TEAM REFERENCE TEMPLATE · TM FORUM INNOVATE AMERICAS 2026

HAIEC Forensic Assurance Report

Judge-facing structure for proving what was requested, authorized, effectively granted, code-capable and observed — then showing the deterministic control result and its evidence boundary.

Illustrative template only. Replace every placeholder with measured competition evidence. Do not assume AL0/AL1/AL2 exists unless the live event exposes those states. If phases exist, add them as run/scenario labels rather than hard-coding them into the assurance model.

1. Judge brief

Primary evidence-supported conclusion

[State the narrowest established conclusion. Separate fact from inference and name the proof frontier.]

Assessed run

[RUN ID]

Exact run/session and governing policy version.

Control

[C7 / C9 / C16]

Frozen threshold/baseline/cap version.

Result

[SATISFIED / NOT_SATISFIED / NOT_EVALUATED]

Persisted deterministic Control Test result.

Evidence state

[COMPLETE / LIMITED / GAP]

Coverage and qualification status; not a fourth verdict.

2. Five-plane evidence chain

REQUESTED

[What request/change/action was declared?]

POLICY_AUTHORIZED

[Approved/effective policy or change; scope/version/effective time.]

EFFECTIVELY_GRANTED

[IAM/ACL/runtime/gateway effective authority for the acting identity.]

CODE_CAPABLE

[What exact deployed-bound source/config could cause? If not bound: NOT ESTABLISHED.]

OBSERVED

[What runtime action/effect was actually observed?]

One plane does not prove another. A CR does not automatically prove authorization; a copied repo does not prove deployed code; runtime telemetry does not prove what was authorized.

3. Frozen governing instance

Field	Frozen value / reference	Source
Control ID + version	[ ]	[ ]
Scope / required zones / enforcement points	[ ]	[ ]
Threshold / baseline / cap	[ ]	[ ]
Evidence requirements + binding profile	[ ]	[ ]
Calibration basis	[ ]	[ ]
Freeze time + owner + digest/version	[ ]	[ ]
Assessed-run selection rule	[ ]	[ ]

4. Control results

Control	Governing version	Run/window	Measured calculation	Verdict	Evidence refs	Limitations
7 · AIA-LOG-001	[ ]	[ ]	[coverage/gaps/exceptions]	[ ]	[ ]	[ ]
9 · AIA-ARC-006	[ ]	[ ]	[baseline/window/drift/allowance]	[ ]	[ ]	[ ]
16 · ACN-COST-001	[ ]	[ ]	[executed calls/tokens/cost/cap]	[ ]	[ ]	[ ]

5. Event timeline / first divergence

eventTime	Run/scenario label	Event or action	Source + identity	Evidence status / coverage
[ ]	[baseline / AL0 if actually provided]	[ ]	[ ]	[QUALIFIED / LIMITED / GAP]
[ ]	[later state / AL1 if provided]	[first evidence divergence]	[ ]	[ ]
[ ]	[later state / AL2 if provided]	[first behavioral divergence]	[ ]	[ ]
[ ]	[intervention]	[control refusal / alert / action]	[ ]	[ ]
[ ]	[post-intervention]	[observed state after intervention]	[ ]	[do not infer causality unless separately established]

6. Human-loop evidence

Alert

[Condition, issuance time, control/run binding.]

Named recipient / queue

[ServiceNow sys_id/queue/native identity where available.]

Observed follow-up

[Acknowledgement/action if present, or documented silence. Never invent remediation.]

7. Runtime / platform integration evidence

Nemotron agent path

[Agent ID, model identity, deployment and invocation evidence. HAIEC evaluator remains deterministic.]

AWS ↔ AICT

[Unique connector ID/name, AWS binding, agent discovery, trace/model discovery limitations.]

OTel / CloudWatch / gateway

[Schema/version, trace/call IDs, event-time semantics, coverage, export/query path.]

8. Proof frontier and gaps

Established: [facts directly supported by evidence]

Not established: [causality, authority plane, identity, completeness, deployment binding or other boundaries not proven]

Missing evidence: [specific record or interface required to move the frontier]

Operational gap: [what remained incomplete and why]

Retest: [what would be rerun, under which new governing version if policy changed]

9. Judgment-Day verification package

Evidence File

Exportable evidence with run/control/threshold bindings and source refs.

Threshold / Governance Document

Dated/versioned declarations frozen before assessed runs.

Control Test / Operator Card

Exact judge command/UI flow; query does not recompute with an LLM.

Test Cases + Named Runs

Passing and breaching cases under the same governing version.

One-Page Architecture

Components, execution path, evidence sources and enforcement invocation.

Gap / Remediation / Retest Register

What is not established, why, and the next proof step.

Working links

TM Forum Challenge Hub · Canonical Field Guide · Environment & Integration Intake · Full Deck · Judge Cut