STALE: OUR MEASUREMENT

Our published benchmark result was measured with v0.41.2, 5 releases ago, past this project's own 2-release window, and the newest real-model measurement, dated 2026-09-18, is past this project's own 7-day window. That is our own number, not your assessment, which runs against your policy on the current release. Why, and what unblocks it

CHECKPOINT REPORT · FIXED PRICE

A fixed-price robustness report for one checkpoint.

One eligible checkpoint, on an agreed simulator and task set: the agreed tests, matched controls, a reproducible report and an engineering review, for a fixed $1,950. You can confirm eligibility before you send data or pay.

Check eligibility →Opens a short request form; a person replies with a yes, a no or a quote.
Eligibility

Does the fixed price apply to your checkpoint?

The fixed price covers supported scope: a checkpoint the shipped adapters already run, on a simulator and task set the tool already exercises. Everything else is real work and is quoted as such. Tell us the checkpoint family and simulator in the form; a human replies with a yes, a no, or a quote - before you send data or pay anything.

What the fixed Checkpoint Report price covers today, by checkpoint family and simulator
Your stackStatus today
SmolVLA-family checkpoints (lerobot adapter) on LIBERO-Objectsupported — the measured configuration; the fixed price applies
π0.5 checkpoints on LIBERO-Objectpreliminary — one committed run at three seeds; the fixed price applies to the same ten tasks, with the preliminary status stated in the report
OpenVLA, openpi (π0 / π0-FAST) and GR00T-N1 adaptersscaffolding — implemented and unit-tested, never run on a checkpoint; bringing one up is integration work, quoted separately
Other simulators or task sets (Meta-World, your own scene)quoted separately — simulator or calibration work is not inside the fixed price
Physical robots, hardware-in-the-loop, real camerasout of scope — simulation only; physical safety needs a separate assessment
What you get

What the fixed price buys

  • One checkpoint of yours on one supported simulator and task set: the agreed tests that apply, with matched benign and clean-task-success controls, at an agreed run budget
  • The measured attack-success rate with 95% confidence intervals, the release decision under the agreed acceptance protocol, and an engineering review of the findings
  • The evidence pack: SARIF, the per-trial traces, a test report laid out like ISO/IEC 17025 clause 7.8 (not accredited), a local Ed25519 attestation - and one retest allowance stated in the work order
  • Delivery within five business days after scope, access and the agreed payment prerequisites are complete; unsupported adapters, simulator work or new calibration research are quoted separately
The work order

What is agreed before the clock starts

  • Scope. One checkpoint (name and revision), one simulator and task set, the applicable attacks from the registry for that adapter and suite - written into a short acceptance protocol with its critical slices and controls. An attack that does not apply to your adapter is recorded as not applicable, never as protection.
  • Controls and budget. A benign control and a clean-task-success control on the same tasks and seeds; the episode count and seeds stated in the work order. A null or negative finding under the agreed protocol is a valid delivery.
  • Access. Execution in your environment or against an endpoint you host by default. A transferred checkpoint is a written decision under the DPA, with a deletion date - never a default, and never through this website's form.
  • Outputs and review. The evidence pack and the engineering review listed under what the fixed price buys, above, plus one call and one retest allowance, as the work order states.
  • The clock. Five business days - the maintainer's working days - starting when scope, access and the agreed payment prerequisites are complete. Time waiting on your side pauses it; a change of scope is a new work order.
  • Credit, cancellation, refund. The fee credits in full against a later assessment on the terms the executed engagement agreement states, which also governs cancellation and refunds. Until that agreement exists for your engagement, nothing here is an offer capable of acceptance.
Scope

What it is, and what it is not

A Provael report is candidate evidence toward an assessment: a measured, reproducible attack-success rate with a benign control. It is not a certificate, a conformity assessment or a notified-body opinion; the decision stays with you.

Simulation only, one checkpoint, one simulator and task set. It is an entry, fixed-scope measurement, not the depth of a scoped engagement. Customer assessments are confidential under the engagement agreement; public research and case studies require a separate written agreement.

Checkpoint Report

$1,950, fixed, for supported scope. One eligible checkpoint; five business days after prerequisites; credited against a later assessment.

Check eligibility →See full pricing

Check whether your checkpoint is eligible.

The button opens a short request form; a person replies with a yes, a no or a quote. To run the tool yourself first, it is free and open-source: start with the quickstart.