Repository navigation
fix(conformance): correct 5 stale gf16 vectors + wire oracle into CI (R7-loop #2) - #1673
Merged
Merged
Conversation
conformance/gf16_ref.py exited non-zero on 5 named-constant vectors (phi_inverse, consciousness_threshold, gravity_strength, quantum_spacing, phase_transition). Proven by brute force over all 65536 codes: the encoder picks the NEAREST GF16 code for each, and decode is FPGA-consistent + roundtrips 19/19 -> the vectors' expected.decoded values were phantom (not representable by any GF16 code). Set expected.decoded to the true constant and tolerance_abs to the format's quantization bound (widened phase_transition 0.0002 -> 0.0005, whose ULP error is 3.9e-4). Oracle now 35 pass / 0 fail. Added .github/workflows/gf16-conformance.yml so the oracle runs on every change to it or its vectors (it was never in CI, which is how the drift went unnoticed). Closes #1579 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
Satisfies the required check-now-freshness gate (t27#141): every PR to master must update docs/NOW.md. No code change.
gHashTag
enabled auto-merge
August 5, 2026 12:51
Contributor
PR DashboardGenerated at: 2026-08-05 12:52:01 UTC
Summary
Seal Status
|
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem (#1579)
conformance/gf16_ref.pyexited non-zero on 5 named-constant vectors, and no CI ran it — so the drift went unnoticed.Root cause (proven, not guessed)
Brute force over all 65536 GF16 codes shows the encoder picks the nearest code for each constant, decode is FPGA-consistent (32/32) and roundtrips (19/19). The vectors'
expected.decodedvalues (e.g.0.617919922for phi_inverse) are not representable by any GF16 code — phantom/stale constants.Fix
expected.decodedto the true mathematical constant,tolerance_absto the format's quantization bound (only phase_transition needed widening, 0.0002 → 0.0005; its ULP error is 3.9e-4). Semantic is now "GF16 encoding of X lands within one ULP of X" — meaningful, not tautological..github/workflows/gf16-conformance.yml(stdlib-only) so the oracle runs on every change to it or its vectors.Oracle: 35 pass / 0 fail (was 30/5), exit 0.
Not in scope: #1580 (
gf_ref.pydenormalFractionblowup) is a separate architectural rewrite.Closes #1579
🤖 Generated with Claude Code