Role ExposureFintech evidence map
Snapshot 18 Oct 2026PUBLIC

■ MODELS · VINTAGE AETHER-FIN-2026-10-18 · REGISTRY, NOT A LEADERBOARD

No invented eval scores.

A named model appears here only through a published Role Exposure pack run. Its cell holds the verdict of that run, never a score for the model.
Empty is untested, not 0. Literature is a blotter, not a cell.

OUR PACK RUNS

2

P01 recovered display · P02 signed run

NAMED MODELS RUN

1

claude-sonnet-5 · P02 · works

LAB MODELS ON OUR PACKS

1 OF 4

P02 only · other packs untested

PACKS SPECIFIED

01–04

01 recovered · 02 run · 03–04 no run

INVENTED SCORES

NONE

refused

■ CANDIDATE × PACK · THIS VINTAGE ONLY

Scroll horizontally to inspect every pack →

CandidateKindP01 · Bank ↔ bookP02 · Name-screening L1P03 · AML TM L1P04 · Payments exceptions
Historical P01 displayRecovered public display; open the pack for its record
Historical record
Recovered displayOpen pack →
Untested
Untested
Untested
claude-sonnet-5Anthropic Messages API · thinking disabled · run 16 Oct 2026 · P02-PUBLIC, SAMPLE book
Named model · pack run
Untested
WorksReceipt model-20261016T152109Z-ae7e77Open pack →
Untested
Untested
Named lab modelsA Models cell requires a published, versioned Role Exposure pack run
Lab model
Untested
Untested
Untested
Untested

What a cell may hold

Recovered display

A published historical pack display with its evidence status and limitations.

Untested

No versioned Role Exposure pack result exists for that cell. Not 0. Not “failed.”

Cited · not our pack

External literature remains in Watch. It does not fill a Models cell.

Open Watch →