YY public evidence desk

Proof before promises. Limits beside results.

Open the artifact, inspect the observed result, and read the boundary that keeps the conclusion honest. These are public release fixtures and reproducible research evidence—not client testimonials or broad guarantees.

  • Public artifacts
  • Exact observations
  • Explicit non-claims

Reproducible case files

Three claims you can inspect now

The useful unit is not a screenshot or a promise. It is a named input, a reviewable output, and a conclusion that stops when the evidence stops.

  • Paid software · release fixture

    AgentSafe

    The 0.0.1-rc7 fixture records six PASS findings, zero FAIL findings, zero not-observable findings, and a passing portability gate for one instrumented FunctionTool workflow.

    Boundary: this proves only what the named release fixture recorded under the supported OpenAI Agents SDK 0.19.x adapter. It is not proof that an agent is safe or secure.

    • 0.0.1-rc7
    • 6 PASS · 0 FAIL
    • Exact event IDs
    • Raw JSON retained
  • Open research · behavioral model

    Analog Matmul

    Release v0.1.0 publishes an executable 8 × 8 model, driver-facing contract, 35 Python tests, synthetic experiments, and a canonical receipt that CI regenerates on Python 3.11 and 3.12.

    Boundary: this is behavioral-model evidence. There is no verified transistor design, layout, fabricated device, silicon measurement, or physical performance result.

    • v0.1.0
    • 35 Python tests
    • Canonical SHA-256
    • 0 silicon measurements

Demonstrations—not client results

See the deliverable before sharing files

Both service samples use fictional source material. No client engagement was performed; the files demonstrate output shape, review flags, and the boundary of the paid work only.