Method and data

How findings are ranked

You choose an atmosphere: oxygen O, pressure P and airflow F. Each finding has its own test conditions o, p and f.

closeness    = 1 − mean( |O−o| / 70,  |P−p| / 101,  |F−f| / 50 )
               conditions the source does not report are skipped
relevance    = safety relevance (1, 2 or 3) / 3
completeness = reported conditions / 3
score        = closeness × relevance × completeness

The divisors are the width of each axis: 70 % O₂, 101 kPa and 50 cm/s. A finding that leaves out a condition is never dropped and never given a guessed value. It loses score through completeness and its card says “not reported”.

This formula is our own design for this tool, not a published NASA method. It is simple on purpose so that every score can be checked by hand.

Worked example

Atmosphere: exploration atmosphere, 34 % O₂, 56.5 kPa, 15 cm/s. Finding: saffire-005, Cotton-fiberglass fabric (SIBAL).

oxygen       1 − |34 − 31| / 70 = 0.957  (% O₂)
pressure     1 − |56.5 − 54.1| / 101 = 0.976  (kPa)
airflow      1 − |15 − 20| / 50 = 0.900  (cm/s)

closeness    0.944
relevance    3 / 3 = 1.000
completeness 3 / 3 = 1.000
score        0.944

Safety relevance

Each finding carries a relevance tag set when the dataset was built:

• 3: directly informs material flammability or extinguishment.
• 2: informs detection or flame behaviour.
• 1: fundamental combustion science.

Tested, thin and untested

A finding is in close range of a point when it reports both oxygen and pressure and lies within ±5 % O₂ and ±15 kPa of it. Findings taken from the same burn share one oxygen and pressure pair, so they count as one test condition.

• Tested: 3 or more test conditions in close range.
• Thin: 1 or 2.
• Untested: none.

The same rule gives the verdict for your atmosphere, the hatching on the chart and the gaps panel. For the ISS cabin preset it returns tested: 3 findings from 3 test conditions. For the exploration preset it returns thin: 4 findings from 2 test conditions. A gap is not evidence of safety. It means this dataset holds no source-checked finding there.

Data rules

• A missing value is shown as “not reported”. It is never guessed.
• A finding appears only after every value was compared with its cited source document. This check was AI-assisted. Independent human review of each row is tracked separately and is not yet complete; the Verify panel shows the status of each finding.
• A flow direction the source does not state is shown as not reported.
• Every card links to its source document and names where in it to look.

How AI is used

• AI helped extract values from NASA documents into the dataset and wrote the investigation summaries and the safety implications for the two preset atmospheres.
• AI-written text is marked with a violet border and the label “AI-generated from NASA PSI records”.
• Each implication cites finding ids. One that cites a finding outside the ranked evidence on screen is not shown. With fewer than 3 cited findings the panel says “Insufficient evidence”.
• The ranking, the coverage verdict, the chart and the gaps use no AI. With every language model switched off they are unchanged.

Limits

• Not a combustion simulator or a predictive model.
• Not a certified safety tool or a source of design limits.
• Not a replacement for NASA's PSI repository. It is a way into it.
• The dataset is small: 28 findings from 7 investigations, of which 11 report both oxygen and pressure.