fix(tui): freeze the completion duration at Stop

[dennisonbertram]

Closes #1434

Process note first: this fix was written, committed and pushed hours ago, and I never opened the PR — I was interrupted mid-flow and did not come back to it. The branch sat stranded on the remote while main kept the bug. It surfaced again only because the untrusted reviewer independently re-found the same defect on a later pass over this package. Rebased onto current main and opened now.

The bug

Worked for <d> kept climbing while it was on screen:

tick 0: "· Worked for 5.0s"
tick 1: "· Worked for 5.3s"
tick 2: "· Worked for 5.6s"
tick 3: "· Worked for 5.9s"

The run is already finished. View re-rendered ElapsedSeconds() — a live wall-clock read — on every tick of the completion window, so the figure a user reads was inflated by up to the window's length (~1.2s at 10 ticks x 120ms). For a short run that is a large relative error: a 0.4s run reports as 1.6s.

Fix

Stop freezes the elapsed duration into stoppedAfter, and the completion branch renders that. ElapsedSeconds() keeps its meaning for a live spinner, so nothing else changes.

Two cleanups from the same original review ride along, both in this file:

• shortenLabel carried an unused full parameter — removed.
• Tick indexes holds[m.step] while advancing modulo len(pulse), so an init now enforces that the two slices stay the same length rather than leaving a future edit to panic at runtime.

Verification

The test asserts successive renders are identical rather than asserting a number — a number would be timing-dependent and flaky, and identity is the property that actually matters. A control keeps the frozen value reflecting the real elapsed time, so freezing at zero would fail.

tick 0: "· Worked for 5.0s"
tick 1: "· Worked for 5.0s"
tick 2: "· Worked for 5.0s"
tick 3: "· Worked for 5.0s"

Spinner package green after rebase onto current main.

Provenance

Originally found by openai-gpt-oss-120b ($0.07/M) during a cheap-model evaluation, on a file gpt-6-astra ($10/M list) had already reviewed and passed — Astra had looked at the same code and drawn the wrong conclusion, filing it as snapshot-test nondeterminism, which is a false positive since those tests set startTime explicitly.

Then re-found by Astra on a later pass, which is what recovered this stranded branch.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WJGxhoFhA8JjkwZFcLGdS5

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.