v2: count drift audit -- 81 formats with full disclosure

Replaces v1 (count=77, stale). Live codegen yields 81; full audit table 77/81/83/84 in README. Tracking issue: gHashTag/t27#1064.

README.md

@@ -1,104 +1,81 @@
 ---
-license: mit
+license: cc-by-4.0
 language:
 - en
 tags:
 - numeric-formats
 - floating-point
 - conformance
-- p3109
-- goldenfloat
-- bf16
-- fp8
-- mxfp4
-- posit
+- audit-disclosed
 size_categories:
 - n<1K
-pretty_name: "Numeric Format Catalog (77 formats, Trinity S3AI / t27)"
-configs:
-- config_name: default
-  data_files:
-  - split: formats
-    path: formats_catalog.jsonl
 ---
 
-# Trinity S3AI Numeric Format Catalog
+# numeric-format-catalog v2 (audit-disclosed)
 
-77-format Single Source of Truth (SSOT) for the Trinity S3AI / t27 stack: one row per numeric format with bit-layout, exception policy, and cross-walk metadata.
+**Version:** v2.0 (2026-06-10) -- supersedes v1 (count=77, withdrawn).
 
-## Source of truth
+**Count:** 81 formats across 13 clusters.
 
-This dataset is a **discoverability mirror**. The canonical source is GitHub + Zenodo. Updates flow GitHub -> Zenodo -> Hugging Face (one-way, not the other direction).
+## Honest count-drift disclosure
 
-- **GitHub (primary):** [gHashTag/t27/gen/numeric/formats_catalog.json](https://github.com/gHashTag/t27/blob/master/gen/numeric/formats_catalog.json)
-- **Zenodo DOI:** [10.5281/zenodo.19456875](https://doi.org/10.5281/zenodo.19456875)
-- **Anchor preprint:** [arXiv:2606.05017](https://arxiv.org/abs/2606.05017)
+This dataset is shipped with a known count-drift across four artefacts:
 
-## What is inside
+| Source | Count | Status |
+|---|---|---|
+| `arXiv:2606.09686` Table 1 (paper #3) | 84 | aspirational, MXFP6 split + standalone E8M0 |
+| `specs/numeric/formats_catalog.t27` SSOT raw `// CATALOG:` lines | 83 | live ground truth |
+| **THIS DATASET (codegen LIVE re-run)** | **81** | parser drops `gf512` + `gf1024` on bias-formula int() cast |
+| `gen/numeric/formats_catalog.json` (committed in repo) | 77 | stale, pre-#1051 snapshot |
+| v1 of this dataset (uploaded 2026-06-10 AM) | 77 | inherited stale gen |
 
-77 rows, one per numeric format, covering:
+**Why 81 and not 83 or 84:** the codegen `tools/gen_formats_catalog.py` cannot evaluate bias formulas of the form `2^N-1`, which are the canonical bias values for `gf512` and `gf1024`. Those two rows are silently dropped. We ship what the codegen actually produces, not a cherry-picked number. Patch tracked in [t27#1064](https://github.com/gHashTag/t27/issues/1064).
 
-- IEEE 754 binary family: `binary16`, `binary32`, `binary64`, `binary128`, `binary256`
-- IEEE 754 decimal family: `decimal32`, `decimal64`, `decimal128`
-- Extended precision: `x87_fp80`, `double_double`, `quad_double`
-- AI-oriented: `bfloat16`, `tf32`, `fp8_e4m3`, `fp8_e5m2`, `fp6_e3m2`, `fp6_e2m3`, `fp4_e2m1`
-- Block-scaled: `mxfp8`, `mxfp6`, `mxfp4`, `nf4`, `afp`
-- Posit family: `posit8`, `posit16`, `posit32`, `posit64`
-- Takum family: `takum8`, `takum16`, `takum32`
-- GoldenFloat family: `gf4`, `gf8`, `gf12`, `gf16`, `gf20`, `gf24`, `gf32`
-- ...plus integer, fixed-point, log-number-system, and other rows.
+**Why ship 81 today instead of waiting for the patch:** because v1 shipped 77 yesterday. Replacing a stale 77 with an honest 81 + full audit is closer to the truth than leaving 77 up while the patch lands.
 
-## Load
+## What's in the dataset
 
-```python
-from datasets import load_dataset
+- 81 row entries, each carrying: `id`, `name`, `bits`, `s` (sign bits), `e` (exponent bits), `m` (mantissa bits), `bias`, `phi_distance`, `storage`, `cluster`, `status`, `standard`, `use_case`, `source`
+- 13 cluster groupings: IEEE754 binary, IEEE754 decimal, MLLowPrecision, GoldenFloat, Posit/Unum III, OCP MX, LNS, IntegerFixed, HistoricalVendor, Theoretical, Compression, Extended, QuantTuned
+- Source SSOT: [`gHashTag/t27` `specs/numeric/formats_catalog.t27`](https://github.com/gHashTag/t27/blob/master/specs/numeric/formats_catalog.t27) HEAD `6ecad30`
+- Regen output SHA-256: `546ba6fff1daec0b64483f599778ab63a9410405592c2e9dbc9d90dbb63bc48a`
 
-ds = load_dataset("playra/numeric-format-catalog", split="formats")
-print(ds[0])
-```
-
-## Schema (per-row)
+## Known gaps vs paper #3 (arXiv:2606.09686)
 
-Each row carries the format's `id`, bit-layout fields (sign/exponent/mantissa/special), rounding mode default, exception policy, encoding examples, and provenance metadata. Field set follows the FORMAT-SPEC-001 schema.
+| Cluster | Paper says | This dataset | Delta |
+|---|---|---|---|
+| GoldenFloat | 16 | 16 | 0 (both pre-#1051; current SSOT has 22, of which gen lifts 20) |
+| OCP MX | 5 | 3 | -2 (no MXFP6 E2M3/E3M2 split; no standalone E8M0 block) |
+| MLLowPrecision | 8 | 7 | -1 (NF4 placement) |
+| Total | 84 | 81 | -3 |
 
-## Verification
+## Reconciliation
 
-SHA-256 of the canonical bytes is in `SHA256SUMS.txt`. Reproduce locally:
-
-```bash
-sha256sum formats_catalog.jsonl
-# should match the line in SHA256SUMS.txt
-```
+- **t27#1064** -- tracking issue, five-step plan: parser patch -> regen -> CI invariant -> HF re-upload v3 -> arXiv erratum if needed
+- Expected v3 ship count after parser patch: 83
+- Expected v4 ship count after MXFP6/E8M0 reconciliation with paper: 84 (or paper erratum to whatever codegen produces)
 
 ## Citation
 
-```
-@misc{vasilev2026goldenfloat,
-  author = {Vasilev, Dmitrii},
-  title  = {GoldenFloat: a minimal numeric anchor at phi^2 + 1/phi^2 = 3},
-  year   = {2026},
-  eprint = {2606.05017},
-  archivePrefix = {arXiv},
-}
+If you use this dataset, please cite the paper it accompanies:
 
-@dataset{vasilev2026t27,
-  author = {Vasilev, Dmitrii},
-  title  = {Trinity S3AI / t27 conformance corpus},
-  year   = {2026},
-  doi    = {10.5281/zenodo.19456875},
+```bibtex
+@misc{vasilev2026numeric,
+  title={An 84-Format Numeric Catalog with Bit-Exact Conformance Vectors},
+  author={Vasilev, Dmitrii},
+  year={2026},
+  eprint={2606.09686},
+  archivePrefix={arXiv},
+  primaryClass={cs.MS}
 }
 ```
 
-## License
-
-MIT. See [t27 LICENSE](https://github.com/gHashTag/t27/blob/master/LICENSE).
-
-## Maintainer
-
-- Dmitrii Vasilev (gHashTag), ORCID [0009-0008-4294-6159](https://orcid.org/0009-0008-4294-6159)
-- Trinity S3AI
+Note: paper claims 84; this dataset ships 81 due to the parser bug. See "Honest count-drift disclosure" above.
 
-## Related
+## Provenance
 
-- Companion dataset: [playra/numeric-conformance-packs](https://huggingface.co/datasets/playra/numeric-conformance-packs) -- bit-exact JSON vectors for the GoldenFloat family and phi-identity invariants
-- Paper page: [hf.co/papers/2606.05017](https://huggingface.co/papers/2606.05017)
+- Uploader: `playra` (Dmitrii Vasilev, ORCID `0009-0008-4294-6159`)
+- Affiliation: **Trinity S3AI**
+- Repo: [`gHashTag/t27`](https://github.com/gHashTag/t27)
+- License: CC BY 4.0
+- Generated: 2026-06-10 +07 from HEAD `6ecad30`

formats_catalog.json

@@ -1,5 +1,5 @@
 {
-  "count": 77,
+  "count": 81,
   "formats": [
     {
       "id": "binary16",
@@ -710,7 +710,7 @@
       "storage": "u32",
       "cluster": "GoldenFloat",
       "status": "Experimental",
-      "standard": "this work; L15 PHI_BIAS=1364",
+      "standard": "this work; rule e=round(23/phi^2)=9; normative bias=2^(e-1)-1=255; empirical PHI_BIAS=1364 (=L15) OPEN",
       "use_case": "server inference",
       "gf_relation": "self",
       "source": "specs/numeric/gf24.t27 (spec only)"
@@ -751,54 +751,122 @@
     },
     {
       "id": "gf6",
-      "name": "GF6 (predicted)",
+      "name": "GF6 (rule-derived)",
       "bits": 6,
       "s_bits": 1,
       "e_bits": 2,
       "m_bits": 3,
       "bias": 1,
-      "phi_distance": 0.05,
+      "phi_distance": 0.049,
       "storage": "u8_packed",
       "cluster": "GoldenFloat",
-      "status": "Experimental",
-      "standard": "this work; e=round(5/phi^2)=2, fills FP6 gap",
+      "status": "Open",
+      "standard": "this work; rule e=round(5/phi^2)=2; FP6 E2M3 bridge",
       "use_case": "OPEN R&D: bridge GF4-GF8; FP6 E2M3 hint",
       "gf_relation": "experimental",
-      "source": "section12.5"
+      "source": "specs/numeric/gf6.t27"
+    },
+    {
+      "id": "gf10",
+      "name": "GF10 (rule-derived)",
+      "bits": 10,
+      "s_bits": 1,
+      "e_bits": 3,
+      "m_bits": 6,
+      "bias": 3,
+      "phi_distance": 0.118,
+      "storage": "u16",
+      "cluster": "GoldenFloat",
+      "status": "Open",
+      "standard": "this work; rule e=round(9/phi^2)=3; bridge GF8-GF12",
+      "use_case": "OPEN R&D: tight-precision activations",
+      "gf_relation": "experimental",
+      "source": "specs/numeric/gf10.t27"
+    },
+    {
+      "id": "gf14",
+      "name": "GF14 (rule-derived)",
+      "bits": 14,
+      "s_bits": 1,
+      "e_bits": 5,
+      "m_bits": 8,
+      "bias": 15,
+      "phi_distance": 0.007,
+      "storage": "u16",
+      "cluster": "GoldenFloat",
+      "status": "Open",
+      "standard": "this work; rule e=round(13/phi^2)=5; bridge GF12-GF16; lowest phi-dist below GF48",
+      "use_case": "OPEN R&D: drop-in for fp16 with tighter phi alignment",
+      "gf_relation": "experimental",
+      "source": "specs/numeric/gf14.t27"
+    },
+    {
+      "id": "gf48",
+      "name": "GF48 (rule-derived)",
+      "bits": 48,
+      "s_bits": 1,
+      "e_bits": 18,
+      "m_bits": 29,
+      "bias": 131071,
+      "phi_distance": 0.003,
+      "storage": "u64_padded",
+      "cluster": "GoldenFloat",
+      "status": "Open",
+      "standard": "this work; rule e=round(47/phi^2)=18",
+      "use_case": "OPEN R&D: between GF32 and GF64; tightest phi-dist of the wide rungs",
+      "gf_relation": "experimental",
+      "source": "specs/numeric/gf48.t27"
+    },
+    {
+      "id": "gf96",
+      "name": "GF96 (rule-derived)",
+      "bits": 96,
+      "s_bits": 1,
+      "e_bits": 36,
+      "m_bits": 59,
+      "bias": 34359738367,
+      "phi_distance": 0.008,
+      "storage": "u128_padded",
+      "cluster": "GoldenFloat",
+      "status": "Open",
+      "standard": "this work; rule e=round(95/phi^2)=36",
+      "use_case": "OPEN R&D: between GF64 and GF128 (phi-aligned extended)",
+      "gf_relation": "experimental",
+      "source": "specs/numeric/gf96.t27"
     },
     {
       "id": "gf128",
-      "name": "GF128 (predicted)",
+      "name": "GF128 (rule-derived)",
       "bits": 128,
       "s_bits": 1,
-      "e_bits": 48,
-      "m_bits": 79,
-      "bias": 0,
-      "phi_distance": 0.008,
+      "e_bits": 49,
+      "m_bits": 78,
+      "bias": 281474976710655,
+      "phi_distance": 0.01,
       "storage": "u128",
       "cluster": "GoldenFloat",
-      "status": "Experimental",
-      "standard": "this work; e=round(127/phi^2)=48 (Open: bias TBD)",
+      "status": "Open",
+      "standard": "this work; rule e=round(127/phi^2)=49 (corrects v1.1 typo e=48)",
       "use_case": "OPEN R&D: phi-aligned binary128 alternative",
       "gf_relation": "experimental",
-      "source": "section12.5"
+      "source": "specs/numeric/gf128.t27"
     },
     {
       "id": "gf256",
-      "name": "GF256 (predicted)",
+      "name": "GF256 (rule-derived)",
       "bits": 256,
       "s_bits": 1,
       "e_bits": 97,
       "m_bits": 158,
-      "bias": 0,
-      "phi_distance": 0.005,
+      "bias": 79228162514264337593543950335,
+      "phi_distance": 0.004,
       "storage": "u256_software",
       "cluster": "GoldenFloat",
-      "status": "Experimental",
-      "standard": "this work; e=round(255/phi^2)=97 (Open: bias ~2^71 unconfirmed)",
+      "status": "Open",
+      "standard": "this work; rule e=round(255/phi^2)=97; normative bias=2^96-1",
       "use_case": "OPEN R&D: phi-aligned binary256 alternative",
       "gf_relation": "experimental",
-      "source": "section12.5; bias Open per skill"
+      "source": "specs/numeric/gf256.t27"
     },
     {
       "id": "gf8_bfp",

formats_catalog.md

@@ -7,7 +7,7 @@ Status labels: Verified | EmpiricalFit | Open | Risk | Retracted |
 Experimental | Historical. phi_distance: lower = more phi-aligned;
 -1 sentinel = undefined (non-radix-2 or non-FP).
 
-Total formats: 77.
+Total formats: 81.
 
 | ID | Bits | S:E:M | Bias | phi_dist | Storage | Cluster | Status | Standard | Use case | GF rel. |
 |----|-----:|------|-----:|--------:|---------|---------|--------|----------|----------|---------|
@@ -52,12 +52,16 @@ Total formats: 77.
 | gf12 | 12 | 1:4:7 | 7 | 0.047 | u16 | GoldenFloat | Verified | this work; L0/F3 | mid-range / audio | self |
 | gf16 | 16 | 1:6:9 | 31 | 0.049 | u16 | GoldenFloat | Verified | this work; PHI_BIAS=60; FPGA 35/35 at 323 MHz Artix-7 | training and inference (production) | self |
 | gf20 | 20 | 1:7:12 | 63 | 0.035 | u32 | GoldenFloat | Experimental | this work; 17-squared empirical PHI_BIAS=289 | high-precision edge | self |
-| gf24 | 24 | 1:9:14 | 255 | 0.025 | u32 | GoldenFloat | Experimental | this work; L15 PHI_BIAS=1364 | server inference | self |
+| gf24 | 24 | 1:9:14 | 255 | 0.025 | u32 | GoldenFloat | Experimental | this work; rule e=round(23/phi^2)=9; normative bias=2^(e-1)-1=255; empirical PHI_BIAS=1364 (=L15) OPEN | server inference | self |
 | gf32 | 32 | 1:12:19 | 2047 | 0.014 | u32 | GoldenFloat | Verified | this work; F0 resolved | fp32 drop-in | self |
 | gf64 | 64 | 1:24:39 | 8388607 | 0.003 | u64 | GoldenFloat | Verified | this work; EXP_MAX - BIAS | scientific / double | self |
-| gf6 | 6 | 1:2:3 | 1 | 0.050 | u8_packed | GoldenFloat | Experimental | this work; e=round(5/phi^2)=2, fills FP6 gap | OPEN R&D: bridge GF4-GF8; FP6 E2M3 hint | experimental |
-| gf128 | 128 | 1:48:79 | 0 | 0.008 | u128 | GoldenFloat | Experimental | this work; e=round(127/phi^2)=48 (Open: bias TBD) | OPEN R&D: phi-aligned binary128 alternative | experimental |
-| gf256 | 256 | 1:97:158 | 0 | 0.005 | u256_software | GoldenFloat | Experimental | this work; e=round(255/phi^2)=97 (Open: bias ~2^71 unconfirmed) | OPEN R&D: phi-aligned binary256 alternative | experimental |
+| gf6 | 6 | 1:2:3 | 1 | 0.049 | u8_packed | GoldenFloat | Open | this work; rule e=round(5/phi^2)=2; FP6 E2M3 bridge | OPEN R&D: bridge GF4-GF8; FP6 E2M3 hint | experimental |
+| gf10 | 10 | 1:3:6 | 3 | 0.118 | u16 | GoldenFloat | Open | this work; rule e=round(9/phi^2)=3; bridge GF8-GF12 | OPEN R&D: tight-precision activations | experimental |
+| gf14 | 14 | 1:5:8 | 15 | 0.007 | u16 | GoldenFloat | Open | this work; rule e=round(13/phi^2)=5; bridge GF12-GF16; lowest phi-dist below GF48 | OPEN R&D: drop-in for fp16 with tighter phi alignment | experimental |
+| gf48 | 48 | 1:18:29 | 131071 | 0.003 | u64_padded | GoldenFloat | Open | this work; rule e=round(47/phi^2)=18 | OPEN R&D: between GF32 and GF64; tightest phi-dist of the wide rungs | experimental |
+| gf96 | 96 | 1:36:59 | 34359738367 | 0.008 | u128_padded | GoldenFloat | Open | this work; rule e=round(95/phi^2)=36 | OPEN R&D: between GF64 and GF128 (phi-aligned extended) | experimental |
+| gf128 | 128 | 1:49:78 | 281474976710655 | 0.010 | u128 | GoldenFloat | Open | this work; rule e=round(127/phi^2)=49 (corrects v1.1 typo e=48) | OPEN R&D: phi-aligned binary128 alternative | experimental |
+| gf256 | 256 | 1:97:158 | 79228162514264337593543950335 | 0.004 | u256_software | GoldenFloat | Open | this work; rule e=round(255/phi^2)=97; normative bias=2^96-1 | OPEN R&D: phi-aligned binary256 alternative | experimental |
 | gf8_bfp | 8 | 1:3:4 | 3 | 0.132 | u8_plus_shared_exp | GoldenFloat | Experimental | this work; per-tile shared exponent | OPEN R&D: LLM-quantization-friendly GF8 | experimental |
 | gf_lns_hybrid | 16 | 1:6:9 | 31 | 0.049 | u16_plus_lns_path | GoldenFloat | Experimental | this work; mul in log-space, accumulate Lucas-closed | OPEN R&D: dual-space arithmetic | experimental |
 | mxgf6 | 6 | 1:2:3 | 1 | 0.050 | u8_packed_plus_e8m0 | GoldenFloat | Experimental | this work; OCP MX block + GF6 | OPEN R&D: phi-aligned MX-6 candidate | experimental |