An animated solve must fit its budget, and a burst must never be cut in half

[JWriter20]

overallSolveTimeoutMs / overall_solve_timeout_ms is 45 s and it is sized for rounds — "a round costs ~4-7 s, so six is the budget", which no-progress.test.ts pins. Recording an animated challenge is not a round. It is a fixed extra stage costing the burst, the slice, one multi-image inference (six keyframes, several times a still's) and the wait for the widget to come back round to the frame the model chose. Nothing in the 45 s was ever set aside for it.

So every escalation to a recording ran the clock out, and reported a timeout — a message about the model being slow, for a budget that had no room for what the solver had just decided to do.

Measured

CaptchaKrakenFinetune Tier 3 run 32596340560, 2026-08-22:

port	fixture	elapsed	reported as
python	hcaptcha_fish_swim_different	45.7s, 48.5s, 49.2s	exceeded overall_solve_timeout_ms during recording the animated challenge
python	hcaptcha_number_with_highest_value_video	49.4s, 50.5s	same
python	hcaptcha_tile_flip_video	52.6s, 52.1s	same
js	hcaptcha_click_image_by_traits	52.4s, 49.7s	Captcha solve timed out after 45000ms (attempt 6/6)
js	hcaptcha_connect_path	50.2s	same
js	hcaptcha_grid_3x3_property	49.7s, 49.4s	same

The same fixtures solve in 11-20 s on the rounds where the still path happens to answer them. It is the escalation that does not fit, not the puzzle.

The escalation buys its own budget, once

video_budget_ms() = burst + keyframe wait + one multi-image inference; 18 s on the defaults, granted the first time a recording starts and never again in the same solve. Derived, so a longer burst carries its own budget rather than quietly reintroducing this. Worst case 63 s, still bounded.

An extension, not a looser default: a solve that never escalates keeps exactly the deadline the caller configured, and video_solve_enabled=False gets no grant at all — that switch already means "fail fast rather than spend the recording time".

Both ports, same arithmetic, because CLAUDE.md 1c — a fixture that passed on one port and timed out on the other reads as a driver bug.

A half-recorded burst is worthless

The python port checked the overall deadline per frame, so a burst starting at 41 s died at frame 27 of 40 — throwing away both the frames and the ~3 s spent making them. The slicer reads a clip's temporal structure; stopping early does not yield a shorter answer, it yields a recording that may not contain the screen the answer is on.

Checked once now, before the first frame, and if the budget cannot fit a whole burst it says so and names the knobs. A separate, looser bound still catches a genuinely hung screenshot and reports it as that.

Tests

9 new. Both regression tests fail on the old behaviour with the exact production message:

FAILED test_a_burst_is_never_abandoned_partway_for_being_over_budget
  'captcha solve exceeded overall_solve_timeout_ms (45000ms) during recording the animated challenge'
FAILED test_recording_extends_the_deadline_once_and_only_once

python/tests 405 passed / 15 skipped, js 85 passed, tsc --noEmit clean.

Pre-existing on main and untouched here: 4 errors in test_browser_compat.py — identical on a clean origin/main, and already fixed on #10's branch.

🤖 Generated with Claude Code

[JWriter20]

✓ Tier 3 driver-gate — pass

Aggregate: 0.852 · 6/10 families passing · 13 pair(s) not solved

Port	Pairs solved
js	0.886
python	0.818

Vendor family	Solve rate
botdetect	1.000
geetest	0.786
hcaptcha	0.885
lemin	1.000
mtcaptcha	1.000
prosopo	1.000
recaptcha	0.500
tencent	0.000
yandex	1.000
yidun	1.000