Leadership Under Uncertainty

SCO Judgment Layer Compatible

Data platforms record what is known. Models infer. Agents act. Workflow systems execute. None of them carries a durable representation of what an organisation currently believes, how uncertain it is, what alternatives it has entertained, why a judgment changed, and whether prior judgments were borne out.

That representation either survives an exchange between two systems or it is quietly destroyed. This profile makes the difference checkable — and says nothing at all about whether the judgment is any good, which is a separate measurement.

Four independent classes

Not a ladder. A read-only dashboard can be a Consumer without ever being a Producer; an attestation service can be Attested without parsing a single view.

Consumer
Ingests judgment artifacts and preserves their epistemic content.
No flattening a calibrated probability into a label, no dropping the citation that bound a claim to its source, no conflating the first-person posture vocabulary with the third-person stance vocabulary.
Producer
Emits artifacts another system can consume without private coordination.
Revisions are recorded as change events carrying a reason, not applied as destructive overwrites. A judgment that moved without saying why is not a judgment another system can audit.
Round-trip
Closes the loop from judgment to decision to returned outcome.
This is what separates a judgment layer from a reporting layer. Without a returned outcome, nothing in the system can ever be shown to have been wrong.
Attested
Produces or verifies artifacts whose integrity a third party can recompute offline.
Demonstrated against a fixture set containing one clean artifact and two tampered ones. A verifier that passes everything is not a verifier.

Registry

leadership-under-uncertainty-reference-implementation 2026.072026-07-28 · SCO-JLC 1.0-draft
Classes attained: producer, roundtrip, attested (assessed: producer, roundtrip, attested)
Self-assessment of the reference implementation. The first run attained no class (five failures, published unedited); all five were fixed and this is the re-run. Consumer remains unassessed — it needs a live re-emission.
reportHash 92fd5f4b6d024f854aadb574d60c391c27d97cc1caadeea5ca1291ba1e0d07a8

The first entry is the maintainers’ own system. Its first run attained no class — five failures, published unedited — which is the ordinary result of running a check you did not design around your own implementation. All five were then fixed and the entry re-run; the statement keeps the record of what they were rather than presenting a clean sheet. Entries are append-only: a system that later regresses is annotated, never deleted.

Run it yourself

The validator is deterministic and offline: no network access, no model calls, no account. A compatibility claim that can only be checked by the party making it is not a claim worth publishing.

sco-jlc assess --input ./bundle --system my-platform --out statement.json
sco-jlc verify-report statement.json

What a statement does not establish

That the judgments carried are true. That the reasoning behind them is sound — that is what the companion benchmark measures. That a recorded rationale is the reason the system actually reached its output. That everything material was recorded. That the artifact reflects the model’s internal state.

And, at the system level: no security or privacy property is assessed, a listing is not an endorsement, and a statement describes only the version assessed on the date assessed.

Status

SCO-JLC 1.0-draft is a candidate profile published under Apache 2.0. Assessment is self-run with the open validator and re-verified by the maintainers before listing. Until an independent system is listed, this profile is technically open and socially closed — the same limitation the external bake-off framework acknowledges of itself.

Want this for your team?

Drop your email and we'll be in touch when access opens for your vertical.