feat(numeric): GF ladder expansion v1.2 — closed-form rule normative (+GF6/10/14/48/96/128/256/512/1024) - #1051
Merged
Merged
Conversation
Normative change (FORMAT-SPEC-001 v1.2): the exp/mant split for every
rung of the binary GoldenFloat ladder is generated by ONE closed rule:
e = round((N - 1) / phi^2)
m = N - 1 - e
bias = 2^(e - 1) - 1
exp_max = 2^e - 1
Anchored at the frozen-silicon ground truth GF16 = 1+6+9, bias 31
(tt-trinity-gamma/src/gf16_v2_mul.v). Lucas identity phi^2 + phi^-2 = 3 = L_2.
Files changed:
- conformance/FORMAT-SPEC-001.json -> v1.2, full 18-rung ladder,
closed_form_rule marked normative, frozen_silicon_anchor recorded.
- docs/NUMERIC_FORMATS_SSOT.md -> normative-rule section, 18-rung
table, GF256 caveat, PHI_BIAS empirical-OPEN list, implementation
status; uniqueness softened to Conj.
- specs/numeric/goldenfloat_family.t27 -> 9 -> 17 binary rungs
(GFTernary/TF3 documented as living outside the array).
- specs/numeric/formats_catalog.t27 -> GF128 corrected (e=49 m=78,
was typo e=48 m=79), GF24 normative bias clarified, added 8 new
rule-derived entries: GF10, GF14, GF48, GF96, GF512, GF1024.
- specs/numeric/{gf6,gf10,gf14,gf48,gf96,gf128,gf256,gf512,gf1024}.t27
-> 9 new per-format spec files (encode/decode constants).
All new rungs ship as status=Open / Conj (no RTL on this repo, no
external DOI; the rule produces exact field widths, that's Verified).
GF16 stays the only Verified rung (production, FPGA 35/35 at 323 MHz).
Per-format empirical PHI_BIAS values stay OPEN and MUST NOT be invented
via Fibonacci/Lucas coincidence; only the normative bias = 2^(e-1) - 1
is canonical.
Refs: skill scientific-works-canon (to update -> v1.3 with 18-rung table).
Co-Authored-By: Perplexity Computer <computer@perplexity.ai>
Closes #1052
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
PR DashboardGenerated at: 2026-06-07 11:15:41 UTC
Summary
|
gHashTag
force-pushed
the
feat/gf-ladder-expansion-v1.2
branch
from
June 7, 2026 11:19
0e95824 to
a1e4937
Compare
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
PR DashboardGenerated at: 2026-06-07 11:19:13 UTC
Summary
|
gHashTag
added a commit
that referenced
this pull request
Jun 13, 2026
…) (#1065) * fix(codegen): handle bias formulas in numeric format SSOT (closes #1064) The bias field in formats_catalog.t27 can carry either a literal int or a rule-derived formula like 2^N-1. Two formats use the formula form: gf512: bias = 2^194-1 gf1024: bias = 2^390-1 Both were added by PR #1051 as limit-of-ladder phi-aligned formats. The parser did int(bias_str) unconditionally, which raised ValueError, so both rows were silently dropped with a "malformed CATALOG line" warning. This is why count drifted: committed gen/ JSON shipped 77 (pre-#1051), post-#1051 SSOT carries 83 raw lines, codegen yielded 81 even after a fresh re-run, paper #3 (arXiv:2606.09686) claims 84. Four sources of truth, none matching. See issue #1064 for the full audit. Fix: - parse_bias(s) helper: tries int(); falls back to evaluating simple 2^N+offset forms within int64 range; otherwise returns sentinel -2 with the raw string in a new bias_formula field - Format dataclass: adds bias_formula: str (preserves the raw SSOT string for every row, including the literal-int cases for symmetry) - All existing emitters (Rust struct, C header, TS interface, Python, Markdown table, etc.) continue to render the int bias field as before; bias_formula appears only in JSON and python emitter where asdict() picks it up automatically -- non-breaking for the typed emitters that hardcode their Format schema. Verification after fix: $ python3 tools/gen_formats_catalog.py specs/numeric/formats_catalog.t27 gen/numeric/ $ python3 -c "import json; print(json.load(open('gen/numeric/formats_catalog.json'))['count'])" 83 gf512 and gf1024 now appear in the JSON with bias=-2 and the formula preserved in bias_formula. Downstream tools that need the actual i64 value can carry the string forward and evaluate at use-site (mpmath or big-int arithmetic) without losing provenance. This does NOT yet reconcile SSOT 83 vs paper #3 claim of 84; that delta is the MXFP6 split + standalone E8M0 in Table 1 vs the SSOT's collapsed mxfp6 row + Microscaling cluster layout. See #1064 for the follow-up plan. The CI invariant (raw line count == JSON count) will land in a separate PR. Refs: #1051 (GF ladder expansion), #1063 (HF Discovery thread), #1064 * fix(codegen): revert auto-generated gen/ artefacts, add NOW.md entry PR #1065 originally committed regenerated gen/numeric/* outputs alongside the source fix in tools/gen_formats_catalog.py. Per L2 generation rules, gen/ artefacts must be auto-regenerated post-merge by ./scripts/tri gen, not committed in source PRs. This commit: - reverts gen/numeric/* back to master state - keeps the source fix (tools/gen_formats_catalog.py) intact - adds docs/NOW.md entry per NOW Sync Gate Closes #1064 --------- Co-authored-by: gHashTag <playra@trinity-s3ai.org> Co-authored-by: gHashTag <admin@t27.ai>
gHashTag
pushed a commit
that referenced
this pull request
Jun 14, 2026
Regenerate gen/numeric/* from SSOT formats_catalog.t27 (// CATALOG: lines = 83, HEAD 16042f4). Committed gen JSON was stale at 77 (pre-#1051 snapshot). - regen all 16 tracked codegen targets: SSOT == regen == committed gen == 83 - add tools/check_catalog_count.py + .github/workflows/catalog-count-gate.yml CI invariant: SSOT line-count == codegen count == committed JSON count - ERRATA_2026-06-14.md: arXiv:2606.09686 declares 84; canonical live = 83 - docs/: P3109 4-param cross-walk, kappa FP8 E4M3 note, conformance-layer positioning Closes #1079
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
2 participants
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.
Closes #1052
Summary
Promotes the GoldenFloat ladder exp/mant split to a single normative closed-form rule:
Anchored at frozen silicon
gf16_v2_mul.v(GF16 = 1+6+9, bias 31,tt-trinity-gamma). Lucas identityphi^2 + phi^-2 = 3 = L_2.Ladder (18 rungs, rule-derived)
GFTernary (special) | GF4 | GF6 (NEW) | GF8 | GF10 (NEW) | GF12 | GF14 (NEW) | GF16 (Verified, silicon) | GF20 | GF24 | GF32 | GF48 (NEW) | GF64 | GF96 (NEW) | GF128 (CORRECTED 49/78, was 48/79) | GF256 | GF512 (NEW) | GF1024 (NEW) | TF3 (container)
Files
conformance/FORMAT-SPEC-001.json-> v1.2 + frozen_silicon_anchordocs/NUMERIC_FORMATS_SSOT.md-> 18-rung table + PHI_BIAS-empirical-OPENspecs/numeric/goldenfloat_family.t27-> 9 -> 17 binary rungsspecs/numeric/formats_catalog.t27-> fix GF128 typo, +8 rule-derived entriesspecs/numeric/gf{6,10,14,48,96,128,256,512,1024}.t27-> 9 new spec filesStatus discipline
Conj(no RTL on this repo, no external DOI).Verifiedarithmetic.Verifiedend-to-end rung (FPGA 35/35 @ 323 MHz).OPEN. Do NOT invent via Fibonacci/Lucas coincidence.Companion PRs to follow
tt-trinity-corona: fix 6 broken rungs intools/gen_rom.pyCATALOG cluster 3 (GF16/24/32/48/64/96/128/256), expand RECORD_COUNT 80 -> 82 for GF512/GF1024 spec recordstt-trinity-gamma: add GF512/GF1024 tospecs/numeric/tri_net_formats.t27goldenfloat-preprint: extend table to 11 GF rungs (v19->v20)arith2027-goldenfloat: align paper + OUTLINE + READMEclaim-audit-lab: NEW CASE-09 "Corona ROM CATALOG vs closed-form rule self-audit" (Conj,Fpath)