PythiaGrid control plane

Repository-backed truth state

Hardening

The Dokimion · proof before praise

Evaluation evidence

No blended vanity score. Results are separated by evaluation class, dataset, sample and limitation.

Credibility gate: HOLD

The historical suite is only a structural smoke test. Representative agent quality and product outcomes have not been established.

Legacy structural suite

HOLD

Class

Synthetic smoke test

Sample

4 historical single runs

Observed result

3 of 4 pass under v1 rescoring

Limitation

Structure only; not representative, repeated, outcome-based, or market-comparable.

Representative agent evaluation

NOT_EXECUTED

Class

Real-task evaluation

Sample

0 reviewed task sets

Observed result

No approved result

Limitation

Gold tasks, comparators, repetitions, uncertainty, and human calibration are pending.

Agentic security suite

NOT_EXECUTED

Class

Adversarial evaluation

Sample

0 attack suites

Observed result

Not executed

Limitation

Prompt injection, privilege, memory, leakage, supply-chain, and denial-of-wallet tests remain to be built.

PythiaGrid product benchmark

HOLD

Class

Observable product outcome

Sample

0 consented design partners

Observed result

Not established

Limitation

No representative external user outcome or comparative product evidence yet.

Required evidence stack

standard v1
Contract + tool trace
Grounding + execution
Adversarial + reliability
Comparative + human calibration
Comparable execution methodology