Cross-check validation against the current (2026) pyfrbus release - #1
Merged
Conversation
The vendored reference is pyfrbus 1.0.0 (2022). Repeating the shock comparison against the Board's 1.1.1 (2026-02-05) gives 1.4e-8 max abs, while the Fed's own two releases differ from each other by 1.3e-8 — our residual is the same scale as the reference implementation's version-to-version noise. Tracking under 1.1.1 is 1.1e-8 vs our 5.6e-17. Also records that the data package the Board labels 'Updated: May 5, 2026' is byte-identical to the vendored April vintage. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Documentation only — no code or gate changes.
Our headline validation compares against the vendored pyfrbus 1.0.0 (2022-04-06). The Board now ships 1.1.1 (2026-02-05) with every module rewritten, including
solver.py,newton.pyandjacobian.py— the exact paths our cross-validation exercises. So the check was repeated against it to confirm the agreement isn't an artifact of one old release.model.xmland LONGBASE are byte-identical between the two releases, so this isolates solver differences:The third row is the point: the Fed's own two releases differ from each other by as much as we differ from either. All three residuals sit at solver-tolerance scale in the same near-zero series (
wpsn,zgap05). 1.1.1 reuses its LU factorization across Newton iterations, recomputing only when the residual fails to halve — that changes the iterate path, not the fixed point.On the tracking invariant, pyfrbus 1.1.1 reproduces LONGBASE to 1.1e-8; ours is 5.6e-17.
The committed reference stays on 1.0.0 so the gate keeps a fixed anchor. Existing gates (< 1e-6) pass comfortably against 1.1.1 output either way. Only the VAR path was exercised; 1.1.1's
jacobian.pylead-substitution fix touches the MCE path, which we don't yet support.Separately: the data package the Board's page labels "Updated: May 5, 2026" is byte-identical to our vendored April vintage (same SHA-256) — noted in
vendor/README.mdso nobody re-does that chase.🤖 Generated with Claude Code