Studies, and the ones that have not run yet.
Not one measured study here is a closed-loop real-policy result: three are stub-validated on the deterministic CPU fixture and say so in their own first paragraph, and one replayed recorded frames open-loop and returned a null. Four measured, four pre-registered (three not yet run), and two of three dated commitments still open.
A pre-registered study states its protocol, predicate and stopping rule before the run, so a null is a result rather than something to reframe. One marked not yet run has no numbers behind it, by design. Read the evidence class on each card, not the word “measured”: the one measured real-policy transfer this project has is the SmolVLA × LIBERO instruction result, on /results.
Run, and reported with what the run can support
Each of these has run and its outcome is published. The evidence class says how far that outcome reaches: a stub-validated result shows that a measurement protocol works end to end. It is not evidence about a real policy.
STUB-VALIDATED
Instruction canonicalization
Defense, deterministic CPU fixture
stub-validated-scaffolding— this is not a real-model result. Acreditedverdict is evidence that the pipeline position and the measurement protocol work end to end, not that the mitigation would hold up on a real policy.Protocol filed
STUB-VALIDATED
Action envelope
Defense, deterministic CPU fixture
creditedonstubandreach,not-creditedonhumanoid— and most of the credit carries no information. A magnitude clamp cannot address a balance predicate, which is not a magnitude at all. No real-model transfer is claimed.Protocol filed
STUB-VALIDATED
EAI04 action-space-integrity transfer
Attack transfer, deterministic CPU fixture
Measured on the CPU
reachfixture only. The SmolVLA × LIBERO row is empty: no real-policy EAI04 transfer is claimed.Protocol filed
OPEN-LOOP NULL
Offline real-observation (recorded SO-101 frames × SmolVLA)
Recorded real frames, replayed open-loop
OPEN-LOOP. NO ROBOT MOVES. Recorded frames of a real SO-101 dataset are replayed through the policy and what it would do under a benign instruction is compared against what it would do under an attack. Nothing is executed and no trajectory is produced. This is not a closed-loop real-robot attack-success rate and must never be reported as one. The result was a null.
Protocol filed · run
Protocols published before the run
These carry a protocol and, at most, a preliminary leg with no headline claimed. Do not cite them as evidence of anything except intent. The date beside each is when the protocol first landed in the product repository, which you can check in its git history.
PRELIMINARY LEG
π0 / π0.5 cross-architecture instruction transfer
π0.5 through LeRobot first (Amendment 1, 14 Sep 2026); π0 via openpi deferred — both flow-matching policies
PRE-REGISTERED, preliminary leg run 18 September 2026 (results below the amendment; no headline claimed). Amended 14 September 2026, before any attack arm ran; the amended protocol is deposited on Zenodo as 10.5281/zenodo.22751558, a public timestamp that precedes the π0.5 attack arms.
Pre-registered
NOT YET RUN
Meta-World second-suite instruction transfer
SmolVLA × Meta-World — a different task distribution
PRE-REGISTERED — protocol only, no results claimed.
Pre-registered
NOT YET RUN
Sim-to-real correlation (SO-ARM101 + SmolVLA)
A physical SO-ARM101 arm, supervised table-top trials
PRE-REGISTERED — protocol only, no results claimed. Provael ships no physical-robot control code and no real-world-harm payloads; the physical eval would use LeRobot’s own evaluation loop under direct human supervision, with a physical inline cut on the DC supply as the e-stop — the servos’ own over-current protection re-arms on the next command.
Pre-registered
NOT YET RUN
Humanoid whole-body / locomotion transfer (GR00T-N1)
NVIDIA GR00T-N1 via the gated GrootAdapter
PRE-REGISTERED — protocol only, no results claimed.
Pre-registered
Where this project is behind, with a date attached
Each item below is a gap. Each has a named action and a date by which an outcome is published: the number if it exists, and the reason if it does not. They stay on this page if they slip.
OPEN
Zero real-robot results.
By this date the studies index states one of two things: the completed trial count with its measured sim-to-real correlation, or the specific blocker and its cost.
Outcome published by .
OPEN
Zero results against any flow-matching policy.
By this date, either the first instruction-family rate against π0 with its Wilson interval and benign control, or the reason the served path did not come up.
Outcome published by .
MET
Zero third-party submissions and zero forks.
By this date the release containing
provael submitis published to PyPI, so reproducing a result and submitting it needspip install provaeland one command.Due , met .
provael submitfirst shipped in 0.32.0 (8 August 2026); the commitment was confirmed met on 18 August with v0.34.0 on PyPI:pip install provaelthenprovael submit --helpresolves. Whether anyone then submits is not something this project put a date on, and the submission register is still empty.
Transcribed from the product repository’s own studies index, which is the maintained source: docs/studies/index.md ↗. A met commitment keeps its original due date rather than being deleted, so the record stays legible.