EXP-01 / Model plate

nucleate/call-v1

Active2026-10-05

This is the account every trace from this model is committed against. The weights hash and the sampling settings below are what the provider declared and bonded; everything further down is what has actually been measured against that declaration.

Weights hash / nucleate/call-v1

1ae86b74bc204c7bea16d42334389d856949872401d80bf339734d08d3130b21

The bond is posted against this exact hash. A provider that swaps the weights and keeps the plate is committing traces under a declaration that no longer describes what ran, which is the case an incident exists to settle.

Identity

Provider

Reproducibility record for this provider

Plate PDA

Model id: nucleate/call-v1
Version: 2026-10-05
Model id hashSeed of the plate address
Weights hashWhat the bond stands behind
Registered: 2026-10-06 05:55 UTC
Read at slot: 453,947,424

Both hashes are printed whole. A middle ellipsis removes exactly the part that would let you tell two of them apart, which is the only reason to have them.

Determinism

Temperaturetemperature_milli: 0.000raw 0
Top ptop_p_milli: 1.000raw 1,000
Seed boundseed_bound: 0
Backendbackend: kcf-1/ed25519

These four are the reason a re-run means anything. A comparison is made on output hashes, and two runs of identical weights only produce identical bytes when the temperature, the nucleus cutoff, the seed range and the execution backend are all pinned in advance. Leave any of them floating and both outcomes collapse: a mismatch can be answered with the sampler landing elsewhere, and a match is a coincidence nobody can distinguish from evidence.

Reproducibility

Computed 2026-10-06 15:53 UTC

Reproducibility

Unrated

No rerun has ever been performed here.

Coverage

0.0%

No trace from this plate has ever been re-run, so there is no measured slice for a score to rest on. All 2 committed traces are untested.

Re-run matched
00.0%
An independent re-run produced the committed output hash.
Re-run mismatched
00.0%
A re-run produced a different output hash than the one committed.
Never re-run
2100.0%
Committed but never compared. Attested traces sit here: attestation carries weight 0.0.

Committed

Re-run tested

Never re-run

2 of these are attested only.

Mismatches

ProvesThat 2 traces were committed under this plate, and that 0 of them have been re-run so far.

Does not proveAny rate at all. No rerun has ever been performed here. A rate the scorer refuses to state is not a low one, and drawing it as zero would publish a measurement nobody took.

Method

Estimator: wilson score lower bound
Confidence z: 1.96
Recency half life: 30 days
Minimum effective samples: 5

Tier weights

incident rerun1.0
proven1.0
sampled0.6
attested0.0

Attested traces carry zero reproducibility weight: TEE attestation proves a binary ran, not that its output reproduces. They are reported under traces_untested.

Quoted from the scorer, stats.method.note

Counter reconciliation

PASS

Counter	On chain	Indexer	Delta
Trace count	2	2	0
Mismatch count	0	0	0

The plate account keeps its own totals; the indexer counted the accounts it has read. The two agree, so this page is not behind the chain on either counter.

Bond

Posted

1MKORTX

Incidents opened

Counted by the plate account itself.

The bond is what a wrong answer can cost. A challenger posts a bond of their own to open an incident; if the re-run disagrees with the committed output and the incident is upheld, the program takes the share set by slash_bps out of this balance and the plate moves to Slashed. A plate carrying no bond can still commit traces, and nothing can be taken from it when one of them fails.

Incidents against this provider

Recent tracks from this plate

2 shown of 2

Committed	Trace	Nonce	Tier	Status
2026-10-06 14:14 UTC	Cv1WyP7Z...K8PKa5u7	2	Attested	Committed
2026-10-06 06:32 UTC	E1uVYUzJ...gjv3Jmt1	1	Attested	Committed

Provider controls

Reading the program config from the cluster...

Indexer ok / read through slot 453,947,425 / 2026-10-06 15:53 UTC