+ NASA PUBLIC RESEARCH / FLAME IN FREEFALL
Explore the experiments behind a cabin environment. See what matches, what differs, and what remains unknown.
◇ Source-checked preview. Independent human verification is pending; verified coverage counts only reviewed physical runs.
Loading indexed evidence
Conditions and source locators will appear here.
Abstracts and unlocated legacy passages are document context. They do not count as physical runs.
Pinned walkthrough examples · each answers a different part of the scenario.
Loading comparison sources…
Oxygen × absolute pressure · unique physical runs
A briefing will appear after the evidence loads.
DETERMINISTIC RESEARCH GAPS
Runs are filtered by research lane, endpoint and the chosen gravity scope. Exact material matches are grouped ahead of general context. Each record is compared against one coherent condition phase.
O₂ ±2 pp · pressure ±5 kPa · gravity ±0.03 g · flow ±5 cm/s
These are search controls, never engineering thresholds. A documented microgravity regime may be mapped to 0 g for search, without claiming a measured residual acceleration. Oxygen values require a documented mole or volume basis.
Coverage uses 5 percentage-point oxygen bins and 10 kPa pressure bins. Each physical run counts at most once in a cell. Aggregate or unidentified observations are excluded. The default grid counts independently human-verified records; the optional source-checked preview is explicitly unreviewed.
Numerical outcome differences are calculated only when endpoint, definition, units, phase and protocol agree. Differences in material, geometry, gravity or flow remain visible as comparison limits.
Curated public reports supplement the uploaded literature baseline. No October 28 PSI dataset package was supplied. PSI catalog entries are identified separately from downloaded report evidence. Abstracts are document context and never counted as executed runs.
The existing fine-tuned EmbeddingGemma 2 checkpoint supplies retrieval where its verified index is available. Actual retrieval mode is shown per request. Saved training scores use synthetic and unreviewed teacher questions; independent human evaluation is pending.
Gemini, when configured, drafts cited scope notes. Structural checks cannot establish scientific correctness; generated claims require human review. Without a valid response, a deterministic evidence-only briefing remains available.
Development used AI for code, source extraction assistance and original decorative chamber art. The art depicts a fictional chamber. It is not an image of a NASA experiment. All scientific values link to source records.
