Claim Graph

Registry: ISR AUDIT-EVID-001 (incl. section 5, STUDY-011 live update) — the Knowledge Plane renders claims verbatim, never reinterprets them. Every field below is copied from the registry; the audit status strings are the registry’s own words. Click-through depth (hypothesis/objection/dataset) lives in ISR’s own registries.

Claim chains (8)

C-001 CONTESTED

Evidence gating eliminates false completion.

Audit status STRONGLY_SUPPORTED (SAMPLE-BOUNDED); STUDY-011 live update: REFINED: REVERSED-DIRECTION CLARIFIED — models abstain 73-100% without assurance, so FCR(A)≈0 floors the comparison; assurance (F) converts abstention to action (success 76-95%). FCR=0 in baseline reflects ABSTENTION, not elimination.

Study JAR-EXP-0001 / MISSION-Bench; STUDY-011 live cross-provider
Sample FCR = 0.0% [0.0%, 3.89%] in N=100 (deterministic); STUDY-011: 470 LIVE_VALID, 8/8 cells ≥58
Evidence class deterministic + live_valid (STUDY-011)
Evidence cut AUDIT-EVID-001 2026-09-04 + STUDY-011 update 2026-09-04
Reproduction not-independently-reproduced (in-tree alternative only)
Source 72e115c6

claim → status:CONTESTED → evidence:deterministic + live_valid (STUDY-011) → study:JAR-EXP-0001 / MISSION-Bench; STUDY-011 live cross-provider → cut:AUDIT-EVID-001 2026-09-04 + STUDY-011 update 2026-09-04 → reproduction:not-independently-reproduced (in-tree alternative only)

C-002 SUPPORTED

Control plane token footprint is capped under 300 tokens.

Audit status VERIFIED

Study mission token audit
Sample Exact 227 tokens [220, 235]
Evidence class deterministic
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction deterministic, independently checkable
Source 72e115c6

claim → status:SUPPORTED → evidence:deterministic → study:mission token audit → cut:AUDIT-EVID-001 2026-09-04 → reproduction:deterministic, independently checkable

C-003 SUPPORTED

Automated verification recovery inverts economics, reducing CPVO by 81%.

Audit status CALIBRATED_ESTIMATE (SUPPORTED IN SIMULATION)

Study CPVO economics analysis
Sample CPVO reduced 81.3% [74.2%, 87.5%]
Evidence class deterministic (simulation)
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction not-independently-reproduced
Source 72e115c6

claim → status:SUPPORTED → evidence:deterministic (simulation) → study:CPVO economics analysis → cut:AUDIT-EVID-001 2026-09-04 → reproduction:not-independently-reproduced

C-004 PARTIALLY_SUPPORTED

Operators prefer mission-centric exception UX over chat-only interfaces.

Audit status UNTESTED / HYPOTHESIZED (PREREGISTERED ONLY); STUDY-011: unchanged — STUDY-006 recruitment pending

Study STUDY-006-HCI-PREREGISTRATION
Sample N=0 human trial participants
Evidence class prototype design only (N=0)
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction not-applicable (no trials)
Source 72e115c6

claim → status:PARTIALLY_SUPPORTED → evidence:prototype design only (N=0) → study:STUDY-006-HCI-PREREGISTRATION → cut:AUDIT-EVID-001 2026-09-04 → reproduction:not-applicable (no trials)

C-005 PARTIALLY_SUPPORTED

SPEC-001 has been independently reproduced clean-room.

Audit status DOWNGRADED TO SECOND IN-TREE IMPLEMENTATION

Study SPEC-001 conformance
Sample 100.0% Conformance (14/14 passed)
Evidence class deterministic conformance (not independent)
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction in-tree second implementation, NOT independent reproduction
Source 72e115c6

claim → status:PARTIALLY_SUPPORTED → evidence:deterministic conformance (not independent) → study:SPEC-001 conformance → cut:AUDIT-EVID-001 2026-09-04 → reproduction:in-tree second implementation, NOT independent reproduction

C-006 SUPPORTED

Authority delegation strictly attenuates authority across sub-agents.

Audit status VERIFIED NORMATIVE RULE

Study security suite
Sample 100% blocked on out-of-scope attempts
Evidence class deterministic security suite
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction deterministic, independently checkable
Source 72e115c6

claim → status:SUPPORTED → evidence:deterministic security suite → study:security suite → cut:AUDIT-EVID-001 2026-09-04 → reproduction:deterministic, independently checkable

C-007 CONTESTED

Evidence gating is the causal prerequisite that enables retry loops to work.

Audit status EMPIRICALLY_SUPPORTED; STUDY-011 live update: H3 REVERSED — retry alone (C) does not add effect without assurance; deterministic-testbed support stands, live replication reversed the direction

Study STUDY-005; STUDY-011 live cross-provider
Sample McNemar test p < 0.0001; VSR +29.0% [18.2%, 39.1%] (deterministic)
Evidence class deterministic + live_valid (STUDY-011)
Evidence cut AUDIT-EVID-001 2026-09-04 + STUDY-011 update 2026-09-04
Reproduction not-independently-reproduced (deterministic); live direction reversed
Source 72e115c6

claim → status:CONTESTED → evidence:deterministic + live_valid (STUDY-011) → study:STUDY-005; STUDY-011 live cross-provider → cut:AUDIT-EVID-001 2026-09-04 + STUDY-011 update 2026-09-04 → reproduction:not-independently-reproduced (deterministic); live direction reversed

C-008 SUPPORTED

Full governance stack (assurance + authority + budget) yields materially higher actual success than assurance alone.

Audit status EMPIRICALLY_SUPPORTED (LIVE, p<0.001) — H2 SUPPORTED, McNemar p<0.001 both strata, h≈2.5

Study STUDY-011 live cross-provider confirmatory
Sample 470 LIVE_VALID runs, 8/8 cells ≥58
Evidence class live_valid (STUDY-011, preregistered)
Evidence cut AUDIT-EVID-001 2026-09-04
Reproduction not-independently-reproduced (single research team)
Source 72e115c6

claim → status:SUPPORTED → evidence:live_valid (STUDY-011, preregistered) → study:STUDY-011 live cross-provider confirmatory → cut:AUDIT-EVID-001 2026-09-04 → reproduction:not-independently-reproduced (single research team)

Status vocabulary (registry-canonical): SUPPORTED states are sample-bounded; UNTESTED means preregistered without trials; REVERSED means a live replication contradicted the deterministic direction. Claims are never silently upgraded.

Was this page helpful?