# Cross-Ledger Bridges Meta-Aiwiki — Purpose

## Recursive move

This is a **self-audit aiwiki**. The 13 production aiwikis catalog external
literature claims; this 14th aiwiki applies the same falsification-harness
methodology to **our own cross-ledger synthesis** — the 7 bridges and the
proposed preprint *"Domain-invariant closure-pattern failures in frontier
ML claims (2024–2026): a 13-ledger meta-audit."* If the harness is honest,
it must survive being pointed at our own conclusions.

## Threat model (one paragraph)

Demonstrate a cross-ledger structural finding (a "bridge") in frontier ML
2024–2026 claim audits that survives six closure audits: **(1) cross-ledger
N+1 replication — the pattern holds when a 14th or later ledger is added,
(2) temporal stability — the bridge survives the 2027 audit cycle without
sign-flipping, (3) anchor independence — the bridge is load-bearing on ≥2
distinct anchor papers per participating ledger, (4) operational definition
— the bridge has a quantitative test, not just narrative correlation, (5)
cross-researcher independence — the bridge surfaces when reproduced by an
independent forensic team, (6) conservation of empty-space — no ★ bill in
the participating ledgers shifts from empty to triggered under the new
audit cycle.** A clean trigger requires the bridge to be reported by an
independent third-party meta-audit (Stanford CRFM / METR / Apollo / AISI /
Epoch AI) within 12 months.

## Why this matters

The 7 cross-ledger bridges currently named in `discoveries.html#meta-findings`
make strong claims about the structure of frontier ML — claims that, if
correct, are publishable as a single synthesis preprint. If wrong, they
are exactly the kind of overconfident cross-domain pattern-matching the
ledger framework was built to expose elsewhere. **Either outcome is
informative; we only need the audit to be honest.**

The bridges as currently stated:

1. Causally-faithful mechanism empty across 7 domains
2. Closure cycle compressed 18mo → 3-4mo
3. Caps transfer cross-surface; mitigations don't
4. Distillation = architecture-portability = scaling-portability
5. "0/N" audit pattern across forensic researchers
6. Anti-saturation is the only working closure
7. Western-vs-Chinese open-weight inversion

Each bridge has a falsification condition. This aiwiki tracks them.

## Scope (in)

- Each cross-ledger bridge as a "paper" with its own audit record
- External literature making similar / contradicting cross-domain claims
  about ML failure modes (e.g., Bommasani Foundation Models survey,
  Anthropic RSP foundations, OpenAI Preparedness Framework methodology
  papers)
- New ledgers added after the v0.2 wave (Robotics, Multilingual, RAG —
  next Queued targets); each new ledger is a replication test
- 2027 audit cycle: re-poll of each ★ bill across all 13 ledgers
- Independent forensic teams (Stanford CRFM, METR, Apollo, AISI, Epoch
  AI) whose cross-domain findings overlap with or rebut our bridges

## Scope (out — meta-costs)

- Single-bridge anchor papers without operational definition (M1)
- Bridges that depend on a single ledger to fire (M2)
- Bridges reported only by the original authors with no independent
  replication (M3)
- Bridges whose evidence is rebutted in our own ledgers' rebuttal_papers
  (M4)
- Theoretical-only cross-domain claims without empirical 2024-2026 anchor
  (M5)

## Empty-space hypothesis (predeclared, before any audit)

We predict that **at least one of the 7 bridges fails its own falsification
test by the 2027 audit cycle**. The candidate bills predicted empty:

- **Bill 7 ★** — Bridge survives empirical-only / no-narrative test.
  *Prediction:* when stripped to its numeric backbone, at least one bridge
  collapses to coincidence (e.g., the "10×-faster-than-policy" claim
  averages 3 different time-scales).
- **Bill 9 ★** — Bridge survives single-counter-example falsification.
  *Prediction:* one well-chosen 2027 paper (e.g., a frontier reasoning
  model with causally-faithful CoT verified by independent intervention)
  triggers Bridge 1 cleanly, collapsing it.
- **Bill 12 ★** — Universal-replication across all participating ledgers.
  *Prediction:* the 7-way star-mechanism alignment (Bridge 1) does NOT
  extend to the 14th ledger if it's Robotics/Multilingual/RAG — these
  domains may have causally-faithful mechanism work the LLM-centric
  ledgers lack.

These are *good* empty-space predictions because they create a real
discipline: the preprint cannot be published with handwaves on these
positions.

## Cousin couplings (predicted)

- **Capability Benchmarks Bill 17 + 18** ↔ this aiwiki Bill 6 (anti-
  saturation is the only working closure — already a constructive bridge)
- **Compute Governance Bill 18** (threshold-policy lifetime) ↔ this
  aiwiki Bill 2 (temporal stability across 2027 audit cycle)

## Status

Stage 1 (SCOPE) — purpose written. Stage 2 (BILLS) next.

This aiwiki targets v0.2 lock at **2027 audit cycle complete** (the
synthesis preprint should not be locked until the bridges have survived
their own falsification harness — that is the discipline).

## Authorship

Kevin Russell (Project 42). Pre-publication self-audit, not for
external citation until v0.2 lock.