You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Only the Semantic rules example is exposed in the browser app. The other four are unverified and stay hidden until each one asserts what it produces.
Why
The verification runs assert presence, not behaviour. The prove target asserted expect: [".out"] — that an output element exists — so a page rendering NOT PROVABLE with an empty proof tree passed cleanly on every run, and the example shipped broken to the published site (see #1).
The same shape applies to the others:
target
assertion
asserts behaviour?
semantic
fired counts per input, derived values, parameter/facts round-trip, section state, no model download
yes
semantic-dynamic
computed-vector provenance, fired count
yes
address
.pill exists
no
fraud
.decision, .prov exist
no
streaming
button.primary exists
no
prove
.out exists
no
Whether those three besides Proof are correct or merely unasserted is not yet known.
Address, Fraud triage, Streaming — audit the output against an independent calculation, then assert the values rather than the elements.
Bar for re-enabling
An example goes back in the nav when its run asserts the values it computes, corroborated independently of the app — the way the semantic bench is checked against llama-server measurements rather than against itself.
Only the Semantic rules example is exposed in the browser app. The other four are unverified and stay hidden until each one asserts what it produces.
Why
The verification runs assert presence, not behaviour. The
provetarget assertedexpect: [".out"]— that an output element exists — so a page renderingNOT PROVABLEwith an empty proof tree passed cleanly on every run, and the example shipped broken to the published site (see #1).The same shape applies to the others:
.pillexists.decision,.provexistbutton.primaryexists.outexistsWhether those three besides Proof are correct or merely unasserted is not yet known.
Per example
Bar for re-enabling
An example goes back in the nav when its run asserts the values it computes, corroborated independently of the app — the way the semantic bench is checked against
llama-servermeasurements rather than against itself.