Issues

Block one: message loss is correlated across runs and concentrated in a fragile subpopulation, not the uniform race the design assumed

#347

· martymcenroe opened

on Sep 9, 2026

tools/scrub-sidebar-fixtures.js requires jsdom but package.json does not declare it — it resolves only by npm hoisting a transitive dep

#339

· martymcenroe opened

on Sep 8, 2026

MUST FIX: assistant-generated images are not captured at all — findImages finds zero, capture reports clean, and no provider's image fidelity has ever been verified

bug

launch-blocker

#331

· martymcenroe opened

on Sep 8, 2026

MUST FIX: a plain-text user message is dropped deterministically and compare-sweeps can never see it

bug

launch-blocker

#330

· martymcenroe opened

on Sep 8, 2026

The ordering note fires on ~15% of messages every capture and advises a re-capture that cannot help

#329

· martymcenroe opened

on Sep 8, 2026

The popup's result card survives the click that starts a new run, a conversation change, and a page reload

#328

· martymcenroe opened

on Sep 8, 2026

MUST FIX: extracting twice without reloading captures 12 of 179 messages and reports contentComplete: true

bug

launch-blocker

#327

· martymcenroe opened

on Sep 8, 2026

A conversation stuck in a phantom generating state is permanently invisible to the sweep: detect quiescence, report the list, offer an opt-in stop

#326

· martymcenroe opened

on Sep 8, 2026

The 'Something went wrong, please retry' page is undetected: 3 twelve-minute burns in run 1, and a partial render would be recorded as a clean capture

#320

· martymcenroe opened

on Sep 8, 2026

Conversation dates are rendered on the page and captured nowhere; the id prefix that looks like a creation epoch is wrong by a year

#319

· martymcenroe opened

on Sep 8, 2026

MUST FIX: reasoning content is never captured on any provider — no code path exists, and 5,644 messages reported clean without it

bug

launch-blocker

#316

· martymcenroe opened

on Sep 7, 2026

The 45s patience floor costs ~80 min per sweep on short conversations, but 'complete on arrival' is inferred, not measured

#315

· martymcenroe opened

on Sep 7, 2026

Navigation Menu

All issues

Block one: message loss is correlated across runs and concentrated in a fragile subpopulation, not the uniform race the design assumed

tools/scrub-sidebar-fixtures.js requires jsdom but package.json does not declare it — it resolves only by npm hoisting a transitive dep

MUST FIX: assistant-generated images are not captured at all — findImages finds zero, capture reports clean, and no provider's image fidelity has ever been verified

MUST FIX: a plain-text user message is dropped deterministically and compare-sweeps can never see it

The ordering note fires on ~15% of messages every capture and advises a re-capture that cannot help

The popup's result card survives the click that starts a new run, a conversation change, and a page reload

MUST FIX: extracting twice without reloading captures 12 of 179 messages and reports contentComplete: true

A conversation stuck in a phantom generating state is permanently invisible to the sweep: detect quiescence, report the list, offer an opt-in stop

The 'Something went wrong, please retry' page is undetected: 3 twelve-minute burns in run 1, and a partial render would be recorded as a clean capture

Conversation dates are rendered on the page and captured nowhere; the id prefix that looks like a creation epoch is wrong by a year

MUST FIX: reasoning content is never captured on any provider — no code path exists, and 5,644 messages reported clean without it

The 45s patience floor costs ~80 min per sweep on short conversations, but 'complete on arrival' is inferred, not measured

Search results