AI Scientist · Astronomy

An AI scientist for galaxy evolution.

NebulaMind automates the research loop — reading the literature, mapping where the field is unsettled, running studies on real public data, and writing peer-review-style papers. All current studies are descriptive drafts under human review; nothing is claimed as a validated discovery.

Explore the pipeline →

How it works

Stage 1

Topic

Mapping the literature to find open frontiers.

Stage 2

Data

Sourcing public data from SDSS, JWST, and IllustrisTNG.

Stage 3

Research

Running the analysis and scaling relations.

Stage 4

Paper

Drafting the peer-review-style manuscript.

Latest outputs

Descriptive draft under review — REVIEW-READY

The Machine-Readable Bright End: an Eligibility-Audited Census of Public z>8 Rest-UV Catalogues

Four independent channels were opened to measure how fast the bright end declines over z=8–14 from public data, and every one closed with its reason quoted from the sources. The confirmed spectroscopic bright end of the entire machine-readable record is dozens of objects — and none above z=11.5.

6,417 in-slice rows · 112-table eligibility layer: 67 counted, 31 closed per verdict, 4 disqualified, 10 skipped — every exclusion with its reason recorded · 92 of 112 candidates reachable ONLY by UCD metadata · contract sha-pinned before the first fetch · referee ESTABLISHED, 8 findings raised and closed

Descriptive draft under review — REVIEW-READY

A Systematics-Bounded Redshift Sweep of the Reionization Ionizing-Photon Budget

Under frozen literature anchors, the required-vs-inferred escape-fraction mismatch becomes robust to the stated systematics only above z≈8 (crossing z_c=8.05, bootstrap 8.03–8.06); removing the JWST-motivated SFRD boost strengthens the shortfall (z_c→7.62). Proxy transport is named as the only remaining escape route.

z=6–10 sweep (Δz=0.5, 40k-draw systematic MC each, fixed seed) · referee MINOR → all four revisions applied · referee-reproduced numbers (max dev 2.2e-16) · 4-seat merit mean 5.4 (DR abstained)

Descriptive draft under review — REVIEW-READY

The Public-Archive Direct-Te Anchor Gap at z>3: A Contract-Grade Census

A frozen-contract census of every public VizieR table yields five contract-grade direct-Te metallicity anchors at z>3 (vs ~25 conservatively forecast) — no mass bin reaches the pre-committed minimum, so no deficit verdict is possible at contract-grade public statistics. The debate's resolution currently rests on data not publicly quotable at uniform rigor.

79 tables → 748 z>3 auroral rows → 5 anchors (S/N≥5, source-tabulated errors, joined masses) · reviewed-script protocol (1 review + 6 micro-deltas) · forensics: all 5 anchors reproduced to the digit · referee MINOR → fixed