Right-size browser fixture histories

[comp615]

This broad audit makes each fixture-backed browser file declare the history it actually exercises instead of paying the default 640 Alpha / 80 Beta signing cost. Small UI and lifecycle journeys retain independent per-test relay/context/storage state with 0–20 rows; pagination, virtualization, scroll/resize restoration, the 500-event unread boundary, and cases that explicitly address message 639 retain deep histories. No cases, permutations, assertions, test bodies, identities, or signing/isolation semantics change.

The initial undersized control failed at the existing geometry (upper requires >400px), 500-unread, and literal message-639 assertions in both engines. Restoring those scenario depths made the full affected set pass in Chromium and WebKit. initial-position.spec.mjs remains untouched for #78; the four RTL-owned files only change setup declarations. CI/discovery configuration is untouched.

Two hosted candidate runs passed all 394 functional browser cases and every full-CI lane. The first candidate attempt completed the full gate in 468s (7m48s), versus pinned-main attempt 2 / attempt 3 at 587s / 716s.

The concurrent fresh comparison reran the same pinned snapshots: baseline attempt 4 and candidate attempt 2.

Metric	Baseline	Candidate
First job start → required-gate completion	688s (11m28s)	550s (9m10s)
Slowest functional browser step	585.858s	464.161s
Summed signed-history generation, all fixture-backed WebKit cases	500.331s	204.881s
Generated history events, all fixture-backed WebKit cases	141,438	59,880
Functional test/project identities	394 passed	Same 394 passed

The repeat improved the full gate by 138s (20.1%) and the slowest browser step by 121.697s (20.8%). Both candidates beat the baseline samples, but two candidates do not establish a stable percentage guarantee. Summed signing cost is not wall-clock saving. The event/signing totals above include every fixture-backed case, correcting the earlier subset totals in this description.

No runner capacity, setup phase, dependency, browser engine, budget or retry setting changed. Build/plugin fixture setup remains unchanged. The only changed setup work is per-case history generation.

🤖 Drafted and implemented with AI assistance.

[wesbillman]

Carl, an automated reviewer, commenting via Wes’s GitHub account.

Review clear

No actionable blockers found for head 3e1fc7d16d6d4dd4035ed0717d9a5706fc471ad7 against base a9194b24a3cc208abf6f9a9a05eed77583e6c0fa.

• The 28-file diff only declares existing historyCounts options. Test bodies, assertions, case permutations, per-test identities/state, fixture signing and CI configuration remain unchanged. Paging, large-history/anchor, 500-unread and message-639 consumers retain their required depth; smaller UI/navigation fixtures retain independent large-sidebar data and real lifecycle coverage.
• Existing hosted CI is green. The candidate's tested merge 3fbf98f0d70e59e398cebd36f8e371068136808b has the pinned base/head as parents and exactly the head's tree. Baseline run 35049707495 attempt 4 and candidate 35117382002 attempt 2 both passed 394 functional browser cases; first-job-start to required-gate completion was 688s versus 550s. The same functional identities are present in both attempts; rounded per-test log durations sum to approximately 3,254s versus 2,518s. These samples do not establish a stable percentage improvement.
• Review was source/metadata-only: no checkout, local execution or CI rerun. The three existing local-only WebKit measurement cases, native/live-account acceptance and the author-reported local fail-then-pass control were not independently exercised.

Non-blocking evidence correction: the baseline 141,438 events / 500.331s signing totals reproduce from the 144 fixture-backed functional WebKit cases alone; label that scope explicitly rather than “all fixture-backed WebKit cases.” I did not independently recover attempt 2’s precise candidate signing total. The currently listed candidate artifacts come from later attempt 3 and cannot substitute for attempt 2’s timing evidence.