ring-096 and formats.t27 agree on f32 (Closes #5746) by gHashTag · Pull Request #5921 · gHashTag/t27

gHashTag · 2026-10-04T06:43:34Z

Closes #5746
Part of #5906

Decision: the spec is the truth, so the ring follows it (option a)

The law that decides it:

docs/T27-CONSTITUTION.md Article SSOT-MATH: "The mathematical, numeric, and physical meaning of Trinity S³AI / t27 has one normative source of truth: specifications in the t27 language (*.t27 files), exercised through the official tri pipeline..."
AGENTS.md: "Specs are source of truth -- behavior belongs in .t27 / .tri"
SOUL.md: "T27 is a spec-first architecture where mathematical truth, not implementation, is the source of truth."
L6 (Constitution): "conformance/FORMAT-SPEC-001.json + specs/numeric/gf16.t27 are the numeric ceiling -- never forked"

The spec was wrong in one place. gf16_to_f32, ternary_to_f32 and quantize_value returned gf16, which lowers to u16, an encoding, although each one produces a float. They now return f32. That is the same boundary the L6 codec already uses: gf16_encode_f32(value: f32) GF16 and gf16_decode_to_f32(gf16: GF16) f32. So the change agrees with L6 and does not fork it.

Changes

specs/numeric/formats.t27
- The three return types change from gf16 to f32.
- Tests no longer call gf16.to_f64, which gf16.t27 never declares. They compare f32 values directly.
- gf16_to_f32_normal_one now uses 0x3E00. Under FORMAT-SPEC-001.json (GF16 bias 31, 2^(E-31) * (1 + M/512)) that is 1.0. The old value 0x3C00 is 0.5; it is binary16's encoding of 1.0.
- ternary_to_f32_is_inverse now checks all three trits.
rings/ring-096-rust/src/lib.rs
- The five functions take and return f32 instead of f64. Internally the encoder still normalises in f64, which is lossless.
- New test f32_boundary_is_lossless_for_every_normal_code: all 63,488 normal GF16 codes go through f32_to_gf16(gf16_to_f32(x)) and come back as the same 16 bits.
- cargo test: 43 pass (42 before).
README: the signatures now say f32.
Seals: .trinity/seals/numeric_Formats.json and .trinity/seals/Formats.json were resealed with t27c seal specs/numeric/formats.t27 --save.
docs/now: new entry for this change.

Drift (`python3 tools/check_ring_spec_drift.py`)

Before (origin/master 7943bc4), rc 1:

DRIFTED ring-096-rust specs/numeric/formats.t27
  hand fns 7, spec fns 6, shared 6, differing 5
    differs: f32_to_gf16 / f32_to_ternary / gf16_to_f32 / quantize_value / ternary_to_f32

After, rc 0:

CONVERGED ring-096-rust specs/numeric/formats.t27
  hand fns 7, spec fns 6, shared 6, differing 0
CONVERGED 3

I did not touch tools/check_ring_spec_drift.py and did not raise any ledger.

Other spec-guards steps (run locally on this branch)

All five self-checks: rc 0.
Seal currency: rc 0, 1283 current, STALE 0.
Duplicate declarations: rc 0.
published_figures self-check: rc 0.
published_figures --check: rc 1, with 8 DRIFT figures. They are identical on master, so this red is pre-existing (Four advisory gates are red on master for inherited reasons, one cause per pull request #5497) and not caused by this PR. On CI the job stops at this step on master and on this branch alike.
diagnostic_capacity: rc 0.
ring_spec_differential.py: ring-090 rc 0 (936/936), ring-099 rc 0 (80/80), ring-096 rc 2. This red is new, and it appears because ring-096 converged: CONVERGED pairs now enter the differential. It cannot be fixed inside this issue's boundary:
- gen-rust emits the spec's u5/u4 constants verbatim, so the harness does not compile (E0425).
- The harness cannot synthesise Trit or Format values. Even with the constants patched, 4 functions are REFUSED.
- The spec functions have no bodies, so the spec side is unimplemented!().
- The harness compares f32 with ==, so code 0xFFFF (NaN on both sides) would always read as a disagreement.
t27c test-report specs/numeric/formats.t27: BLOCKED before and after. On master the cause is gf16.zig FileNotFound. Now it is the spec's bodyless functions (@compileError("not yet implemented")). The seal records tests.blocked. So "its tests pass" can only be met once formats.t27 has bodies.

Left open (outside this boundary)

The ring-096 differential red described above.
formats.t27 has no function bodies.
gf16.t27 contradicts itself:
- its decode formula puts 1.0 at 0x3E00
- its pow2_table[0] and gf16_from_components(0,0,0) tests say 0x3C00
- its encode subtracts the wrong exponent bias
formats.t27 and the ring use NaN 0x7F01, while gf16.t27 uses GF16_NAN = 0xFE01.
The two sides give subnormals different meanings.
The ring's existing subnormal encode is off by a factor of 2 (2^-31 vs 2^-30). This predates the PR.

🤖 Generated with Claude Code

check_ring_spec_drift.py reported ring-096-rust DRIFTED against specs/numeric/formats.t27 on five functions. The constitution decides which side moves: Article SSOT-MATH gives numeric meaning "one normative source of truth: specifications in the t27 language", so the spec is the truth and the ring follows it. The spec itself was wrong in one place: gf16_to_f32, ternary_to_f32 and quantize_value returned `gf16` (a u16 encoding) while their names and callers mean a float. They now return f32, which is the boundary the L6 codec specs/numeric/gf16.t27 already uses (gf16_encode_f32(f32), gf16_decode_to_f32(GF16) f32). Its tests stop calling gf16.to_f64, which gf16.t27 does not declare; gf16_to_f32_normal_one uses 0x3E00, which is 1.0 under FORMAT-SPEC-001.json's GF16 bias 31 (0x3C00 is 0.5). rings/ring-096-rust moves its five functions from f64 to f32. A new test re-encodes all 63,488 normal GF16 codes through the f32 boundary and gets the same 16 bits back. 43 ring tests pass. Both formats seals resealed with `t27c seal specs/numeric/formats.t27 --save`. Drift: ring-096 DRIFTED (differing 5, rc 1) -> CONVERGED (differing 0, rc 0). Part of #5906. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

github-actions · 2026-10-04T06:43:56Z

PR Dashboard

Generated at: 2026-10-04 06:43:47 UTC

ring-096 and formats.t27 agree on f32 (Closes #5746) #5921 ring-096 and formats.t27 agree on f32 (Closes #5746) | Created: 2026-10-04 | Updated: 2026-10-04 | 28 failing / 36 total
build(t27c): reqwest on native-tls, drop the aws-lc-sys build #5920 build(t27c): reqwest on native-tls, drop the aws-lc-sys build | Created: 2026-10-04 | Updated: 2026-10-04 | 8 failing / 37 total
spec(crm-story-reel): v24 -- drop the stale ElevenLabs flag #5915 spec(crm-story-reel): v24 -- drop the stale ElevenLabs flag | Created: 2026-10-04 | Updated: 2026-10-04 | 1 failing / 32 total
build(t27c): put the web stack behind cargo features (212 -> 49 crates) #5914 build(t27c): put the web stack behind cargo features (212 -> 49 crates) | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 40 total
fix(rust): preserve writes to by-value parameter bindings #5911 [GOLD-RING] Infer writable Rust value parameter bindings | Created: 2026-10-04 | Updated: 2026-10-04 | 4 failing / 41 total
fix(t27c): var tuple destructure silences never-mutated Zig locals (Closes #5886) #5902 fix(t27c): var tuple destructure silences never-mutated Zig locals (Closes #5886) | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 46 total
chore(deps): drop unused candle-core/candle-nn from t27c (clean build 142.7 s -> 97.9 s) #5900 chore(deps): drop unused candle-core/candle-nn from t27c (clean build 142.7 s -> 97.9 s) | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 40 total
fix(tri tick): dirty work nobody holds is an anomaly, released claim or not #5896 fix(tri tick): dirty work nobody holds is an anomaly, released claim or not | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 34 total
fix(tri gates shell): a step whose run: key comes first is a step #5894 fix(tri gates shell): a step whose run: key comes first is a step | Created: 2026-10-04 | Updated: 2026-10-04 | 1 failing / 43 total
spec(crm-client-workspace): v6 -- client work carries the client's marks #5893 spec(crm-client-workspace): v6 -- client work carries the client's marks | Created: 2026-10-04 | Updated: 2026-10-04 | 1 failing / 34 total
Port gHashTag/trios:crates/trios-openclaw/rings/OC-01/src/lib.rs (Rust, 3 functions) to specs/port/trios/crates/trios-op #5888 Port gHashTag/trios:crates/trios-openclaw/rings/OC-01/src/lib.rs (Rust, 3 functions) to specs/port/trios/crates/trios-op | Created: 2026-10-04 | Updated: 2026-10-04 | 2 failing / 35 total
Port gHashTag/trinity:fpga/openxc7-synth/d_oscillator.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_o #5887 Port gHashTag/trinity:fpga/openxc7-synth/d_oscillator.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_o | Created: 2026-10-04 | Updated: 2026-10-04 | 5 failing / 35 total
Port scripts/test-agent-bridge.sh (Shell, 1 function) to specs/port/scripts/test-agent-bridge.t27 #5883 Port scripts/test-agent-bridge.sh (Shell, 1 function) to specs/port/scripts/test-agent-bridge.t27 | Created: 2026-10-04 | Updated: 2026-10-04 | 2 failing / 35 total
Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/api/services/a2a/a2a-liveness.ts (TypeScript, 2 functions) to #5879 Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/api/services/a2a/a2a-liveness.ts (TypeScript, 2 functions) to | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 35 total
feat(tri pr-state): --why, and each name under the heading tri printed it under #5868 feat(tri pr-state): --why, and each name under the heading tri printed it under | Created: 2026-10-04 | Updated: 2026-10-04 | 1 failing / 30 total
Port gHashTag/trinity:fpga/openxc7-synth/led_both_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/le #5866 Port gHashTag/trinity:fpga/openxc7-synth/led_both_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/le | Created: 2026-10-04 | Updated: 2026-10-04 | 4 failing / 35 total
fix(tri tick): a released claim is not held, and the dead-tick fix names its files #5864 fix(tri tick): a released claim is not held, and the dead-tick fix names its files | Created: 2026-10-04 | Updated: 2026-10-04 | 1 failing / 34 total
Port gHashTag/trios:crates/trios-scarab-types/rings/SR-01/src/lib.rs (Rust, 6 functions) to specs/port/trios/crates/trio #5855 Port gHashTag/trios:crates/trios-scarab-types/rings/SR-01/src/lib.rs (Rust, 6 functions) to specs/port/trios/crates/trio | Created: 2026-10-04 | Updated: 2026-10-04 | 4 failing / 35 total
feat(tri pr ready): --why compares what a pre-existing failure printed #5853 feat(tri pr ready): --why compares what a pre-existing failure printed | Created: 2026-10-04 | Updated: 2026-10-04 | 3 failing / 38 total
feat(tri): tri harness measures a page signed in -- --save-state / --storage-state, sessionStorage included #5849 feat(tri): tri harness measures a page signed in -- --save-state / --storage-state, sessionStorage included | Created: 2026-10-04 | Updated: 2026-10-04 | 2 failing / 35 total
feat(tri pr ready): ask a check's own workflow when the commit window missed it #5839 feat(tri pr ready): ask a check's own workflow when the commit window missed it | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 41 total
Port gHashTag/trinity:src/vibeec/validation_engine.zig (Zig, 3 functions) to specs/port/trinity/src/vibeec/validation_en #5836 Port gHashTag/trinity:src/vibeec/validation_engine.zig (Zig, 3 functions) to specs/port/trinity/src/vibeec/validation_en | Created: 2026-10-03 | Updated: 2026-10-04 | 5 failing / 35 total
feat(tri): tri harness -- a page measured at 375px and 1024px, every visible frame #5828 feat(tri): tri harness -- a page measured at 375px and 1024px, every visible frame | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 37 total
feat(tri census explain): name the file behind a moved census number #5826 feat(tri census explain): name the file behind a moved census number | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 39 total
tri pr-state: a loop's pull requests with tri pr ready's verdict, not "open" #5824 tri pr-state: a loop's pull requests with tri pr ready's verdict, not "open" | Created: 2026-10-03 | Updated: 2026-10-03 | 1 failing / 32 total
Port gHashTag/trinity:fpga/openxc7-synth/blink_simple.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/bli #5822 Port gHashTag/trinity:fpga/openxc7-synth/blink_simple.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/bli | Created: 2026-10-03 | Updated: 2026-10-04 | 4 failing / 35 total
Port gHashTag/trinity:fpga/openxc7-synth/uart_echo.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/uart_e #5820 Port gHashTag/trinity:fpga/openxc7-synth/uart_echo.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/uart_e | Created: 2026-10-03 | Updated: 2026-10-04 | 4 failing / 35 total
feat(providers): what the Gonka network sells and what TRI-NET would rent #5800 feat(providers): what the Gonka network sells and what TRI-NET would rent | Created: 2026-10-03 | Updated: 2026-10-04 | 1 failing / 32 total
tri: stranded-work finder and cron tick resume card #5787 tri: stranded-work finder and cron tick resume card | Created: 2026-10-03 | Updated: 2026-10-03 | 1 failing / 31 total
Port gHashTag/trinity:fpga/openxc7-synth/d_srl_blink.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_sr #5783 Port gHashTag/trinity:fpga/openxc7-synth/d_srl_blink.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_sr | Created: 2026-10-03 | Updated: 2026-10-04 | 2 failing / 36 total
Port gHashTag/trinity:fpga/openxc7-synth/d_bufg_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_bu #5781 Port gHashTag/trinity:fpga/openxc7-synth/d_bufg_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_bu | Created: 2026-10-03 | Updated: 2026-10-04 | 3 failing / 36 total
Port gHashTag/trinity:fpga/openxc7-synth/led_on_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/led_ #5780 Port gHashTag/trinity:fpga/openxc7-synth/led_on_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/led_ | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 36 total
Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/api/services/queen-lane-credits.ts (TypeScript, 4 functions) #5778 Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/api/services/queen-lane-credits.ts (TypeScript, 4 functions) | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 32 total
bees/merger/publisher: reviewer bee as a service, merger reads discounted checks, publisher heading passes check (Closes #5776) #5777 bees/merger/publisher: reviewer bee as a service, merger reads discounted checks, publisher heading passes check (Closes #5776) | Created: 2026-10-03 | Updated: 2026-10-04 | 0 failing / 31 total
Port gHashTag/trinity:fpga/openxc7-synth/d_g22_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_g22 #5757 Port gHashTag/trinity:fpga/openxc7-synth/d_g22_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_g22 | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 35 total
Port gHashTag/trinity:fpga/openxc7-synth/d_y0_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_y0_t #5756 Port gHashTag/trinity:fpga/openxc7-synth/d_y0_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_y0_t | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 37 total
Port scripts/tri-issue-create.py (Python, 1 function) to specs/port/scripts/tri-issue-create.t27 #5755 Port scripts/tri-issue-create.py (Python, 1 function) to specs/port/scripts/tri-issue-create.t27 | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 37 total
fix(tri): gates preview reads the PR body the way the issue gate does (Closes #5744) #5745 fix(tri): gates preview reads the PR body the way the issue gate does (Closes #5744) | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 42 total
fix(tri): gates preview asks the contexts the ruleset requires, parse-ratchet among them (Closes #5725) #5732 fix(tri): gates preview asks the contexts the ruleset requires, parse-ratchet among them (Closes #5725) | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 43 total
Port gHashTag/trinity:fpga/openxc7-synth/d_e21_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_e21 #5689 Port gHashTag/trinity:fpga/openxc7-synth/d_e21_test.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/d_e21 | Created: 2026-10-03 | Updated: 2026-10-03 | 2 failing / 32 total
spec(automation): kanban-card-chat v7 -- the bee's own log on a running card, live #5680 spec(automation): kanban-card-chat v7 -- the bee's own log on a running card, live | Created: 2026-10-03 | Updated: 2026-10-03 | 5 failing / 32 total
docs(constitution): entrench Article CREDIT and seal it (v1.3) #5667 docs(constitution): entrench Article CREDIT and seal it (v1.3) | Created: 2026-10-03 | Updated: 2026-10-03 | 4 failing / 40 total
specs/port/fpga/verilog/ternary_mac_synth.t27: rewrite w_code_to_mult/product in t27 form (explicit return, element-wise #5664 specs/port/fpga/verilog/ternary_mac_synth.t27: rewrite w_code_to_mult/product in t27 form (explicit return, element-wise | Created: 2026-10-03 | Updated: 2026-10-03 | 1 failing / 36 total
Port gHashTag/trinity:fpga/openxc7-synth/symbiflow_blink.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/ #5663 Port gHashTag/trinity:fpga/openxc7-synth/symbiflow_blink.v (Verilog, 1 module) to specs/port/trinity/fpga/openxc7-synth/ | Created: 2026-10-03 | Updated: 2026-10-03 | 3 failing / 36 total
docs(igla): re-take the corpus figures on master, 1150 specs (Refs #5497) #5610 docs(igla): re-take the corpus figures on master, 1150 specs (Refs #5497) | Created: 2026-10-02 | Updated: 2026-10-02 | 1 failing / 34 total
The Queen's accept is the merge: dispatch.t27 says so, and its harness measures it #5452 The Queen's accept is the merge: dispatch.t27 says so, and its harness measures it | Created: 2026-10-01 | Updated: 2026-10-02 | 6 failing / 36 total
docs(tri): mark the Solana tri-mining program and depin.prove rewards superseded by mint-on-acceptance (Closes #5412) #5413 docs(tri): mark the Solana tri-mining program and depin.prove rewards superseded by mint-on-acceptance (Closes #5412) | Created: 2026-10-01 | Updated: 2026-10-02 | 6 failing / 32 total
Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/lib/container/image-loader.ts (TypeScript, 2 functions) to sp #5320 Port gHashTag/BrowserOS:trios/agent-server/apps/server/src/lib/container/image-loader.ts (TypeScript, 2 functions) to sp | Created: 2026-09-30 | Updated: 2026-10-03 | 0 failing / 3 total
A "using" in prose became an import; and the per-spec delta check compared strings #5215 A "using" in prose became an import; and the per-spec delta check compared strings | Created: 2026-09-29 | Updated: 2026-10-03 | 10 failing / 42 total
Auto-loop 2026-09-29: silent-discard parser family + loop infrastructure #5084 Auto-loop 2026-09-29: silent-discard parser family + loop infrastructure | Created: 2026-09-28 | Updated: 2026-10-02 | 10 failing / 45 total

Summary

Status	Count
Total Open PRs	50
PRs with Failing Checks	47
PRs with All Checks Green	3
READY	2
FAILING	47
PENDING	0
NO CHECKS YET	0

These columns do not partition: 2 + 47 + 0 + 0 = 49, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

⚠️ STALE -- sha256(compiler.rs)=b45a356c2eb6 != manifest seal=87e5cbd3ad94.
The committed NMSE numbers were certified against an older compiler.rs.
Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

github-actions · 2026-10-04T06:44:30Z

📓 NotebookLM Notebook linked to this PR

Notebook ID: b83263109fb055dc
URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.

github-actions Bot mentioned this pull request Oct 4, 2026

bees/merger/publisher: reviewer bee as a service, merger reads discounted checks, publisher heading passes check (Closes #5776) #5777

Open

gHashTag mentioned this pull request Oct 4, 2026

EPIC: master's three inherited reds and the review-speed follow-ups (2026-10-04) #5906

Open

Navigation Menu

ring-096 and formats.t27 agree on f32 (Closes #5746) - #5921

ring-096 and formats.t27 agree on f32 (Closes #5746)#5921
gHashTag wants to merge 1 commit into
masterfrom
fix/ring-096-formats-drift

gHashTag commented Oct 4, 2026

Uh oh!

github-actions Bot commented Oct 4, 2026

Uh oh!

github-actions Bot commented Oct 4, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

gHashTag commented Oct 4, 2026

Decision: the spec is the truth, so the ring follows it (option a)

Changes

Drift (python3 tools/check_ring_spec_drift.py)

Other spec-guards steps (run locally on this branch)

Left open (outside this boundary)

Uh oh!

github-actions Bot commented Oct 4, 2026

PR Dashboard

Summary

Seal Status

Uh oh!

github-actions Bot commented Oct 4, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Drift (`python3 tools/check_ring_spec_drift.py`)