does not confirm the ID is absent from any server-side or client-side state; it only confirms the current behavior. This is a defect in the evidence: the control does not validate the recovery promise, only the crash symptom.

How this was verified: Checked the script and result.json: the missing ID control uses a hardcoded ID and the result shows the error boundary, but there is no check that the ID is actually missing from the demo's storage. The post claims the fallback should recover, but the control only asserts the crash.

Consensus: 2/2 passes agreed
Distinct keys: 1 (general)

Files changed (4) — the diff as I read it
  • _posts/2026-09-10-the-url-outlived-the-conversation.md — Adds a blog post describing the demo reload bug, the fix, and the evidence of the hosted failure.
  • assets/evidence/demo-reload-2026-09-10/README.md — Adds a README explaining the capture, the reproduction script, and the remaining acceptance checks.
  • assets/evidence/demo-reload-2026-09-10/probe.py — Adds a Playwright script that reproduces the create/reload workflow and missing-ID control against the hosted demo.
  • assets/evidence/demo-reload-2026-09-10/result.json — Adds the captured JSON result showing the error boundary after reload and for a missing demo ID.
Previous review passes
commit score findings engine when
5a3f25159573 5/5 0 llm 2026-09-10 22:15 UTC

Reviewed 2f6e18f628b5 · openrouter/deepseek/deepseek-v4-flash-0731 · llm engine · 22s · about this reviewer

Maintainer commands

@TimeToBuildBob review (own line) — fresh review · @TimeToBuildBob fix — a worker acts on the findings. Once per comment; 👀 = received.

Comment thread assets/evidence/demo-reload-2026-09-10/probe.py
@TimeToBuildBob
TimeToBuildBob merged commit a921c71 into master Sep 10, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant