Parent: #588 — step 7 of 8.
Problem
Fidelity testing needs a Figma token and a captured design baseline. That is the differentiated thing, and it is also a wall in front of a customer who just wants to know whether today's regeneration broke yesterday's output.
There is a second, much cheaper question the same machinery answers: did this change alter any render at all, measured against the last renders we accepted. No Figma access, no manifest staleness, no variant join ambiguity — and it is what the original zero-flake measurement actually measured: 142 of 142 stories, three full runs, exactly 0 differing pixels on unchanged stories.
Potential solution(s)
- An accepted-render tree beside the Figma baselines, keyed identically.
specs testing visual accept [--components <keys...>] [--all] promotes renders from the current run into it.specs testing visual diff --against accepted scores against it — the flag already exists from step 5; this makes it work.accept refuses in fidelity mode, with a sentence saying why. Percy and Chromatic are built around approving a changed snapshot; here the Figma export is the design, and a render that disagrees with it is wrong by definition. An accept that could write over a Figma baseline would turn the tool into one that agrees with whatever the emitter last did.
Acceptance criteria
Case data
Notes
Depends on step 5. Independent of steps 6 and 8.
Parent: #588 — step 7 of 8.
Problem
Fidelity testing needs a Figma token and a captured design baseline. That is the differentiated thing, and it is also a wall in front of a customer who just wants to know whether today's regeneration broke yesterday's output.
There is a second, much cheaper question the same machinery answers: did this change alter any render at all, measured against the last renders we accepted. No Figma access, no manifest staleness, no variant join ambiguity — and it is what the original zero-flake measurement actually measured: 142 of 142 stories, three full runs, exactly 0 differing pixels on unchanged stories.
Potential solution(s)
specs testing visual accept [--components <keys...>] [--all]promotes renders from the current run into it.specs testing visual diff --against acceptedscores against it — the flag already exists from step 5; this makes it work.acceptrefuses in fidelity mode, with a sentence saying why. Percy and Chromatic are built around approving a changed snapshot; here the Figma export is the design, and a render that disagrees with it is wrong by definition. Anacceptthat could write over a Figma baseline would turn the tool into one that agrees with whatever the emitter last did.Acceptance criteria
acceptfollowed immediately by a diff against accepted reports 0 differing pixels across the board.acceptwith neither a component scope nor--allrefuses rather than promoting everything.acceptwhile the configured mode is fidelity refuses with an explanation and writes nothing.Case data
Notes
Depends on step 5. Independent of steps 6 and 8.