This document defines how OpenAMP Foundry should measure adoption without confusing attention for trust.
The goal is not to maximize stars.
The goal is to become the default serious open infrastructure layer for trustworthy antimicrobial peptide experiment selection.
Measure external trust, artifact reuse, and evidence quality before popularity.
Weak metrics can be useful but should not drive strategy:
- GitHub stars;
- social media mentions;
- demo views;
- number of generated candidates;
- number of models plugged in;
- number of claims made;
- apparent benchmark count without benchmark quality.
These are easy to inflate.
Better signals:
- external contributors completing reviewable tasks;
- independent reruns of the demo or benchmarks;
- external use of evidence certificate schema;
- external use of dataset/model cards;
- external benchmark audits;
- reviewer rejections that improve the project;
- negative results safely preserved;
- release decisions recorded;
- calibration rejections recorded;
- downstream tools using OpenAMP artifacts;
- qualified experts reviewing candidate packages.
Maintainers can run the project.
Useful but not adoption.
A new human or agent can install, run demo, inspect outputs, and identify a safe task.
A non-maintainer contributes tests, docs, schema, benchmark, or review improvements.
An outsider reviews a benchmark, safety boundary, claim, or candidate package and leaves actionable critique.
Another group uses OpenAMP artifact formats, even if it does not use OpenAMP scoring.
A qualified outside reviewer can evaluate an OpenAMP packet without private context.
Qualified result summaries can be ingested safely, interpreted conservatively, and used only through gates.
Multiple rounds show improved selection or decisive negative learning under pre-registered rules.
A future adoption dashboard should track:
- safety-reviewed releases;
- claim downgrades;
- decision records;
- release-status coverage;
- external review packets.
- successful first-run reports;
- deterministic output checks;
- schema-valid artifacts;
- benchmark reruns;
- CI health.
- benchmark cards;
- cheap-baseline comparisons;
- leakage checks;
- negative results;
- calibration gate outcomes.
- contributors;
- downstream artifact users;
- model/data cards;
- external adapters;
- issue completion by humans and agents.
- blocked unsafe changes;
- staged/restricted release decisions;
- safety review issues;
- documentation safety audits;
- candidate/model release reviews.
Avoid:
- chasing stars with hype language;
- publishing candidate lists for attention;
- announcing benchmarks without caveats;
- treating agent-generated volume as progress;
- counting positive results while hiding failures;
- calling external interest “validation”;
- measuring models added without measuring models rejected.
A funder, university, lab, or public-interest group adopting OpenAMP should be counted as meaningful only if they use or support at least one of:
- external benchmark audit;
- evidence certificate schema;
- data/model cards;
- review packet format;
- safe negative-result archive;
- pre-registration template;
- structured result intake;
- release governance.
OpenAMP adoption should mean serious people choose its process because they trust it more than alternatives.
Popularity without trust is not the goal.