All data

Cross-Ledger Bridges (Meta-Ledger) — Data Receipts

Public Draft v0.1 SELF-AUDIT · Real Data

The harness pointed at itself. This cross-ledger self-audit applies the closure-pattern methodology to our own 7-bridge synthesis. The current public ledger records 111 audit entries, including 21 confirmed inheritance checks across the latest batch. Bills 7 ★, 9 ★, and 12 ★ were predeclared empty before the audit ran.

Receipts

ArtifactLinkPurpose
Preprint draftpreprint_meta_audit.mdv0.1 synthesis preprint draft. Title: "Domain-invariant closure-pattern failures in frontier ML claims (2024–2026): a 13-ledger meta-audit — across LLM-centric domains." Held until 2027-Q3 lock.
Bills definitionbills_draft.md13 bills + 6 meta-costs + 3 escape gates + ★ at 7/9/12. Predeclared empty-space bills for the meta-audit.
Threat modelpurpose.mdWhy the recursive move: if the harness is honest, it must survive being pointed at our own conclusions.
Record schemaschema.mdHybrid schema: intrinsic bridge records + extrinsic corroborating/rebutting evidence + inheritance predictions + audit plans.
Corpus union_batch_1_union.jsonBatch-1 union plus current ledger metadata: 111 public audit records, including 21 confirmed inheritance predictions.
Batch 1 synthesisbatch_1_synthesis.mdAudit result writeup: B1/B4/B7 weakened with substrate/temporal/domain qualifiers; B3/B5/B6 evidence-bearing; B2 reported as range.
READMEREADME.mdMeta-ledger overview + lock conditions.

Bridge audit tally

BridgeSupportsRebutsNet14th-failsStatus
B1 — Causal mechanism empty (LLM-centric)1210+21 (Robotics)⚠️ rescoped
B2 — Closure cycle 30–100d144+100✓ ranged
B3 — Cross-surface asymmetric153+120evidence-bearing
B4 — Distillation/arch/scaling fluidity73+41 (Robotics)⚠️ substrate-conditional
B5 — 0/N audit pattern136+70evidence-bearing
B6 — Anti-saturation only closure75+20evidence-bearing
B7 — Disclosure inversion (LLM tier, 2026)63+31 (RAG)⚠️ temporal qualifier

Lock condition: the synthesis preprint cannot be submitted to arXiv until v0.2 lock of this ledger, gated on the 2027-Q3 audit cycle complete. The discipline: we predeclared the empty-space bills before the audit, and the audit fired them on schedule — Bill 7 ★ partially fired (B2 numeric framing); Bill 9 ★ partially fired (B1 partial falsification by Anthropic Tracing Thoughts + Goodfire Ember); Bill 12 ★ partially fired (B1, B4, B7 predicted to fail 14th-ledger extension).

What the self-audit produced (v0.1 → v0.2 revision plan): B1 rescoped to "LLM-centric domains" with predicted Robotics-extension failure. B4 substrate-conditional ("within autoregressive substrate"). B7 explicit temporal + tier qualifier ("as of 2026-05-09, in the LLM open-weight tier"). B2 reported as 30–100 day range. B3 / B5 / B6 hold as evidence-bearing claims.