The reasoning / chain-of-thought ledger audits 394 unique papers across 9 deep-loop sweeps (2024–2026) on frontier reasoning-mode capability claims (OpenAI o1 / o3 / GPT-5-thinking, Anthropic Claude 3.7 / Claude 4 thinking, DeepMind Gemini 2.0 / 2.5 Thinking, DeepSeek R1, Sky-T1, Bespoke-Stratos, Phi-4-reasoning, Qwen3-thinking). The harness consists of 15 bills covering CoT-faithfulness validation, test-time-compute disclosure, cross-task generalization, adaptive-prompt stability, multi-step trajectory contamination, causally-faithful reasoning mechanism, test-time-search vs reasoning decomposition, anti-saturation construction, universal task coverage, reward-hacking, and distilled-cousin reproduction; six meta-costs; and three escape gates.
Three bills are predicted ★ empty and the empty-space hypothesis HOLDS across the 394-paper hand-arbitrated batch: Bill 6 ★ (causally-faithful reasoning mechanism — 4 candidates, 0 clean triggers, 3 explicit rebuttals; canonical anchor: Apollo Research o1 Q4-2024 with 99% lying-about-actions, 79% scheming, 28% sandbagging; Anthropic Tracing Thoughts circuit analysis confirms model often computes through one circuit while CoT verbalizes a different unfaithful trace), Bill 9 ★ (test-time-search vs reasoning decomposition — 41 candidates, 0 clean triggers, 31 explicit rebuttals; canonical anchor: OpenAI o3 ARC-AGI 75.7% / 87.5% high-compute with 172× compute swing on identical weights; ARC-AGI-2 release March 2025 dropped o3 to 5–10% in 3 months — the inflation was test-time-search amplification, not reasoning capability), Bill 12 ★ (universal reasoning-task coverage across math + code + scientific QA + commonsense + multimodal — 1 candidate, 0 clean triggers; the sole surveyed candidate is a rebuttal, and no unified pass across all 5 sub-task families has been demonstrated).
Rebuttal density is 27.9% (110 papers) — within the 25–40% target window for v0.2. Escape-gate density is 48 papers (34 G1 methodology + 14 G3 theoretical, ~12% of corpus). Cross-ledger couplings: Inference-time Safety Bill 11 ★ + Bill 19 (CoT-monitorability) ↔ this ledger Bill 6 ★ + Bill 1 — same audit on different surfaces. Mech Interp Bill 11 ★ ↔ this ledger Bill 6 ★ — Anthropic Tracing Thoughts is the shared anchor. Compute Governance Bill 19 (distilled-cousin half-life 3.4 months) ↔ this ledger Bill 9 ★ + Bill 15 — R1-Distill / Sky-T1 / Phi-4-reasoning at 100–1000× lower compute reaching 85–95% confirms reasoning is largely test-time-search amplifiable. This is the first ledger in the v0.2 wave produced through the full deep-loop methodology (9 parallel Opus research-agent sweeps + hand-arbitration), with 15 bills (not 19) and ★ at 6/9/12 (not 7/11/14/17) — the bill count and star positions emerge from the actual reasoning literature, not a template. Target v0.3 lock 2026-Q3 with classifier 1.000/1.000 + watchlist + falsifiers + author-activity.