Ten scoped reviews of cosmology and gravitation papers: reproduced results, bounded attempts and narrowly supported findings.
Selected cases from the project intake; not a representative sample, error-rate estimate or AI leaderboard. Each outcome applies only to the fixed reviewed version and declared scope.
Editorial correction only: distinguish executable packages from documentary reports, date preparation history, correct citation versions and formula formatting.
Scientific claims, code, data and saved numerical results are unchanged. Previous published versions remain available.
What these results mean
Reproduced in scope: the named checks support the reported result, with the stated exceptions; not a certification of the whole paper.
Partly reproduced: some named results reproduce while important parts remain unverified.
Confirmed finding: the specific discrepancy has author or independent researcher confirmation; downstream consequences may remain unknown.
Technical finding — not externally confirmed: a supported, explicitly limited discrepancy without external confirmation.
Attempt documented: the completed work and reasons it stopped are recorded. Failure to finish does not establish a paper error.
The review supports selected parameter shifts and their attribution to baryon acoustic oscillation measurements, with numerical exceptions and an unfinished modified-gravity fit.
Limit One interval width and tracer amplitude miss tolerance; storage failure stopped the modified-gravity fit before usable parameter samples were retained.
✦ OpenAI Codex + Anthropic Claude▤ 147.6M tokens processed◷ 51.1 h recorded session time (includes waits)
Partial records · 17/17 sessions with complete token counts · 16/17 with activity time recorded
◎ OpenAI Codex · gpt-6-astra · reasoning: high
✦ Anthropic Claude · claude-opus-5-5 · reasoning: high · settings reported by the project owner
Algebra and selected parameter summaries reproduce. Checks of the exponential dark-energy model and logarithmic-model probability tails remain unresolved.
Limit Supernova parameter agreement and precision with the full Planck likelihood remain unmet for the exponential model. Causes are unresolved; further computation is paused.
✦ OpenAI Codex + Anthropic Claude▤ 93.6M tokens processed◷ 18.0 h recorded session time (includes waits)
Partial records · 6/6 sessions with complete token counts · 6/6 with activity time recorded
◎ OpenAI Codex · gpt-6-astra · reasoning: high
✦ Anthropic Claude · claude-opus-5-5 · reasoning: high · settings reported by the project owner
The pressure–distance equations and null-energy-condition identities reproduce. Numerical agreement depends on our particular reconstruction implementation.
Limit The headline significance and location remain unverified. Our reconstruction was repaired after inspecting earlier results; equivalence to the authors' pipeline is unestablished.
✦ OpenAI Codex + Anthropic Claude▤ 41.9M tokens processed◷ 4.5 h recorded session time (includes waits)
Partial records · 3/5 sessions with complete token counts · 1/5 with activity time recorded
◎ OpenAI Codex · model not recorded · reasoning: not recorded
◎ OpenAI Codex · gpt-6-astra · reasoning: high
✦ Anthropic Claude · claude-opus-5-5 · reasoning: high · settings reported by the project owner