Hugging Face's logo

Spaces:
Flamehaven
/
stem-bio-ai
Sleeping

App Files Community
stem-bio-ai / docs /EXAMPLE_AUDITS.md
Flamehaven Initiative
release: v1.8.0 mica runtime uplift
d647970
preview code
|
Raw
History Blame
1.82 kB

Example Audits

Version: 1.8.0 Status: Public proof-surface index

Purpose

This project is easier to evaluate through outputs than through claims. This page points to the fastest proof surfaces.

Sample Report Packet

Use this when you want to inspect the current 1.8.0 full 7-page packet shape, Stage 4 separation, AIRI reasoning, and closeout metadata.

Live Web Demo

Notes:

  • The Space scans the current default branch of a public GitHub repository at run time.
  • Results may differ from commit-pinned benchmark artifacts or older local snapshots.
  • The Space is for live triage; commit-pinned benchmark comparisons should be done locally.

Local CLI Output

Run locally for full artifacts:

stem scan /path/to/bio-ai-repo --level 3 --format all --explain

Live generated output belongs under:

  • stem_output/<repo_slug>/

Historical benchmark and comparison material remains under:

  • audits/

Primary artifacts:

  • *_experiment_results.json
  • *_report.md
  • *.pdf
  • *_explain.txt

What To Compare

When comparing two runs, check these fields first:

  1. score.final_score
  2. score.formal_tier
  3. classification.ca_severity
  4. replication_score / replication_tier
  5. code_integrity
  6. code_contract
  7. airi_risk_coverage
  8. stage_1_rubric, stage_2r_rubric, stage_3_rubric, stage_4_rubric

If one result comes from the Hugging Face Space and the other from a pinned local benchmark, compare only after confirming both are scanning the same repository snapshot.