mirror of
https://github.com/heygen-com/hyperframes.git
synced 2026-09-01 19:42:03 +00:00
docs(engine): point the depth-sign comment at the ticket that would cover it
The comment already admits the self-verify net cannot catch a re-inversion here. Three reviewers asked for the render-parity comp that could. That is not a fixture drop — the harness has no channel for `HF_FAST_CAPTURE_3D`, the golden has to be produced on the Linux/Docker CI platform, and the sibling fast-capture fixture is itself excluded from every shard — so it is scoped in PRINFRA-570 and named here, along with the fact that until it lands this sign has no CI coverage at all.
This commit is contained in:
@@ -816,7 +816,11 @@ export async function initThreeDProjectionInPage(): Promise<ThreeDProjectionResu
|
||||
// rendering a two-plane preserve-3d comp on both capture paths and
|
||||
// comparing; the self-verify net CANNOT catch a regression here, because
|
||||
// it captures ground truth after this rewrite and would compare wrong
|
||||
// geometry against wrong geometry.
|
||||
// geometry against wrong geometry (measured: a 65 dB "pass" over a
|
||||
// truly-broken 18.8 dB render). The only backstop that can see this class
|
||||
// of regression is a render-parity comp in a regression shard, which
|
||||
// needs harness plumbing rather than a fixture drop — PRINFRA-570 carries
|
||||
// the scope. Until it lands, THIS SIGN IS UNCOVERED BY CI.
|
||||
const ndc: Mat4 = [
|
||||
2 / canvasW,
|
||||
0,
|
||||
|
||||
Reference in New Issue
Block a user