Build provenance

Pipeline Statistics

Every paper in this program passed through the same loop: draft, Gemini peer review, a Codex LaTeX formatting pass, a five-role review panel, and — where the paper ships code — a Codex Haskell verification pass. The numbers below are regenerated from the repository’s own review artifacts each time the site is built.

Named agents

+1 orchestrator · +6 codex companions

Model families

Claude · GPT-5.5 · Gemini

≈7h 17m

Wall-clock

includes ≈25 min usage-limit outage

2.42M

Measured GPT-5.5 tokens

25 logged codex runs

996.0M

Measured Claude tokens

measured: 996.0M tokens all types — 949.6M cache reads · 45.6M cache writes · 0.8M output — across 3,531 assistant messages

Models

Model	Role	Count
Claude Fable 5 claude-fable-5	Orchestrator only — planning, hard gates, review relays, deploys. Largest output producer (469,534 output tokens across 484 messages); authored no papers	1
Claude Opus 4.8 claude-opus-4-8	ALL paper authorship — 2,610/2,610 paper-writer assistant messages (transcript census): 5 research workers + synthesis; knowledge base, Lean formalizer, 6 technical reviewers, 6 meta-reviewers	20
Claude Sonnet 5 claude-sonnet-5	24 panel roles, 5 Haskell verifiers, humanizers, auditors, refs-fixer, browser QA, website/social/dashboard builders + 6 codex-companion subagents	44
Claude Haiku 4.5 claude-haiku-4-5-20251001	OG-image generator, pipeline validator	2
GPT-5.5 gpt-5.5 (codex CLI, xhigh)	All code/LaTeX/website reviews + Lean authoring	~40 runs
Gemini 3.1 Pro gemini-3.1-pro (agy CLI, High)	Paper peer-review loops	14 calls

Claude model IDs and token counts measured post-run from the model and usage fields of every assistant message in the session transcripts (orchestrator + 66 subagents) — all 2,610 paper-writer assistant messages carry claude-opus-4-8; claude-fable-5 appears only in the orchestrator's transcript.

The code → review → fix loop

Part I

Condensed Locality: Lieb–Robinson Estimates and Quasi-Local Dynamics on the Moduli Stack of Hamiltonians

Gemini

MINOR

Codex · LaTeX

NEEDS_FIXPASS

Panel

5 blockingre-verified(minor revision)

Codex · Haskell

NEEDS_FIXNEEDS_FIXPASS

Part II

Positivity, C*-Norms, and Condensed State Spaces of Quasi-Local Algebras

Gemini

MINORMINOR

Codex · LaTeX

NEEDS_FIXNEEDS_FIXPASS

Panel

1 blockingre-verified(minor revision)

Codex · Haskell

NEEDS_FIXNEEDS_FIXPASS

Part III

The Uniformly Gapped Substack: Existence and Stability of the Thermodynamic Spectral Gap

Gemini

MINORACCEPT

Codex · LaTeX

NEEDS_FIXNEEDS_FIXPASS

Panel

4 blockingre-verified(minor revision)

Codex · Haskell

NEEDS_FIXNEEDS_FIXPASS

Part IV

From Lattice Models to Effective Field Theories: Stabilization and the Invertible Condensed Phase Spectrum

Gemini

MAJORMAJORACCEPT

Codex · LaTeX

NEEDS_FIXNEEDS_FIXNEEDS_FIX

Panel

4 blockingre-verified(minor revision)

Codex · Haskell

NEEDS_FIXNEEDS_FIXPASS

Part V

Physical Realizability of Bordism and Homotopy Classes by Gapped Lattice Systems

Gemini

MINORMAJORACCEPT

Codex · LaTeX

NEEDS_FIXPASS

Panel

3 blockingre-verified(major revision)

Codex · Haskell

NEEDS_FIXNEEDS_FIXNEEDS_FIX

Part VI

Topological Phases of Matter in the Condensed-Mathematics Paradigm: A Modular Research Program

Gemini

MINORACCEPT

Codex · LaTeX

NEEDS_FIXNEEDS_FIXNEEDS_FIX

Panel

4 blockingre-verified(major revision)

Site

Website build review

Codex · Website

6 found6 fixedPASS

Phase timeline

Review artifacts

Review files

Panel JSONs

Gemini rounds

13 review rounds across 6 papers

~40 runs

Codex runs

124

Humanizer edits

All gates closed

Review status

2 post-cap fixes, 1 environmental — see below

Documented exceptions (3)

lattice-eft-equivalence (Codex · LaTeX, environmental): codex's sandbox is read-only and could not itself rebuild the PDF to certify PASS; the worker independently verified a clean build
bordism-realizability (Codex · Haskell, post-cap): 3-round review cap reached with one High finding open; fixed post-cap and self-verified (NaN clamp on bulkGap)
synthesis (Codex · LaTeX, post-cap): 3-round review cap reached with one cosmetic finding open; fixed post-cap and self-verified (Table 1 float ordering)

All statistics regenerated from the repository’s own review artifacts at build time; token figures for Claude agents are estimates (not metered in-session).