ProjectH / MethodSource · identity · compatibility · publication
ProjectH.
Latest profile
Publication method

A visible number keeps its receipt.

ProjectH separates what a source reported from what has been qualified for comparison. The public interface currently demonstrates that boundary with published source-category records; it does not claim a ProjectH score.

Traceability

Start with the record, not the shape.

Every displayed result should resolve to a source revision, observation or publication date, retrieval date, metric identity, unit, and immutable input. Presentation never becomes evidence by itself.

New evidence creates a new snapshot. Corrections point to the superseded record instead of silently rewriting history.

Identity

A model name is not a deployment.

Capability evidence belongs to the system that produced it: deployment or release identity, reasoning mode, harness, prompts, tools, generation settings, and evaluation date where available. Unknown configuration remains explicit.

Qualification questions

Five questions before publication.

  1. What exactly was published?Source revision, artifact identity, reported value, and dates.
  2. What system produced it?Deployment identity and compatible run configuration—or an explicit unknown.
  3. Can the metric join this profile?Metric version, unit, transform, weight, and manifest compatibility.
  4. Can the evidence be used and republished?Technical review and visible rights status remain separate gates.
  5. What can the result support?Coverage, freshness, uncertainty, limitations, and final publication review.
Unknown handling

Missing does not mean zero.

If required evidence is missing or incompatible, the axis remains unknown. ProjectH does not reweight the remaining metrics to manufacture a complete profile, infer configuration from a marketing name, or use polygon area as an overall score.

Current boundary

Source evidence is visible. Public qualification is still gated.

The current public surface reproduces published source-category records and makes their display calculation inspectable. Backend M0 evidence review, rights approval, deployment identity, compatible manifests, uncertainty policy, and public score authorization remain separate gates.

Private-profile intake, structured corrections, pricing, and response commitments will open only with named owners and published data-handling terms.