# TrustFortress / cubie-tf — Grounded Claims Registry

> The four headline claims, each with pinned source evidence, direct downloads, reproduce commands, and honest scope boundaries.

**This file is generated — do not edit by hand.** Canonical source: [`apps/resource-library/src/claims-data.json`](../../apps/resource-library/src/claims-data.json). Regenerate with `python tools/gen_claims_md.py`. Live registry: [https://lib.trustfortress.ai/claims](https://lib.trustfortress.ai/claims) · JSON: [https://lib.trustfortress.ai/api/claims](https://lib.trustfortress.ai/api/claims).

- **Generated:** 2026-07-04 · **Fact-checked:** 2026-07-04
- **Pins:** cubie-tf `88ef7e8303ed82b4e0af398960be172f8bb560db` · proof snapshot `c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08` · cubie-research `1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4`

Single source of truth for the four headline claims. Every claim was independently fact-checked on 2026-07-04 and corrections were applied where needed (the full fact-check audit trail is in the in-repo CLAIMS.md). Evidence is pinned to immutable commit SHAs. Source links marked "private" require cubie-tf / cubie-research repository access (both proprietary); public downloads are served from lib.trustfortress.ai. Numbers quantifying repo state are provenance snapshots — re-run the reproduce commands to re-measure.

| Claim | Status | Scope |
|---|---|---|
| [Triple-kernel (Coq/Lean/Verus) CUB theorem corpus with in-crate proof-to-code bindings — mixed-snapshot scoped](#proof-corpus) | 🟡 Scoped | Mixed-corpus snapshot — closure is per-file / manifest-scoped, NOT a blanket zero-axiom certificate. The proofs are standalone: 0/476 Verus specs import the runtime crates, so the corpus does not formally verify the shipping Rust. |
| [Cubie-native TEP fault detector: 100/100/100 FDR at d00 FAR=0.000% on the closed-loop-masked trio (CUB-1921 CUSUM OR-gate)](#tep-far0) | 🟡 Scoped | In-sample calibration identity: PASS-B sets h = 1.5x the observed d00 peak, so FAR = 0 by construction (CUB-1952). The 100/100/100 FDR is in-sample on the Braatz/Rieth d00/d03/d09/d15 set; held-out out-of-sample transfer is NOT yet established (tep_oos_results.md = VERIFIED_BAD). |
| [TF Logs agentic-waste findings: GPU deadlock blast radius, VRAM lock, request-waste rate across ~45.3M real telemetry rows](#agentic-waste) | ✅ Verified | Re-measured from the committed result JSONs (not the prior memory dossier). ~99.5% output<input waste is scoped to the Azure-2024 trace (44.1M of the ~45.3M rows); only the 66.85% blast-radius figure appears in a whitepaper. |
| [EU AI Act coverage (Articles 9/10/11/13/14/15 + Annex IV) via cubie-tf crates/cubie-eu-*](#eu-ai-act) | 🟡 Scoped | A compiling Rust evidence-mapping + Annex IV generation SCAFFOLD (per-article structs + AnnexIvBuilder + /api/v1/annex-iv), NOT a fully populated conformity pack — modules carry placeholder values (e.g. kill_switch_url=example.com, some proof counts hardcoded 0). |

---

<a id="proof-corpus"></a>

## Triple-kernel (Coq/Lean/Verus) CUB theorem corpus with in-crate proof-to-code bindings — mixed-snapshot scoped

**Status:** 🟡 Scoped
**Scope boundary:** Mixed-corpus snapshot — closure is per-file / manifest-scoped, NOT a blanket zero-axiom certificate. The proofs are standalone: 0/476 Verus specs import the runtime crates, so the corpus does not formally verify the shipping Rust.

### Claim

cubie-tf and cubie-research ship a triple-kernel formal proof corpus (Coq 8.18, Lean 4, Verus) keyed by CUB theorem IDs. The published cubie-tf snapshot (baseline commit c3d4ae27, verified SHA-256 0a60569...fbbd) is a MIXED corpus: coq/ 487 files, lean/ 563 files, verus/ 476 files — containing both closed proofs and statement-form/stub material. It is NOT a blanket zero-axiom / proof-closure certificate; its own published library MANIFEST (/objects/cubie-tf/verified-proofs/2026-07-04/MANIFEST.md) says "Do not cite the full ZIP as uniformly verified evidence" and names known-unclosed files (coq/CUB_2000_CubeObjectSize192.v Admitted; verus/cubie_adversarial_robustness_theorem_spec.rs ensures-true stub). Live re-measured flagged counts on the snapshot: Coq 138/487 files contain an `Admitted` token, Lean 258/563 contain a `sorry` token, Verus 142/476 contain a stub marker (external_body 42, ensures-true 77, assume 51) — but these are coarse file-presence upper bounds: many hits are in comments (e.g. only ~52/563 Lean files have `sorry` on a non-comment line), so genuinely-unclosed file counts are materially lower. The proofs are STANDALONE: 0/476 verus specs import the crates, and CI runs the Verus kernel check with `verus --crate-type lib` on the spec files STANDALONE (proof-kernel-check.yml over the spec corpus, exit 1 on failure; proof-kernel-verification.yml a hard gate, continue-on-error:false, on the narrow CUB-1940..1966 TEP band proofs/*/tep/CUB_19[4-6][0-9]_*), so the corpus does NOT formally verify the runtime crate code. Proof-to-code binding is by CUB-ID cross-reference plus behavioral Rust tests, not Verus-over-runtime: CUB-1921 (Page-CUSUM aggregator) has coq/CubieCusumAggregator.v + lean/CubieCusumAggregator.lean + verus/cubie_cusum_aggregator_spec.rs (proofs), cubie-tep/src/cusum.rs (code), and cubie-tep/tests/cusum_wiring.rs (behavioral OR-gate wiring test naming CUB-1921). Separately, in-crate `verus! {}` blocks exist (cubie-core/src/replay.rs, cubie-platform/src/uah.rs, etc.) but erase to plain Rust under normal cargo build. The v6 TEP whitepaper (repo cubie-math revision, 2026-05-26) claims a broader envelope — CUB-0704..CUB-1966 (1,966 unique IDs), 1,145 IDs with full triple-kernel parity, and substantive (no admit/sorry) CI-enforced bodies specifically for the CUB-1940..1966 band; those are whitepaper-revision numbers, not the cubie-tf snapshot's live verified-ID count. cubie-research holds a SEPARATE, tightly-scoped triple-kernel corpus (proofs/coq 36, lean4 22, verus 23; provenance commit 1e8c3e26): CUB-TSP-001–006, 008–012, 014–016 (14 proven in all three kernels) with Coq coqchk `Axioms: <none>`, Lean on 3 standard axioms (no sorry), Verus no assume/admit — but its README explicitly bounds this as a seal over axiom-free Coq DEFINITIONS ("trivial" on definitions), not a P vs NP result.

### Direct downloads (public — lib.trustfortress.ai)

- [⬇ Published proof-corpus MANIFEST (scoping: mixed snapshot, not uniformly verified)](https://lib.trustfortress.ai/objects/cubie-tf%2Fverified-proofs%2F2026-07-04%2FMANIFEST.md)
- [⬇ Published proof-corpus ZIP snapshot (2723732 bytes, SHA-256 verified live)](https://lib.trustfortress.ai/objects/cubie-tf%2Fverified-proofs%2F2026-07-04%2Fcubie-tf-verified-proofs-corpus-20260704.zip) — `sha256 0a60569028a0b6653444510343e435d35b57af9cdf044c39e808cca43a19fbbd`
- [⬇ v6 TEP whitepaper (CUB-0704..1966, 1145-ID triple parity, CUB-1940..1966 substantive; cubie-math revision 2026-05-26)](https://lib.trustfortress.ai/objects/cubie-tf%2Fv6-whitepaper-evidence%2F2026-07-03%2Fv6_cubie_tep_whitepaper.pdf)

### Source evidence (pinned)

| Artifact | Source | sha256 |
|---|---|---|
| cubie-tf coq/ proof dir (487 .v files at snapshot baseline) | [🔒 cubie-tf/coq/CubieCusumAggregator.v](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/coq/CubieCusumAggregator.v)@`c3d4ae27` |  |
| cubie-tf lean/ proof dir (563 .lean files at snapshot baseline) | [🔒 cubie-tf/lean/CubieCusumAggregator.lean](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/lean/CubieCusumAggregator.lean)@`c3d4ae27` |  |
| cubie-tf verus/ proof dir (476 .rs files; standalone, 0 import crates) | [🔒 cubie-tf/verus/cubie_cusum_aggregator_spec.rs](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/verus/cubie_cusum_aggregator_spec.rs)@`c3d4ae27` |  |
| CUB-1921 proof-to-code binding: cusum.rs runtime code | [🔒 cubie-tf/cubie-tep/src/cusum.rs](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/cubie-tep/src/cusum.rs)@`c3d4ae27` |  |
| CUB-1921 behavioral wiring test binding proof-ID to live TepDetector | [🔒 cubie-tf/cubie-tep/tests/cusum_wiring.rs](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/cubie-tep/tests/cusum_wiring.rs)@`c3d4ae27` |  |
| In-crate verus! block (erases under cargo build; not CI-verified against runtime) | [🔒 cubie-tf/cubie-core/src/replay.rs](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/cubie-core/src/replay.rs)@`c3d4ae27` |  |
| CI Verus kernel check runs `verus --crate-type lib` on the spec files standalone; proof-kernel-verification.yml is a hard gate (continue-on-error:false) on the CUB-1940..1966 TEP band, and 0/476 specs import the runtime crates — so CI does not formally verify the runtime crate code. | [🔒 cubie-tf/.github/workflows/proof-kernel-verification.yml](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/.github/workflows/proof-kernel-verification.yml)@`88ef7e83` |  |
| Known-unclosed example flagged by manifest (Admitted) | [🔒 cubie-tf/coq/CUB_2000_CubeObjectSize192.v](https://github.com/iamdatanick/cubie-tf/blob/c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08/coq/CUB_2000_CubeObjectSize192.v)@`c3d4ae27` |  |
| cubie-research proofs/coq (36 .v, CUB-TSP band, coqchk Axioms: none on definitions) | [🔒 cubie-research/proofs/coq/CubTsp001.v](https://github.com/iamdatanick/cubie-research/blob/1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4/proofs/coq/CubTsp001.v)@`1e8c3e26` |  |
| cubie-research proofs/lean4 (22 .lean, 3 standard axioms, no sorry) | [🔒 cubie-research/proofs/lean4/CubTsp001.lean](https://github.com/iamdatanick/cubie-research/blob/1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4/proofs/lean4/CubTsp001.lean)@`1e8c3e26` |  |
| cubie-research proofs/verus (23 .rs, no assume/admit) | [🔒 cubie-research/proofs/verus/cub_tsp_001.rs](https://github.com/iamdatanick/cubie-research/blob/1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4/proofs/verus/cub_tsp_001.rs)@`1e8c3e26` |  |
| cubie-research proof SHA-256 ledger (127 files, provers coq 8.18.0 / lean4 4.31.0 / verus 0.2026.06.28) | [🔒 cubie-research/ledger/proof_hashes.json](https://github.com/iamdatanick/cubie-research/blob/1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4/ledger/proof_hashes.json)@`1e8c3e26` |  |
| cubie-research canonical verification script | [🔒 cubie-research/tools/verify_proofs.ps1](https://github.com/iamdatanick/cubie-research/blob/1e8c3e2690c0b0bc3d0d4d6e70c73d09ad2fb5d4/tools/verify_proofs.ps1)@`1e8c3e26` |  |

> 🔒 Source links require **cubie-tf / cubie-research** repository access (both proprietary). Public downloads above are on lib.trustfortress.ai.

### Reproduce this result

```bash
# 1. Verify the published cubie-tf snapshot ZIP hash.
git archive --format=zip -o cubie-tf-verified-proofs-corpus-20260704.zip c3d4ae2783f4de9403dd3a3f02abcc1cc8a53a08 coq/ lean/ verus/
sha256sum cubie-tf-verified-proofs-corpus-20260704.zip   # git-archive ZIP container hashes are Git/toolchain-sensitive; use the published object below as canonical
# Download the published object directly and hash it:
curl -sL -o corpus.zip 'https://lib.trustfortress.ai/objects/cubie-tf%2Fverified-proofs%2F2026-07-04%2Fcubie-tf-verified-proofs-corpus-20260704.zip' && sha256sum corpus.zip   # must equal 0a60569028a0b6653444510343e435d35b57af9cdf044c39e808cca43a19fbbd
# 2. Re-measure file counts and real-vs-flagged binding counts on the snapshot:
unzip -q corpus.zip -d ex
find ex/coq -name '*.v' | wc -l        # 487
find ex/lean -name '*.lean' | wc -l    # 563
find ex/verus -name '*.rs' | wc -l     # 476
grep -rlE '\bAdmitted\b' ex/coq --include='*.v' | wc -l     # 138 files (coarse upper bound; some in comments)
grep -rlE '\bsorry\b' ex/lean --include='*.lean' | wc -l    # 258 files raw; ~52 with sorry on a non-comment line
grep -rlE 'external_body|ensures[[:space:]]+true\b|assume\(' ex/verus --include='*.rs' | wc -l  # 142 files
grep -rlE 'use +crate::|use +cubie_core|extern crate cubie' ex/verus --include='*.rs' | wc -l    # 0 -> standalone, not runtime-wired
# 3. Per-kernel verifier checks on a CLOSED file (scoped, not the whole ZIP):
coqc ex/coq/CubieCusumAggregator.v && coqchk ex/coq/CubieCusumAggregator.vo   # expect Axioms: <none> on closed modules
lake env lean ex/lean/CubieCusumAggregator.lean   # then '#print axioms <thm>' per theorem
verus --crate-type lib ex/verus/cubie_cusum_aggregator_spec.rs   # standalone per-file check in CI; NOT report-only
# 4. cubie-research triple-kernel corpus (clone iamdatanick/cubie-research at 1e8c3e26):
pwsh tools/verify_proofs.ps1 -Build   # runs all proofs + coqchk Axioms: <none> audit + SHA-256 ledger integrity
python tools/gen_sha_ledger.py --check # verify ledger/proof_hashes.json root_sha256 7881e3a9...d54
```

### Verification checks

- MANIFEST self-scoping: 'mixed corpus snapshot ... Do not cite the full ZIP as uniformly verified evidence' — status must be scoped/partial, never 'verified'.
- Snapshot SHA-256 reproduced live 2026-07-04: 0a60569028a0b6653444510343e435d35b57af9cdf044c39e808cca43a19fbbd (2723732 bytes) — matches manifest exactly.
- File counts reproduced on the verified ZIP: coq 487, lean 563, verus 476 (match manifest table).
- Flagged-file counts are file-presence UPPER BOUNDS: e.g. Lean raw sorry-in-file=258 but sorry-on-non-comment-line~=52; treat as 'files containing marker', not 'unverified theorems'.
- 0/476 verus specs import crates + CI runs verus --crate-type lib per-file STANDALONE (proof-kernel-verification.yml is a hard gate on the CUB-1940..1966 band) => corpus is standalone; 'proofs verify the runtime' would be overstated. Binding = CUB-ID cross-ref + behavioral tests.
- Whitepaper numbers (1966 IDs, 1145 triple-parity, CUB-1940..1966 substantive) are cubie-math revision 2026-05-26 claims — attribute to whitepaper, do NOT present as cubie-tf snapshot's live verified-ID count.
- cubie-research 'Axioms: <none>' is scoped by its own README to axiom-free Coq DEFINITIONS ('trivial' on definitions); 14 CUB-TSP IDs proven all-three-kernels, no P-vs-NP claim.

### Known boundaries / open gaps

- Exact verified-vs-unclosed THEOREM count per kernel on the cubie-tf snapshot is not published as a machine-readable inventory; only file-presence marker counts (measured here) and two named example files are authoritative. A per-theorem coqchk/#print-axioms/verus sweep would be needed to produce true closed-theorem counts; the MANIFEST deliberately declines to make that blanket claim.
- Whitepaper's 1,145-ID triple-kernel-parity and 1,966-unique-ID figures are stated for the cubie-math revision (2026-05-26) and were not re-derivable against the cubie-tf snapshot within this task; the snapshot's live unique-CUB-ID parity count is unmeasured.
- The v6 whitepaper PDF header cites github.com/iamdatanick/cubie-math (frozen/absorbed source-mirror), so its counts predate the cubie-tf monorepo snapshot; treat as historical-revision provenance, not current cubie-tf state.

### Fact-check corrections applied (2026-07-04)

- CI description corrected: both proof-kernel-check.yml and proof-kernel-verification.yml run `verus --crate-type lib` on the spec files STANDALONE; proof-kernel-verification.yml is a hard gate (continue-on-error:false) on the narrow CUB-1940..1966 TEP band, not report-only over verus/**. The conclusion (corpus is standalone / does not verify the runtime crates) is unchanged.
- Clarified that the self-scoping quote and snapshot SHA-256 come from the PUBLISHED library MANIFEST (verified live: line 48 / lines 28,57,61), not a repo-tracked file — a repo-tree grep will not find them.

---

<a id="tep-far0"></a>

## Cubie-native TEP fault detector: 100/100/100 FDR at d00 FAR=0.000% on the closed-loop-masked trio (CUB-1921 CUSUM OR-gate)

**Status:** 🟡 Scoped
**Scope boundary:** In-sample calibration identity: PASS-B sets h = 1.5x the observed d00 peak, so FAR = 0 by construction (CUB-1952). The 100/100/100 FDR is in-sample on the Braatz/Rieth d00/d03/d09/d15 set; held-out out-of-sample transfer is NOT yet established (tep_oos_results.md = VERIFIED_BAD).

### Claim

The cubie-tep detector (a no_std Rust crate) achieves FDR IDV-3 = 100.00%, FDR IDV-9 = 100.00%, FDR IDV-15 = 100.00% at d00 fault-free FAR = 0.000% (0/960 false alarms over 960 d00 samples) on the Tennessee Eastman Process (TEP, teP = Tennessee Eastman, NOT TSP) closed-loop-masked trio. The result comes from CUB-1921, a Page (1954) CUSUM aggregator OR-gated with the pre-existing CUB-1832 MetaCube binomial-bounce aggregator; CUSUM catches the slow-onset drift regime (e.g. IDV-15 condenser CW valve stick, ~85 sub-threshold samples post-injection) that the binomial bound misses by construction. Per-cell CUSUM parameters k and h are derived from the d00 fault-free baseline via two-pass calibration (no hardcoded constants). SCOPING (mirrored from both the whitepaper and the repo audit docs, load-bearing for accuracy): the FAR=0 boundary is a definitional CALIBRATION IDENTITY (CUB-1952), not statistical inference — PASS B sets h[c] = 1.5 x max_t s_t[c] observed on d00, so P(CUSUM fires on d00) = 0 by construction. The FDR figures are IN-SAMPLE on the Braatz/Russell-Chiang/Rieth d00/d03/d09/d15 set; held-out out-of-sample transfer (Rieth-2017 FaultFreeTest, DOI 10.7910/DVN/6C3JR1) is explicitly NOT YET evaluated (docs/audit/tep_oos_results.md marks the current default-config OOS proxy run VERIFIED_BAD). The peak strictly dominates PCA-T2 (IDV-3 6%, IDV-9 3%, IDV-15 10%) on all three faults. Reproduced across 60 runs (20 starting-layout seeds x 3 search algorithms: greedy hill-climb, Pareto-frontier, simulated annealing) at a 1000-iteration budget; 60/60 converge to the saturation peak (re-validated 2026-05-27). Triple-kernel formal backing: CUB-1950..CUB-1959 (Coq+Lean4+Verus), notably CUB-1952 (calibration identity), CUB-1953 (OR-gate completeness), CUB-1954 (V3 layout dominance).

### Direct downloads (public — lib.trustfortress.ai)

- [⬇ Public source pack — cubie-tep-source.zip (git-archive of the TEP detector crate: src/cusum.rs Page-CUSUM, two-pass calibration, OR-gate, src/bin/tep_detect.rs + tep_layout_search.rs). Enables source inspection without repo access; end-to-end build/run requires placing the pack in a matching cubie-tf workspace checkout. Bytes and regeneration notes are tracked in docs/resource-library/2026-07-04-source-packs.md.](https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2Fcubie-tep-source.zip) — `sha256 fbdd757730dbcd9f883849ac79b7a9a67601ed0cee574d9800b810f88d140394`
- [⬇ V6 TEP whitepaper PDF — exact claim + method + scoping (title: 'Cubie-Native No-Training Tennessee Eastman Fault Detection at Zero False Alarms'; author Nick Venezia / Centillion.AI; dated 2026-05-26). States FDR IDV-3/9/15 = 100.00% at d00 FAR = 0.000%, calibration-identity framing, 60/60 saturation, strict PCA-T2 dominance.](https://lib.trustfortress.ai/objects/cubie-tf%2Fv6-whitepaper-evidence%2F2026-07-03%2Fv6_cubie_tep_whitepaper.pdf) — `sha256 8569EC5FF26A77E3B55591E11FF8A75DB26EBEB82B523E94F376696AC60B9B05`
- [⬇ TEP work evidence pack (library) — publishes the raw Rieth-2017 RData files (>1GB compressed) that are gitignored in the repo, plus cubie-tep crate, layouts, baseline stats, and benchmark/Opus-4.8 evidence. Manifest key cubie-tf/tep-work/2026-07-03/MANIFEST.md](https://lib.trustfortress.ai/objects/cubie-tf%2Ftep-work%2F2026-07-03%2FMANIFEST.md)

### Source evidence (pinned)

| Artifact | Source | sha256 |
|---|---|---|
| cubie-tep crate lib.rs — TEP fault detector crate root (STICKER layer over cubie-core; IDV-3/9/15 closed-loop-masked trio; Q16.16 f64-firewall; all algorithms executable, no stubs) | [🔒 cubie-tf/cubie-tep/src/lib.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/cubie-tep/src/lib.rs)@`88ef7e83` |  |
| cubie-tep/src/cusum.rs — CUB-1921 Page(1954) CUSUM aggregator: s_t=max(0,s_{t-1}+(\|z\|-k)); calibrate_two_pass (PASS A k[c]=E\|z\|+0.25 sigma, PASS B h[c]=1.5 x observed d00 peak => FAR=0 by construction); OR-gate any_fired() | [🔒 cubie-tf/cubie-tep/src/cusum.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/cubie-tep/src/cusum.rs)@`88ef7e83` |  |
| docs/audit/empirical_peak.md — authoritative peak record: 100.00/100.00/100.00 @ d00 FAR=0.000% (0/960), commit 721394f, CUB-1921 CUSUM OR-gate; reproduced seeds {2024,5,73}; explicit in-sample calibration-identity scope disclosure; PCA beat factors 16.67x/33.33x/10.00x; full history table | [🔒 cubie-tf/docs/audit/empirical_peak.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/docs/audit/empirical_peak.md)@`88ef7e83` |  |
| docs/audit/tep_recovery_status.md — closure of the 'perfect FAR=0% + three perfect FDRs' debt items; 100/100/100 across 3 V3-derived layouts; Pareto-improvement over prior 717e689 peak (41.12/31.62/29.12 @ FAR=0%) | [🔒 cubie-tf/docs/audit/tep_recovery_status.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/docs/audit/tep_recovery_status.md)@`88ef7e83` |  |
| docs/audit/tep_oos_results.md — HONEST OOS boundary: status VERIFIED_BAD; default-config held-out proxy on d03/d09/d15 reports 0.00% FDR proxy; states the 100/100/100 headline 'remains an in-sample CUSUM calibration identity, not a held-out product claim' | [🔒 cubie-tf/docs/audit/tep_oos_results.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/docs/audit/tep_oos_results.md)@`88ef7e83` |  |
| cubie-tep/src/bin/tep_detect.rs — CLI detector over Rieth-2017 CSV schema (55 cols; fault inject sample 161); emits per-fault FDR proxy, FAR proxy, p99 detection latency | [🔒 cubie-tf/cubie-tep/src/bin/tep_detect.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/cubie-tep/src/bin/tep_detect.rs)@`88ef7e83` |  |
| cubie-tep/src/bin/tep_layout_search.rs — layout-search harness that reproduces the peak (--max-iters, --seed, --cusum default-on, --algorithm pareto) | [🔒 cubie-tf/cubie-tep/src/bin/tep_layout_search.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/cubie-tep/src/bin/tep_layout_search.rs)@`88ef7e83` |  |
| data/tep/layouts/best_found_by_search.json — search-discovered sticker permutation (seed 1729) used for the peak | [🔒 cubie-tf/data/tep/layouts/best_found_by_search.json](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/data/tep/layouts/best_found_by_search.json)@`88ef7e83` |  |
| data/tep/layouts/_perfect_3wins_seed2024_cusum.json — one of the three reproduced perfect-3 layout artifacts (seed 2024) | [🔒 cubie-tf/data/tep/layouts/_perfect_3wins_seed2024_cusum.json](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/data/tep/layouts/_perfect_3wins_seed2024_cusum.json)@`88ef7e83` |  |
| data/tep/README.md — dataset provenance (Russell/Chiang/Braatz d00/d03/d09/d15 from camaramm/tennessee-eastman-profBraatz; .dat files gitignored); local reproduction recipe. NOTE: the --use-braatz-baseline path alone underperforms PCA; the 100/100/100 peak requires the search layout + CUB-1921 CUSUM | [🔒 cubie-tf/data/tep/README.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/data/tep/README.md)@`88ef7e83` |  |
| proofs/coq/tep/CUB_1952_CalibrationIdentity.v — Coq proof that FAR=0 on the calibration set is a definitional identity of h=alpha*max_t s_t with alpha>1 | [🔒 cubie-tf/proofs/coq/tep/CUB_1952_CalibrationIdentity.v](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/proofs/coq/tep/CUB_1952_CalibrationIdentity.v)@`88ef7e83` |  |
| proofs/lean/tep/CUB_1952_CalibrationIdentity.lean — Lean4 kernel of the CUB-1952 calibration identity | [🔒 cubie-tf/proofs/lean/tep/CUB_1952_CalibrationIdentity.lean](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/proofs/lean/tep/CUB_1952_CalibrationIdentity.lean)@`88ef7e83` |  |
| proofs/verus/tep/CUB_1952_calibration_identity_spec.rs — Verus spec of the CUB-1952 calibration identity | [🔒 cubie-tf/proofs/verus/tep/CUB_1952_calibration_identity_spec.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/proofs/verus/tep/CUB_1952_calibration_identity_spec.rs)@`88ef7e83` |  |
| proofs/coq/tep/CUB_1953_OrGateCompleteness.v — Coq proof: OR-gate completeness (CUSUM OR binomial catches both fast and slow regimes) | [🔒 cubie-tf/proofs/coq/tep/CUB_1953_OrGateCompleteness.v](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/proofs/coq/tep/CUB_1953_OrGateCompleteness.v)@`88ef7e83` |  |
| proofs/coq/tep/CUB_1954_V3LayoutDominance.v — Coq proof: V3/search layout dominance | [🔒 cubie-tf/proofs/coq/tep/CUB_1954_V3LayoutDominance.v](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/proofs/coq/tep/CUB_1954_V3LayoutDominance.v)@`88ef7e83` |  |
| coq/CUB_1832_TepBelnapBaselineThreshold.v + coq/CubieCusumAggregator.v — the binomial-bounce (CUB-1832) and CUSUM aggregator kernels the OR-gate combines | [🔒 cubie-tf/coq/CubieCusumAggregator.v](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/coq/CubieCusumAggregator.v)@`88ef7e83` |  |

> 🔒 Source links require **cubie-tf / cubie-research** repository access (both proprietary). Public downloads above are on lib.trustfortress.ai.

### Reproduce this result

```bash
# Public source inspection: download the detector crate source pack.
curl -sL -o cubie-tep-source.zip 'https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2Fcubie-tep-source.zip'   # hashes in https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2FMANIFEST.md
# This partial crate pack is not standalone-buildable: unzip into a matching cubie-tf checkout to build (`cargo run -p cubie-tep --bin tep_detect`), or read cubie-tep/src/cusum.rs directly for the CUSUM + calibration-identity logic.
# 1. Get the dataset (gitignored in repo; published in the library tep-work pack). Repo path: data/tep/README.md documents source camaramm/tennessee-eastman-profBraatz (d00/d00_te/d03_te/d09_te/d15_te .dat).
git clone https://github.com/iamdatanick/cubie-tf.git && cd cubie-tf
# 2. Build the detector + search harness (release):
cargo build --release --package cubie-tep --bin tep_detect --bin tep_layout_search
# 3. Reproduce the search-discovered peak layout (seed 1729, 100 iters, CUSUM default-on):
python3 tools/tep_layout_search.py --max-iters 100 --seed 1729   # emits data/tep/layouts/best_found_by_search.json
# (Rust harness equivalent: ./target/release/tep_layout_search --seed 1729 --max-iters 100)
# 4. Run the detector per fault (Rieth convention: fault injected at sample 161):
./target/release/tep_detect --csv data/tep/d00_test.csv --summary-only                              # expect d00 FAR = 0.000% (0/960)
./target/release/tep_detect --csv data/tep/d03_test.csv --fault-inject-sample 161 --summary-only    # expect IDV-3 FDR = 100.00%
./target/release/tep_detect --csv data/tep/d09_test.csv --fault-inject-sample 161 --summary-only    # expect IDV-9 FDR = 100.00% (canonical window, samples 161+)
./target/release/tep_detect --csv data/tep/d15_test.csv --fault-inject-sample 161 --summary-only    # expect IDV-15 FDR = 100.00%
# 5. Reproduce 60/60 saturation (whitepaper): run across 20 seeds x 3 algorithms (greedy/pareto/simanneal) at 1000-iter budget via tep_layout_search --algorithm {greedy,pareto,simanneal}.
# 6. Verify triple-kernel proofs of the calibration identity (per docker/ toolchains): coqchk on proofs/coq/tep/CUB_1952_CalibrationIdentity.v ; lean on proofs/lean/tep/CUB_1952_CalibrationIdentity.lean ; verus --crate-type lib on proofs/verus/tep/CUB_1952_calibration_identity_spec.rs.
# 7. (Honest boundary) OOS check: python tools/tep_oos_validation.py --dataset-dir data/tep --out docs/audit/tep_oos_results.md  # default config returns 0.00% FDR proxy => confirms in-sample-only scope.
```

### Verification checks

- Whitepaper PDF SHA-256 = 8569EC5FF26A77E3B55591E11FF8A75DB26EBEB82B523E94F376696AC60B9B05 (non-watermarked edition; replaced the watermarked 900a623a copy on 2026-07-05, same 22-page content).
- Whitepaper body (extracted): 'the final detector achieves FDR_IDV-3 = 100.00%, FDR_IDV-9 = 100.00%, FDR_IDV-15 = 100.00% at d00 FAR = 0.000%' and 'the zero-FAR boundary is not statistical inference but a calibration identity' and 'All 20x3 = 60 runs converge to the saturation peak ... 60/60 perfect-3'.
- empirical_peak.md history table row 2026-05-26 commit 721394f: 100.00%/100.00%/100.00%, 3 wins, d00 FAR=0.000% (0/960 by h-calibration construction).
- cusum.rs calibrate_two_pass: PASS B computes h[c] = (peaks[c] as f64 * 1.5) => guarantees P(fire on d00)=0 (calibration identity, matches CUB-1952).
- All 18 cited repo paths confirmed at evidence commit 88ef7e8303ed82b4e0af398960be172f8bb560db via gh api repos/iamdatanick/cubie-tf/contents/<path>?ref=88ef7e8303ed82b4e0af398960be172f8bb560db (each returned a blob sha).
- Triple-kernel suite CUB-1950..CUB-1959 present in proofs/{coq,lean,verus}/tep/ (git ls-files); CUB-1952/1953/1954 are the calibration-identity / OR-gate-completeness / layout-dominance proofs.
- OOS honesty gate: docs/audit/tep_oos_results.md status VERIFIED_BAD, 0.00% FDR proxy at default config, explicitly labels the headline 'in-sample CUSUM calibration identity, not a held-out product claim'.

### Known boundaries / open gaps

- Held-out out-of-sample generalization is NOT established. docs/audit/tep_oos_results.md (status VERIFIED_BAD) shows the current default-config detector reports 0.00% FDR proxy on a held-out proxy; a clean transfer to the Rieth-2017 FaultFreeTest set (Harvard Dataverse DOI 10.7910/DVN/6C3JR1) has not been run. The 100/100/100 @ FAR=0 result is in-sample (calibration identity), and the whitepaper + audit docs both scope it that way. Checked: whitepaper (library), cubie-tf audit docs + code, data/tep. Not a missing-evidence gap — it is an explicit scope boundary of the claim.
- The raw Braatz/Rieth .dat/.RData datasets are gitignored in cubie-tf (only committed baseline_stats.json / layouts / MANIFEST.sha256). They ARE published in the library tep-work evidence pack (>1GB compressed, cubie-tf/tep-work/2026-07-03/), but that manifest does not enumerate per-file object keys, so exact per-dataset object URLs could not be captured without extracting the pack.
- Repo attribution nuance: the whitepaper author line reads github.com/iamdatanick/cubie-math and is dated 2026-05-26 (pre-monorepo). The live, current code + proofs are in iamdatanick/cubie-tf (cubie-math was frozen and absorbed into the monorepo on 2026-05-27). All blob URLs above point to the live cubie-tf/main location.
- cubie-research (iamdatanick/cubie-research) does NOT contain the TEP detector or dataset — it is a TSP (CUB-TSP) triple-kernel stress harness (crates/cubie-tsp-*, data/tsplib/, proofs/{coq,lean,verus}/CubTsp*). Confirmed via full recursive tree listing. Do not cite cubie-research for the TEP result.

---

<a id="agentic-waste"></a>

## TF Logs agentic-waste findings: GPU deadlock blast radius, VRAM lock, request-waste rate across ~45.3M real telemetry rows

**Status:** ✅ Verified
**Scope boundary:** Re-measured from the committed result JSONs (not the prior memory dossier). ~99.5% output<input waste is scoped to the Azure-2024 trace (44.1M of the ~45.3M rows); only the 66.85% blast-radius figure appears in a whitepaper.

### Claim

TrustFortress's "agentic waste" analysis over five real public datasets quantifies structural inefficiency in production LLM inference fleets. Re-measured from the committed analysis result files: (1) TP4 NVLink cascade deadlock freezes 66.85% of GPU fleet capacity during events (105,226 of 157,411 LLM GPU-intervals had SM duty==0; full_tp4_blast_radius.capacity_pct=66.8479, 140/141 containers); (2) 51.76% of all fleet VRAM is locked at any moment by the TP4 cascade (VRAM-GB-interval fraction (222,368.7+1,649,527.9)/3,616,626.1); (3) ~99.5% request waste = output<input, measured over the Azure-2024 trace only (43,872,851 of 44,107,694 requests = 99.47%). The total analyzed corpus re-summed from the result JSONs is ~45.29M distinct dataset rows (Alibaba 157,411 intervals + Azure-2023 28,185 + Azure-2024 44,107,694 + Azure-LMM-2025 1,000,000 + MaverIQ 1,684). SCOPING: Only the 66.85% figure appears in a whitepaper (V8 bare-metal .md, and its footnote marker [3] mis-points to the Rubik's-cube paper). The 51.76%, 99.5%, and row-total figures are backed by the JSON result files + INTC Evidence Pack, NOT by any whitepaper. The prior memory dossier's headline "44.9M rows / 99.5% of all 44M" is an approximation; the re-measured distinct-row total is ~45.3M, and 99.5% is scoped to Azure-2024 (44.1M of the 45.3M), not the whole corpus. The current live public pack at prefix cubie-tf/agentic-waste-evidence/2026-07-04 publishes the four headline JSONs needed for the 66.85% / 51.76% / ~99.5% reproduction; row-total support JSONs remain repo/private until the next owner-authorized publish run.

### Direct downloads (public — lib.trustfortress.ai)

- [⬇ Public reproduction pack — REPRODUCE.md with copy-paste curl+verify+re-derive commands for the 66.85% / 51.76% / ~99.5% figures (no repo access required), plus the intc MANIFEST mapping each number to its JSON path.](https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2FREPRODUCE.md)
- [⬇ intc-v1.0-tp4_nvlink_results.json — TP4 NVLink cascade: full_tp4_blast_radius {intervals:105226, capacity_pct:66.8479, containers:140}; vram_locked_by_tp4.pct_total_vram_locked:51.76; total_llm_intervals:157411. Source of the 66.85% and 51.76% headline numbers.](https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-tp4_nvlink_results.json) — `sha256 7d1a90bfc6036e7d6eb8483ba1ddb4bf69cc69003a306016368d221286119bc9`
- [⬇ intc-v1.0-hunter_results.json — DCGM threshold sweep + ITL backpressure: exact_zero_sm {count:105226, pct:66.8479}; itl_proxy_code.gen1_pct:15.9979 (2,688,232 single-token aborts); retry chains full_chains_ge3:127; lmm context_spikes_10x_median:93157, max effective ctx 28,375,569 tokens.](https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-hunter_results.json) — `sha256 6fb720db2600441e6a7155d4a2ac6ee8cdc023de65fceef016edc7742d979685`
- [⬇ intc-v1.0-azure_2024_unconstrained.json — Azure-2024 request-waste: code total_rows:16,803,695 output_lt_input pct:99.2115; conv total_rows:27,303,999 output_lt_input pct:99.6252. Combined 43,872,851/44,107,694 = 99.47% ≈ the ~99.5% waste headline.](https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-azure_2024_unconstrained.json) — `sha256 4a2a451e4e104bc543cce67aae42f8446513ea7160c7c32a3a6a880c2bfada06`
- [⬇ intc-v1.0-alibaba_v2026_squatter_metrics.json — original GPU-squatter run (pre-TP4-reinterpretation): chronic_squatter_rows:7553, containers:131, fleet_capacity_destroyed_pct:4.7983, wasted_kwh:44.0592. Shows the 4.8% original filter that only caught GPU 0.](https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-alibaba_v2026_squatter_metrics.json) — `sha256 2e4adf2dcd79233d603e1c68498434a469413e292b8843bca7bc5131844d43ad`
- [⬇ Library raw-data pack (LIVE, HTTP 200) — datacenter-crash-logs/2026-07-02 MANIFEST with real whole-file SHA-256s. Raw bundle TF Logs 2 LARGE.zip (410,827,647 bytes, sha256 3141ab66a0ffe8f855aa503670d8664c194733cf36f438942e1230af09c5b7b6) split into part001 (sha 1a399c21...) + part002 (sha 01d7faa5...); plus tf-intc-analysis-pack-20260527-trimmed.zip (263,520 bytes, sha ecceb2a2...) containing the analysis scripts.](https://lib.trustfortress.ai/objects/cubie-tf%2Fdatacenter-crash-logs%2F2026-07-02%2FMANIFEST.md)
- [⬇ Library raw-data object part001 (LIVE, HTTP 200), 209,715,200 bytes — first half of the canonical TF Logs 2 LARGE evidence archive (Alibaba GenAI GPU telemetry, Azure LLM 2023/2024 + LMM 2025 traces, MaverIQ CSVs, vLLM OOM/DCGM artifacts).](https://lib.trustfortress.ai/objects/cubie-tf%2Fdatacenter-crash-logs%2F2026-07-02%2Ftf-logs-2-large-20260412.zip.part001) — `sha256 1a399c21046176a99d1ddabfda9337e9bed4411dc704fc52159ef9c5b5771007`

### Source evidence (pinned)

| Artifact | Source | sha256 |
|---|---|---|
| intc-v1.0-MANIFEST.md — evidence-pack manifest. Lists per-file sha256 that MATCH the committed copies exactly (tp4=7d1a90bf…, hunter=6fb720db…, azure_2024=4a2a451e…, alibaba=2e4adf2d…); there is no upstream-vs-committed divergence. Maps each canonical dashboard number to its JSON path (66.85% -> full_tp4_blast_radius.capacity_pct, 51.76% -> vram_locked_by_tp4, 87.53% DCGM blind spot derived). | [🔒 cubie-tf/demo/evidence/intc-v1.0/intc-v1.0-MANIFEST.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/demo/evidence/intc-v1.0/intc-v1.0-MANIFEST.md)@`88ef7e83` |  |
| INTC_Evidence_Pack_v1.0.md — reproduce documentation: names analysis scripts scripts/tp4_nvlink_breakdown.py and hunter.py, the results/*.json outputs, and the causal chain (99.5% output<input -> KV-cache pin -> 66.85% fleet frozen). Confirms scripts live in the raw bundle, not the git repo. | [🔒 cubie-tf/intel-submissions/cubie-intel-submission/intel-submission/evidence/INTC_Evidence_Pack_v1.0.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/intel-submissions/cubie-intel-submission/intel-submission/evidence/INTC_Evidence_Pack_v1.0.md)@`88ef7e83` |  |
| CUBIE_MASTER_WHITE_PAPER_V8_0_BARE_METAL.md — the ONLY whitepaper that states an agentic-waste number: 'freezing up to 66.85% of total data center fleet capacity without triggering hypervisor alerts [3]'. Caveat: footnote [3] mis-points to Rokicki 2014 (Rubik's cube diameter), a broken citation. Does NOT mention 51.76%, 99.5%, or the row total. | [🔒 cubie-tf/docs/CUBIE_MASTER_WHITE_PAPER_V8_0_BARE_METAL.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/docs/CUBIE_MASTER_WHITE_PAPER_V8_0_BARE_METAL.md)@`88ef7e83` |  |

> 🔒 Source links require **cubie-tf / cubie-research** repository access (both proprietary). Public downloads above are on lib.trustfortress.ai.

### Reproduce this result

```bash
# Public reproduction (no repo access). Full guide: https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2FREPRODUCE.md
curl -sL -o tp4.json 'https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-tp4_nvlink_results.json'
curl -sL -o az2024.json 'https://lib.trustfortress.ai/objects/cubie-tf%2Fagentic-waste-evidence%2F2026-07-04%2Fintc-v1.0-azure_2024_unconstrained.json'
sha256sum tp4.json az2024.json   # tp4 7d1a90bfc6036e7d6eb8483ba1ddb4bf69cc69003a306016368d221286119bc9 ; azure_2024 4a2a451e4e104bc543cce67aae42f8446513ea7160c7c32a3a6a880c2bfada06
```

- Reproduce the 66.85% and 51.76% headlines directly from the JSON: python3 -c "import json;d=json.load(open('tp4.json'));print(d['full_tp4_blast_radius'])"  -> EXPECT {'intervals':105226,'capacity_pct':66.8479,'containers':140,...} ; and d['vram_locked_by_tp4']['pct_total_vram_locked'] -> EXPECT 51.76 (=(222368.7+1649527.9)/3616626.1).
- Reproduce the ~99.5% request-waste figure: fetch intc-v1.0-azure_2024_unconstrained.json, then python3 -c "import json;d=json.load(open('az2024.json'));o=d['code']['output_lt_input']['count']+d['conv']['output_lt_input']['count'];t=d['code']['total_rows']+d['conv']['total_rows'];print(o,t,round(100*o/t,4))"  -> EXPECT 43872851 44107694 99.4676 (i.e. 99.5% output<input, scoped to Azure-2024).
- Repo/private row-total self-check until the row-total support JSONs are uploaded: from demo/evidence/intc-v1.0, Alibaba 157411 + Azure2023 (8819+19366) + Azure2024 (16803695+27303999) + LMM2025 1000000 + MaverIQ 1684 = 45,294,974 distinct rows (~45.3M). tools/publish_reproduce_data.ps1 now includes the Azure-2023, LMM-2025, and MaverIQ support JSONs for the next public pack.
- Regenerate the JSONs from raw data (full reproduce): download the raw bundle from the LIVE library pack — curl -O https://lib.trustfortress.ai/objects/cubie-tf%2Fdatacenter-crash-logs%2F2026-07-02%2Ftf-logs-2-large-20260412.zip.part001 (and .part002), reassemble ('copy /b name.part001+name.part002 name.zip'), verify sha256 == 3141ab66a0ffe8f855aa503670d8664c194733cf36f438942e1230af09c5b7b6 ; unzip; then run the analysis scripts (scripts/tp4_nvlink_breakdown.py -> results/tp4_nvlink_results.json ; hunter.py -> results/hunter_results.json) which ship inside tf-intc-analysis-pack-20260527-trimmed.zip (sha256 ecceb2a2e201b1a8871a4acb558717dcdafffe285a41c1a928ea7e618f221114) from the same library pack, per INTC_Evidence_Pack_v1.0.md.

### Verification checks

- JSON headline metrics reproduce exactly: full_tp4_blast_radius.capacity_pct=66.8479 (105226/157411 zero-SM intervals), vram_locked_by_tp4.pct_total_vram_locked=51.76, Azure-2024 output<input = 43872851/44107694 = 99.4676%. Confirmed by base64-decoding the GH-API-served file content and re-summing.
- Whitepaper anchoring confirmed by grep: ONLY 66.85% appears (V8 bare-metal .md), and its footnote [3] mis-points to Rokicki 2014 (Rubik's cube). Greps for 51.76|99.5|45.3|44.9|VRAM.lock returned zero hits — those three are result-JSON-only.
- Library object availability confirmed by HTTP status: datacenter-crash-logs/2026-07-02 MANIFEST + part001 return 200 with the manifest's SHA-256 table; tf-logs-evidence/2026-07-04 MANIFEST returns 404 (object store empty — it is a locator doc, not a populated pack).
- Committed-copy hash self-consistency confirmed 2026-07-05: the four registry download sha256 values match the committed demo/evidence/intc-v1.0 JSON files and intc-v1.0-MANIFEST.md exactly (tp4=7d1a90bf..., hunter=6fb720db..., azure_2024=4a2a451e..., alibaba=2e4adf2d...).
- Live public-link HEAD verification on 2026-07-06 returned HTTP 200 for REPRODUCE.md, MANIFEST.md, and the four headline JSON objects under cubie-tf/agentic-waste-evidence/2026-07-04/. The Azure-2023, LMM-2025, and MaverIQ row-total support object URLs returned HTTP 404, so they are intentionally not listed as current public downloads.

### Known boundaries / open gaps

- Analysis scripts (scripts/tp4_nvlink_breakdown.py, hunter.py) are NOT committed to the GitHub repo — they exist only inside the raw bundle (tf-intc-analysis-pack-20260527-trimmed.zip on the library, sha ecceb2a2..., per INTC_Evidence_Pack_v1.0.md). A reader can re-hash and re-read the precomputed JSONs from GitHub, but must download the library zip to regenerate them from raw data. Not unfindable (the script names + container zip are documented), but they lack a direct GitHub blob URL.
- The library pack referenced by the repo locator doc docs/resource-library/2026-07-04_tf-logs-evidence.md — prefix cubie-tf/tf-logs-evidence/2026-07-04/ — is a locator only: its object store is EMPTY (MANIFEST returns HTTP 404; api/list returns []). The live, populated raw-data pack is cubie-tf/datacenter-crash-logs/2026-07-02/ (HTTP 200). Referenced-but-not-yet-populated, not a true evidence gap. Checked: library api/list, api/search, direct /objects HEAD.
- Number-reconciliation note (accuracy, not a gap): the prior memory dossier states 44.9M rows and 43,666,866 Azure-2024 rows; the committed result JSONs give Azure-2024 = 44,107,694 and a 5-dataset distinct-row total of 45,294,974 (~45.3M). Registry uses the re-measured JSON figures. Also the dossier/whitepaper phrase '99.5% of all 44M' is scoped in the data to Azure-2024 (44.1M) specifically, not the whole 45.3M corpus.
- Current public-pack boundary: the four headline JSONs are live for no-repo reproduction, but the row-total support JSONs (Azure-2023, LMM-2025, MaverIQ) are not yet uploaded in the 2026-07-04 prefix. The publisher now includes them for the next authorized upload.
- Hash provenance is now aligned for the four committed result JSONs: cite the registry download rows or demo/evidence/intc-v1.0/intc-v1.0-MANIFEST.md for SHA-256 verification of the GitHub copies.

### Fact-check corrections applied (2026-07-04)

- Replaced all four evidence provenance sha256 values with the actual committed hashes, re-measured live 2026-07-04 (tp4=7d1a90bf…, hunter=6fb720db…, azure_2024=4a2a451e…, alibaba=2e4adf2d…). The prior values matched no artifact in the repo or its history.
- Removed the phantom 'MANIFEST upstream hashes differ from committed copies' narrative — the MANIFEST lists the same committed-copy hashes; there is no divergence.
- Completed the truncated INTC Evidence Pack URL to …/evidence/INTC_Evidence_Pack_v1.0.md.

---

<a id="eu-ai-act"></a>

## EU AI Act coverage (Articles 9/10/11/13/14/15 + Annex IV) via cubie-tf crates/cubie-eu-*

**Status:** 🟡 Scoped
**Scope boundary:** A compiling Rust evidence-mapping + Annex IV generation SCAFFOLD (per-article structs + AnnexIvBuilder + /api/v1/annex-iv), NOT a fully populated conformity pack — modules carry placeholder values (e.g. kill_switch_url=example.com, some proof counts hardcoded 0).

### Claim

cubie-tf implements EU AI Act coverage as a Rust evidence-mapping + Annex IV generation scaffold under `crates/cubie-eu-*`, plus a separate documented compliance surface under `apps/trustfortress/docs/eu-ai-act-compliance/`. IMPLEMENTED (Rust, verified to compile + tests pass): `cubie-eu-articles` provides per-article Serde structs that map cubie-core runtime facts to EU AI Act evidence types — one module each for Art 9 (risk register from 17 cubie-core denial codes), Art 10 (data governance + DataProvenanceTag), Art 13 (transparency), Art 14 (human-oversight escalation hooks), Art 15 (accuracy/robustness → ISO 24029-2), Art 16+17 (QMS), Art 19 (registration record), Art 50 (transparency), Art 73 (incident reporting, typed 15-day deadline), plus a NIST RMF module. Article 11 (technical documentation) is realized by the separate `cubie-eu-annex-iv` crate — the `AnnexIv` struct + `AnnexIvBuilder` + JSON Schema ARE the Annex IV technical documentation Art 11 requires; Article 12 (record-keeping) maps to the builder's `with_article_12`→`monitoring` field and to `cubie-eu-ledger`. The `cubie-eu-api` crate exposes the Annex IV generator as `GET /api/v1/annex-iv` (axum, binds 0.0.0.0:8443). ACCURACY SCOPING: (1) These crates are a mapping/generation SCAFFOLD, not a fully populated conformity pack — they carry placeholder values (`kill_switch_url`=example.com, `incident_reporting_email`="TO_BE_CONFIGURED", Art 15 lean/coq/verus proof counts hardcoded to 0, `single_bit_flip_closure_verified`/`constant_time_validator_verified` hardcoded `true`). (2) Only 2 of the 9 article modules have Rust unit tests (article_09, article_73); the annex-iv crate has 2 builder tests — 4 EU tests total, all passing. (3) The V8/V9 master whitepaper's regulatory-mapping chapter (ch14_regulatory_mapping.tex) scopes the mapping MORE NARROWLY than the crate set: it maps only bare-metal/→Art 9, cubie-platform/→Art 15, cubie-eu-ledger/→Art 12 & 13, verus/consent_spec/→Art 14. (4) The richer per-article coverage with high test counts (Art 5:29, Art 6:88, Art 72:53, Art 73:85, Art 86:65, Art 14:24) and the 98/100 readiness score live in the DOCUMENTED surface `apps/trustfortress/docs/eu-ai-act-compliance/` and refer to TypeScript modules in the trustfortress app — NOT the Rust crates; that TS is marked deprecated for new logic per apps/trustfortress/CLAUDE.md (2026-07-01 boundary). The crate-description discrepancy (cubie-eu-articles Cargo.toml lists Art 9/10/11/12/13/14/15/16/17/19/50/73 but has no article_11.rs or article_12.rs) is explained by Art 11=annex-iv crate and Art 12=builder monitoring field, NOT a gap.

### Direct downloads (public — lib.trustfortress.ai)

- [⬇ Public source pack — cubie-eu-source.zip (git-archive of cubie-eu-articles + cubie-eu-annex-iv + cubie-eu-api: per-article evidence-mapping structs, AnnexIvBuilder = Article 11 tech documentation, /api/v1/annex-iv generator). Enables source inspection without repo access; build/test requires placing the pack in a matching cubie-tf workspace checkout. Bytes and regeneration notes are tracked in docs/resource-library/2026-07-04-source-packs.md.](https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2Fcubie-eu-source.zip) — `sha256 acc5178bfc39ba6677a5c9efb40e71c52426f4777e77a0985423f75544358688`

### Source evidence (pinned)

| Artifact | Source | sha256 |
|---|---|---|
| cubie-eu-articles crate — per-article EU AI Act evidence-mapping structs (Art 9/10/13/14/15/16-17/19/50/73 + NIST RMF); lib.rs declares 9 article modules (no article_11/article_12 files by design) | [🔒 cubie-tf/crates/cubie-eu-articles/src/lib.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/lib.rs)@`88ef7e83` |  |
| Article 9 risk management — build_from_denial_cert() maps 17 cubie-core deny codes to Risk structs; test risk_register_has_17_deny_codes passes | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_09.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_09.rs)@`88ef7e83` |  |
| Article 10 data governance — DataGovernance + DataProvenanceTag (RTMR[2] anchor) | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_10.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_10.rs)@`88ef7e83` |  |
| Article 13 transparency — Transparency struct wiring denial_cert + dual-ledger receipt (no unit test in module) | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_13.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_13.rs)@`88ef7e83` |  |
| Article 14 human oversight — HumanOversight escalation hooks (placeholder kill_switch_url=example.com; no unit test in module) | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_14.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_14.rs)@`88ef7e83` |  |
| Article 15 accuracy/robustness — AccuracyRobustness → ISO 24029-2; lean/coq/verus proof counts hardcoded 0, *_verified flags hardcoded true (no unit test in module) | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_15.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_15.rs)@`88ef7e83` |  |
| Article 73 incident reporting — typed INCIDENT_REPORT_DEADLINE_DAYS=15 const; test deadline_is_exactly_15_days_after_detection passes | [🔒 cubie-tf/crates/cubie-eu-articles/src/article_73.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-articles/src/article_73.rs)@`88ef7e83` |  |
| cubie-eu-annex-iv builder — AnnexIv struct + AnnexIvBuilder (= Article 11 technical documentation); with_article_12→monitoring, with_article_13→transparency; 2 tests pass | [🔒 cubie-tf/crates/cubie-eu-annex-iv/src/builder.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-annex-iv/src/builder.rs)@`88ef7e83` |  |
| cubie-eu-annex-iv JSON Schema (draft 2020-12) for Annex IV technical documentation | [🔒 cubie-tf/crates/cubie-eu-annex-iv/src/schema.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-annex-iv/src/schema.rs)@`88ef7e83` |  |
| cubie-eu-api — Annex IV generator route GET /api/v1/annex-iv (annex_iv handler builds AnnexIv doc, returns JSON); crate builds clean (exit 0) | [🔒 cubie-tf/crates/cubie-eu-api/src/lib.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-api/src/lib.rs)@`88ef7e83` |  |
| cubie-eu-api main — axum server binds 0.0.0.0:8443 (Annex IV generator entrypoint) | [🔒 cubie-tf/crates/cubie-eu-api/src/main.rs](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/crates/cubie-eu-api/src/main.rs)@`88ef7e83` |  |
| Whitepaper regulatory-mapping chapter (V8/V9 master) — authoritative narrower scoping: bare-metal/→Art9, cubie-platform/→Art15, cubie-eu-ledger/→Art12&13, verus/consent_spec/→Art14 | [🔒 cubie-tf/docs/whitepaper/chapters/ch14_regulatory_mapping.tex](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/docs/whitepaper/chapters/ch14_regulatory_mapping.tex)@`88ef7e83` |  |
| DOCUMENTED surface (separate from Rust crates) — EU AI Act compliance readiness scorecard 98/100, per-article RED/YELLOW/GREEN; cites TypeScript modules (human-override.ts etc.), deprecated for new logic per 2026-07-01 boundary | [🔒 cubie-tf/apps/trustfortress/docs/eu-ai-act-compliance/evidence/compliance-readiness-scorecard.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/apps/trustfortress/docs/eu-ai-act-compliance/evidence/compliance-readiness-scorecard.md)@`88ef7e83` |  |
| DOCUMENTED surface — existing-compliance-evidence.md: file/line evidence matrix mapping TS components to Art 5/10/12/14/15/50/72/73/86 | [🔒 cubie-tf/apps/trustfortress/docs/eu-ai-act-compliance/evidence/existing-compliance-evidence.md](https://github.com/iamdatanick/cubie-tf/blob/88ef7e8303ed82b4e0af398960be172f8bb560db/apps/trustfortress/docs/eu-ai-act-compliance/evidence/existing-compliance-evidence.md)@`88ef7e83` |  |
| Resource library search for a published Annex IV artifact: https://lib.trustfortress.ai/api/search?q=annex-iv returned [] (no published Annex IV document); ?q=eu%20annex returned only docs/resource-library/CLOUDFLARE_KNOWLEDGEBASE_PLAN.md | [link](https://lib.trustfortress.ai/api/search?q=annex-iv) |  |

> 🔒 Source links require **cubie-tf / cubie-research** repository access (both proprietary). Public downloads above are on lib.trustfortress.ai.

### Reproduce this result

```bash
# Public source inspection: download the EU crates source pack.
curl -sL -o cubie-eu-source.zip 'https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2Fcubie-eu-source.zip'   # hashes in https://lib.trustfortress.ai/objects/cubie-tf%2Fsource-packs%2F2026-07-04%2FMANIFEST.md
# This partial workspace pack is not standalone-buildable: unzip into a matching cubie-tf checkout to build/test (`cargo test -p cubie-eu-annex-iv`), or read crates/cubie-eu-articles/src/*.rs for the article-to-runtime mappings.
```

- Clone + enter repo: git clone https://github.com/iamdatanick/cubie-tf && cd cubie-tf (local: C:\Users\NickV\cubie-tf)
- Run the EU crate tests: `cargo test -p cubie-eu-articles -p cubie-eu-annex-iv` — EXPECTED (verified 2026-07-04): cubie-eu-annex-iv 2 passed (build_fails_without_identifier, minimal_build_succeeds); cubie-eu-articles 2 passed (article_09::tests::risk_register_has_17_deny_codes, article_73::tests::deadline_is_exactly_15_days_after_detection); 0 failed
- Confirm the Annex IV generator compiles: `cargo build -p cubie-eu-api` — EXPECTED: `Finished dev profile` exit 0 (verified 2026-07-04, ~45s cold)
- Run the Annex IV generator (code-read of handler; not executed in this session): `cargo run -p cubie-eu-api` starts axum on 0.0.0.0:8443, then `curl http://localhost:8443/api/v1/annex-iv` — EXPECTED JSON body with system_identifier="CUBIE-EU-v0.1", provider.name="Centillion AI" (override via CUBIE_EU_PROVIDER_NAME), provider.country="US" (override via CUBIE_EU_PROVIDER_COUNTRY), intended_purpose="High-risk AI inference admission with TDX-rooted attestation", plus Annex IV fields (risk_management, data_governance, human_oversight, accuracy_robustness, monitoring, transparency_disclosures, logs, conformity_declaration)
- Confirm the crate-description vs file discrepancy is intentional: `rg -n "article_11|article_12" crates/cubie-eu-articles/src` returns no dedicated files; Art 11=annex-iv crate, Art 12=cubie-eu-annex-iv builder with_article_12→monitoring + cubie-eu-ledger

### Verification checks

- cargo test -p cubie-eu-articles -p cubie-eu-annex-iv  # 4 tests pass, 0 fail (verified)
- cargo build -p cubie-eu-api  # exit 0, generator endpoint compiles (verified)
- rg -c '#\[test\]' crates/cubie-eu-articles/src/*.rs  # only article_09.rs=1, article_73.rs=1 → 7 of 9 article modules have NO Rust test
- rg -n 'TO_BE_CONFIGURED|example.com|lean_proof_count: 0|_verified: true' crates/cubie-eu-articles/src  # confirms placeholder/scaffold values
- gh api repos/iamdatanick/cubie-tf/contents/crates/cubie-eu-articles/src/lib.rs?ref=88ef7e8303ed82b4e0af398960be172f8bb560db --jq .path  # confirms the evidence path resolves at the pinned commit

### Known boundaries / open gaps

- No published Annex IV output artifact in the resource library: `/api/search?q=annex-iv` returns [] and `?q=eu%20annex` returns only docs/resource-library/CLOUDFLARE_KNOWLEDGEBASE_PLAN.md — so no /objects/ library URL exists for a generated Annex IV document (checked: library /api/search). The generator exists in code (cubie-eu-api) but no rendered sample output is published.
- Per-article Rust test coverage is thin: 7 of 9 cubie-eu-articles modules (10,13,14,15,16-17,19,50) have NO unit test; only article_09 and article_73 are tested. The high per-article test counts (Art 6:88, Art 73:85, etc.) belong to the DOCUMENTED TypeScript surface, not the Rust crates.
- Scaffold placeholders unresolved in the Rust crates: Art 14 kill_switch_url=example.com, Art 16-17 incident_reporting_email="TO_BE_CONFIGURED", Art 15 lean/coq/verus proof counts hardcoded 0 and *_verified flags hardcoded true — these are static defaults, not populated from live cubie-core proof/config state.
- No formal proof (Verus/Coq/Lean CUB spec) binds the EU article mapping structs to cubie-core runtime behavior; the whitepaper references verus/consent_spec/ for Art 14 but the cubie-eu-articles structs themselves are unproven plain Rust. (Not searched exhaustively across the full 1800+ CUB corpus for an EU-specific spec.)

---

_Generated from `claims-data.json` by `tools/gen_claims_md.py` · fact-checked 2026-07-04 · live at [https://lib.trustfortress.ai/claims](https://lib.trustfortress.ai/claims)._