01
What it shows
Rows cover data provenance, derived-signal posture, output receipts, evaluation cooperation, and disclosure cadence.
Chris Royse / leapable.ai
Evidence Map - methodology v0.1 - updated 2026-04-25
Google DeepMind — Evidence Map methodology trace.
Public technical reports, Frontier Safety Framework materials, and model materials; no cooperated evidence in v0.
02
What it avoids
There is no single rank, no hidden score, and no claim that public evidence equals internal practice.
03
How to correct it
Send better public evidence through the contact page. The methodology marks cooperated rows and links the disclosure at /governance#issuer-pays.
Evidence only matters when it traces back.
Watch on YouTube· 12:19
Document analysis with provenance back to source files
Chris walks through a document-intelligence pipeline that extracts text, metadata, images, entities, and source locations before asking questions over the corpus. That same proof posture drives the evidence map: claims need source rows, dates, and traceable artifacts.
Watch videoOpen the full video on YouTubeGrade grid
- TDP
- C+
- DSP
- C+
- ORP
- D
- IEC
- C
- DC
- C+
Public documentation of corpus origin, licensing posture, and auditability.
Whether the entity describes how it derives labels, eval data, or supervision from existing corpora.
Whether outputs can carry a verifiable receipt, content credential, or comparable provenance record.
Whether the entity cooperates with third-party evaluators under a published method.
Frequency and completeness of methodology, capability, and limitation publications.
Have one context problem?
Send the audience, data type, target task, proof bar, and sharing limits. The first pass should stay bounded.