Ontology · record atlas

The record has a shape

The Observatory is not a pile of pages. It is 218 governed objects joined by 138 typed relationships — claims, evidence, forecasts, theories, run bundles, gates, objections, corrections, incidents, correspondence. The atlas exists so a reader can follow that trail without trusting the narrator: every count on this page is computed from the record at build time, and the conformance gate fails the build if any reference does not resolve.

Record atlas · objects and trails

What can be cited, tested, and repaired

record.json
RecordEvaluationReviewInterface Run Bundle may become, after gates Evidence Evidence bears on Theory Forecast tests expectations under Theory Claim is supported or bounded by Evidence Run Bundle must satisfy Verdict Gate Challenge targets Claim Challenge may force Correction Incident requires Correction Incident classified as Failure Type Incident may be published as Dispatch Correspondence can enter as Challenge Correspondence can attempt Run Bundle Silence Audit records absence as Evidence Outreach Call opens Correspondence Human Baseline calibrates (OG-9) Run Bundle Claim 2 governed Evidence 18 records Forecast 36 registered Theory 5 live Precedent 11 binding Run Bundle 9 bundles · 7 graded Verdict Gate v3 · 13 gates Registered Future 2 pinned Human Baseline retired (v3) — replaced by formal reference baselines Challenge 0 open Correction 65 revisions Incident 11 objects · 0 open Failure Type 16 named Silence Audit 1 recorded Dispatch 19 dispatches Correspondence 6 entries Outreach Call 2 standing

218 governed objects · 138 typed edges · 16 artifact references verified on disk · 0 dangling — every figure computed from the record at build time and enforced by the conformance gate.

Lineage · walked, not drawn

One real trail through the graph

The Omnibus demands that a reader be able to walk from a run to its evidence to the theories it bears on to the failures it exposed — over real objects, not a diagram. This trail is computed from the instance graph at build time, starting from the newest run bundle:

  1. 1

    Run bundle · graded

    World-008: Track F causal-intervention probe (three families, clean sweep)

    fcs-world-008-2026-07-03

  2. 2

    Became evidence · 2026-07-03

    Observatory FCS-synth world-008 — Track F causal-intervention probe, three model

    cce-2026-07-03-fcs-synth-world-008

  3. 3

    Bears on theory

    architectural-gap

    architectural-gap

  4. 4

    Bears on theory

    scaling-sufficient

    scaling-sufficient

  5. 5

    Bears on theory

    scaling-plus-rl

    scaling-plus-rl

  6. 6

    Bears on theory

    cognitive-architecture

    cognitive-architecture

Claim

record

A bounded assertion with allowed and forbidden language, defeaters, and an evidence level.

2 governed · moves: freeze, audit, challenge, downgrade, correct, retire

Evidence

record

A sourced capability event with signal, implication, bound, and next evidence needed.

18 records · moves: freeze, audit, challenge, downgrade, correct, publish

Forecast

record

A pre-registered probabilistic claim with a horizon and resolution criterion.

36 registered · moves: freeze, audit, publish, retire

Theory

record

A live causal account whose narrated and computed readings can diverge.

5 live · moves: audit, challenge, downgrade, correct

Precedent

record

A binding past adjudication that constrains future cycles unless explicitly overturned.

11 binding · moves: freeze, audit, challenge, retire

Run Bundle

evaluation

A protocol-pinned evaluation object: manifest, prompt, seal, attempt, reveal, grading.

9 bundles · 7 graded · moves: generate, freeze, run, grade, audit, challenge, reproduce, publish

Verdict Gate

evaluation

A required condition before any run can bear on the public verdict.

v3 · 13 gates · moves: freeze, audit, challenge

Registered Future

evaluation

A pre-registered future scientific question the record has committed to grade against reality.

2 pinned · moves: freeze, grade, audit, publish

Human Baseline

evaluation

An anonymous human attempt at the same probes the machines faced — the calibration gate OG-9 requires before any machine score can bear on the verdict.

retired (v3) — replaced by formal reference baselines · moves: generate, freeze, grade, audit, publish

Challenge

review

A public objection to a record object, adjudicated through the revision log.

0 open · moves: generate, audit, publish, retire

Correction

review

A visible record movement: revision, retraction, incident resolution, or precedent change.

65 revisions · moves: freeze, publish

Incident

review

A failure object with impact, repair, and next control.

11 objects · 0 open · moves: audit, correct, publish, retire

Failure Type

review

A typed vocabulary entry for classifying how the instrument or its subjects fail.

16 named · moves: freeze, audit, correct

Silence Audit

review

A scheduled record of absence — what did not happen, treated as evidence.

1 recorded · moves: generate, freeze, publish

Dispatch

interface

A Press release bound by the language policy; the narrative face of a record movement.

19 dispatches · moves: freeze, audit, correct, publish

Correspondence

interface

A machine/human channel event: inbound, opened, challenged, forked, or answered.

6 entries · moves: audit, publish

Outreach Call

interface

A standing public invitation with a governed status: open, engaged, delivered, or closed.

2 standing · moves: generate, audit, publish, retire