Methodology
Science at the level of process
A scientific platform is only as trustworthy as the discipline it enforces on itself. These principles are operating rules — not aspirations — and every campaign is bound by them.
Operating principles
- Evidence over assertion — a claim without traceable support does not enter the state.
- Measurement over opinion — confidence is a measured quantity, never a rhetorical one.
- Baseline before improvement — nothing is called better without a measured baseline.
- Ablation before attribution — a component earns credit only when removal shows its effect.
- Falsifiability as a requirement — a hypothesis that cannot fail is not admitted.
- Failure is data — null and negative results are recorded, catalogued and reported.
- Pre-registration — predictions are registered before data collection begins.
- Calibration — stated confidence is continuously compared with observed accuracy.
- Provenance — every artifact carries a lineage from question to result.
- Reproducibility — campaigns replay from first principles on any compatible substrate.
- Explicit gates — nothing advances silently; every transition is a recorded decision.
- No fabricated authority — the platform never claims consensus, review or replication that has not actually happened.
From principle to practice
Pre-registration
Research questions, metrics and success criteria are locked in the record before results exist — the audit trail shows what was expected, and when.
Calibration tracking
Registered predictions are scored against outcomes. The platform maintains an honest account of when its confidence was earned.
Replication
Key findings are re-run independently — different seeds, fresh state, same protocol — before they are treated as converged.
Negative results
A dedicated catalogue preserves what did not work. Nulls are part of the scientific record, not noise to be discarded.
Ablation
Attribution claims are tested by removal, not by narrative — a lesson inherited from experimental science.
Trace and lineage
Every result can be traced back through experiments, predictions and evidence to the original question.
Oversight and disclosure
The deep mechanics of the platform are confidential by design — the methods themselves are the asset. What gets validated externally are results and performance, under controlled conditions and non-disclosure agreements. Verification never requires handing over the machinery.
Publication of results is a human decision, made case by case — never an automatic platform behavior. The system prepares evidence packages; the decision to release them belongs to people.
Read the method in action
Every campaign in the workspace shows its research state, its registered predictions and its verification record.