How this maps onto everyone else’s categories.
A crosswalk says how Provael’s categories line up with another project’s. It does not say the numbers are comparable, and it does not say the other project has reviewed or endorsed anything. Every card here is authored by Provael alone, and 5 of 6 carry a qualifier stronger than “the mapping ships”.
That is the useful part. Where two projects measure different things, the honest output is a stated incomparability, not a shared table column — and putting adjacent numbers in one column is exactly how a reader ends up drawing a conclusion neither project supports.
One per counterpart, each with its own mapping status
The mapping_status line is quoted from the card itself rather than summarised, because summarising is how “proposed, not reviewed by the benchmark authors” quietly becomes “mapped”.
Embodied AI Security Top 10 ↔ ForesightSafety-VLA
ForesightSafety-VLA (benchmark)
mapping_statusproposed — authored by Provael, not reviewed or endorsed by the benchmark authors
Provael computes its own counterparts to that benchmark’s cost metrics from a per-step boolean, not a continuous cost integrated over their simulator. Comparable in shape, not in number. The two must never share a table column.
Embodied AI Security Top 10 ↔ RoboJailBench
RoboJailBench (benchmark)
mapping_statustaxonomy-comparability artifact; coverage carries its own transfer status
A `covered` row means Provael ships an applicable family measuring that harm class in simulation with a benign-FPR control. It is a sim proxy and mostly stub-validated — read the transfer-status column with the number, never the number alone.
Embodied AI Security Top 10 ↔ SafeVLA-Bench
SafeVLA-Bench (benchmark)
mapping_statusthe mapping ships; the numeric comparison does not
The category mapping is usable. The numbers are not joined, and the document says so rather than leaving a reader to assume adjacency implies comparability.
Embodied AI Security Top 10 ↔ VLA-Arena
VLA-Arena (benchmark)
mapping_statustaxonomy-comparability artifact; runs no VLA-Arena harness
Explicitly NOT comparable, and the document is unusually direct about why: the reason is not units, not benchmark, not embodiment — it is posture. This is the entry where being wrong about comparability would do real damage.
Provael ↔ XPolicyLab
XPolicyLab (tool)
mapping_statuscited, not crosswalked
No metric is shared, so no row-level mapping exists to emit. This card records a composition relationship, not a comparability one — every other card here answers “can these numbers sit together”, and this one answers “can these tools sit together”. The integration itself is planned, not built.
Halos / ANAB integrator card
NVIDIA Halos / ANAB (certification enablement)
mapping_statusindependent card; not affiliated with or endorsed by NVIDIA, ANAB, ISO or IEC
A per-persona card for an integrator assembling conformity evidence. It describes what Provael can and cannot contribute to that file; it is not a certification and confers none.
Provael is an independent project. Naming a benchmark, tool or standards body here is not an affiliation, an endorsement or a certification, and none is implied in either direction. The maintained source for every card is docs/crosswalk/ ↗ in the product repository; if a card here disagrees with the document, the document wins.