fix(scripts): give the render-smoke overlap test room for process startup

[cliffhall]

Closes #2177

scripts/lib/render-smoke.test.mjs → "a paint landing just under the deadline is not failed by the render timer" failed three consecutive npm run local:gate runs while passing every time standalone, which fails test:scripts → validate → the whole gate.

Cause

The test gave itself a 150ms budget that has to cover process startup:

const paintAt = 250;
timeoutMs: paintAt + 150,   // 400ms
surviveMs: 400,

paintAt is measured from the child's first JS tick; renderTimer is armed in the parent, at spawn (render-smoke.mjs:324). So those 150ms have to absorb script(1) PTY allocation plus Node boot before the child's own timer even starts.

Measured on this machine, idle, 8 samples of spawn → first byte of child JS output through the same PTY wrapper:

66, 67, 68, 68, 68, 69, 71, 114 ms   (max 114, budget 150)

~36ms of slack at best with nothing else running. Under gate load it goes over, the render timer fires before the paint, and the run is reported as did not render "MCP Servers" within 400ms — which is exactly the misdiagnosis this test exists to catch.

Fix

Widen the constants, not the assertion. The shape under test is an overlap, and only an overlap:

actual paint  <  timeoutMs  <  actual paint + surviveMs

timeoutMs: 1200 / surviveMs: 1500 keeps that with real margin on both sides — worst-case paint lands near 365ms (well under 1200), and 1200 still falls inside a survival window that closes no earlier than ~1750ms. The comment now spells out why both margins are wide and warns against tightening them back for speed.

Verified by mutation, not just by going green

A widened timing test can pass vacuously, so I checked it still fails for the right reason. Removing the clearTimeout(renderTimer) on first paint (render-smoke.mjs:267) — the line this test guards:

✖ a paint landing just under the deadline is not failed by the render timer (1204.191542ms)
ℹ pass 14
ℹ fail 1

It fails at 1204ms, i.e. the render timer still fires inside the survival window. Restored, 15/15 pass.

Verification

npm run local:gate green end to end, with test:scripts at 292/292.

Cost: about a second of extra runtime on one test.

🤖 Generated with Claude Code

https://claude.ai/code/session_013hzwzS8UvBsr4yZm7o7sev

[Copilot]

Pull request overview

Expands timing margins in the render-smoke overlap regression test to prevent load-related flakes.

Changes:

• Increases render timeout to 1200 ms.
• Increases survival window to 1500 ms.
• Documents the timing relationship being tested.

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[cliffhall]

Correct on both counts, and thank you — the fix was right for the wrong reason. Fixed in 30beabb.

You are right that this path never touches script(1): stub() spawns process.execPath directly, and resolvePtyWrapper lives in smoke-tui.mjs — the caller — not in runRenderSmoke. My 66-114ms measurement was taken through the PTY wrapper, so it was measuring a cost this fixture does not pay.

Re-measured the way the test actually spawns (plain process.execPath, 12 samples, idle):

46, 47, 48, 48, 48, 49, 49, 49, 54, 55, 56, 59 ms

So the old 150ms budget had roughly 90ms of slack idle, not the ~36ms I claimed. That is genuinely useful to know rather than a wording nit: it explains why this presented as an occasional flake instead of an obvious bug, and it means the cause is scheduling delay under load — hitting both the child's boot and the parent's timer callback during a full local:gate run — not a fixed startup cost.

The comment now says that, names stub()'s direct spawn explicitly so nobody infers PTY coverage from this fixture, and drops the bogus "worst case 365ms" figure in favour of the reason 1200ms is safe. The constants and the fix are unchanged.

Worth noting the fixture is still verified by mutation rather than by going green: removing the clearTimeout(renderTimer) on first paint (render-smoke.mjs:267) makes it fail at 1204ms, so it still fires inside the survival window and still catches its regression.