fix(mock): gate BRIA expand ops behind MOCK_PROVIDERS — mock stacks were billing real fal#184
Merged
Conversation
…DERS
Live-caught 2026-07-20: expand_image was the ONE image op without the
mock gate. A stack booted with MOCK_PROVIDERS=1 but EXPAND_MAP_PAN=1
(every local dev boot — .env sets the flag) billed real BRIA outpaint
calls on each Around click. Nobody noticed while the calls succeeded;
today the fal balance ran dry ("User is locked. Reason: Exhausted
balance"), all four pans failed, and the expand-tray spec went red with
an empty tray. CI never caught it because CI doesn't set the pan flag —
i.e. the pan path was ALSO silently untested under mock.
Both BRIA entry points (edge-pinned pan + centered zoomout) now return
the deterministic parchment card, same shape as edit_image/
continue_image. The new test runs with FAL_KEY deleted, so a regression
re-raises the key-missing error instead of billing anyone.
Known remaining hole, deliberately out of scope: video.py (Animate) has
no mock gate either — mock.py has no video helper and no spec exercises
animate. Ledgered.
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
eren23
added a commit
that referenced
this pull request
Jul 20, 2026
…#186) The E5 flip from the interactive-tests campaign (#176–#181): delete continue-on-error from the mock-stack Playwright step. The suite ran green on 16 consecutive CI runs across unrelated PRs (2026-07-19→20), including runs that caught real regressions locally before push (expand-tray red on the mock-billing hole, fixed in #184). From this commit, a failing browser flow fails the PR.
eren23
added a commit
that referenced
this pull request
Jul 21, 2026
… entry point (#187) * fix(mock): gate Animate behind MOCK_PROVIDERS with an embedded real clip Finding #7 from the 2026-07-20 self-play ledger — the last ungated fal entry point, same class as the expand hole (#184): an Animate click on a MOCK_PROVIDERS stack billed a real ltx-video call. mock.py gains mock_video_data_url(): a real 1-second 64x36 H.264 parchment clip (~1.7KB, generated once with ffmpeg, embedded base64) so the mock stack needs no video toolchain. A VALID mp4 matters because the client hands the URL straight to <video src> — a fake payload would render as a broken player and read as a product bug. animate_image returns it before the FAL_KEY check; the test runs under the suite's FAL_KEY-deleted fixture so a regression re-raises loudly, and asserts the ftyp magic to prove the payload is a real mp4. * fix(mock): segment() takes the VLM arm under mock — SAM3 is per-label paid Audit follow-up to the animate gate: grep over _fal_subscribe callers found one more reachable-from-live path — generate.py's extraction refinement calls segment(), and SEGMENTER_PROVIDER=sam3_fal makes that one paid fal call PER LABEL with no gate. Default ("vlm") was already safe via the mocked llm seam; now the flag flip is too. Test spies _segment_sam3 and asserts it never runs under mock.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The find
Self-play follow-through (2026-07-20): the local mock stack's expand-tray spec went red with an empty tray. Backend log:
A stack booted
MOCK_PROVIDERS=1was calling real fal.expand_image(BRIA map-pan) was the one image op without themock.on()gate, and.envsetsEXPAND_MAP_PAN=1— so every Around click in "mock" dev billed four real BRIA outpaints. Nobody noticed while they succeeded; today the balance ran dry and it became a hard failure.CI never caught it because CI's zero-key mock job doesn't set the pan flag — meaning the pan path was also never exercised under mock at all.
The fix
expand_image+expand_image_zoomoutget the exact gateedit_image/continue_imagealready have:mock.on()→ deterministic parchment card, before any key check or upload.Receipts
fal.ai/pan_failed/ upload lines.Known hole left open (deliberate)
video.py(Animate) also has no mock gate — mock.py has no video helper and no spec exercises it. Ledgered rather than scope-crept.🤖 Generated with Claude Code