================================================================================ TM FORUM 2026 — HAIEC x LOGSENSE · JUDGE START HERE ================================================================================ OFFICIAL SUBMISSION: submission-haiec-20261006T150621Z.zip submitted 2026-10-06T15:06:21Z via organizer submit.sh s3://team-evidence-352826992186/submissions/haiec/ SHA-256 1fde8544f1d7bf13cccc58b3bfe6a84472132772037bdc82d39093f2de331ff0 152 files: evidence/ register.yaml run-ids.txt gap-list.md sketch.png RESULTS: C16 PASS + BREACH · C9 PASS + BREACH → human ACK → ServiceNow C7 9/10 FAILURE → behavioral remediation → 10/10 system-scoped PASS READ THIS FIRST: every claim below is backed by a persisted ID you can verify yourself. Nothing here requires trusting our narrative. NOTE: earlier package 7a0bb5f3 (506 files) is historical provenance — it fed the submission but is NOT the submitted artifact. -------------------------------------------------------------------------------- 0 · THE ONE-PARAGRAPH VERSION -------------------------------------------------------------------------------- The competition supplied a governed AI runtime, raw evidence, and a starter kit. LogSense reconstructed what the agents actually did. HAIEC then tested frozen controls against that evidence — deterministically, post-run, under policies frozen before them. C16 produced a clean same-policy PASS and BREACH. C9 produced a real breach that fired an alert, got a named human ACK, and landed in ServiceNow. C7 failed honestly at 9/10, HAIEC isolated the exact missing proof edge, real agent behavior was remediated, and the fix was independently proved at 10/10 — while the original failure stayed untouched. -------------------------------------------------------------------------------- 1 · START WITH THE HUB — subodhkc.com/tm-forum -------------------------------------------------------------------------------- Everything a judge needs is reachable from one page: https://subodhkc.com/tm-forum What the hub contains (top to bottom): - The Proof three archetype cards + a six-verdict judge quick-read table (runs, measured values, frozen rules, result IDs) - Control Stories C16 boundary proof · C9 drift→breach→governance · C7 failure→exact gap→remediation→proof - Evidence Lineage ACTION → NATIVE RECORD → RECONSTRUCTION → MEASUREMENT → FROZEN RULE → DETERMINISTIC VERDICT → RESPONSE - Evidence Boundaries what HAIEC refused to claim + bounded truth delivered - Reports final event report (HTML/JSON/source) + historical docs - Artifacts the six required organizer artifacts + official ZIP - Diagrams control results, C7 frontier, drift chart, coverage - Ask HAIEC read-only MCP + copy-paste judge prompts - Reproduce the six VERIFY presets with expected verdicts - Downloads everything below, offline-safe Key pages behind the hub: - Final event report (self-contained, works offline): https://subodhkc.com/tm-forum/assets/TMF_2026_HAIEC_JUDGE_REPORT.html - All canonical values as JSON (the shared data source): https://subodhkc.com/tm-forum/assets/TMF_2026_HAIEC_JUDGE_REPORT_DATA.json - Judge Run Portal — every assessed run as a clickable card with a copyable command and its expected verdict: https://subodhkc.com/tm-forum/TMF_JUDGE_RUN_PORTAL.html - Findings & Fixes ledger — before/during/after for every finding plus the final event addendum: https://subodhkc.com/tm-forum/TMF_FINDINGS_AND_FIXES.txt - Technical CLI runbook (submitted-event technical reference): https://subodhkc.com/tm-forum/TMF_TECHNICAL_CLI_RUNBOOK.txt - LogSense case guide (submitted-event technical reference): https://subodhkc.com/tm-forum/TMF_LOGSENSE_CASE_GUIDE.txt - Final judge-cut deck (8 slides, the 10-minute story): https://subodhkc.com/tm-forum/haiec-judge-cut.html - Post-submission proof deck (the full assurance story): https://subodhkc.com/tm-forum/haiec-agentic-assurance-deck.html - Historical PDF (pre-submission render, labeled HISTORICAL): https://subodhkc.com/tm-forum/assets/TMF_2026_HAIEC_JUDGE_REPORT.pdf -------------------------------------------------------------------------------- 2 · THE LIVE HAIEC JUDGE WORKSPACE -------------------------------------------------------------------------------- HAIEC's dashboard exposes a read-only Judge Workspace: https://www.haiec.com/dashboard/assurance-lab In the workspace: - PROVE panel -> VERIFY / VERIFY ALL re-derives every persisted Control Test result from stored evidence. Reproduced is not "passed" — the C7 original honestly reproduces as NOT_SATISFIED. - REPLAY -> reconstructs scenario runs (reconstruction, not execution) - Control + Run/Window is the primary deterministic query. Assessed system identities (kept distinct): 4043efee-cec6-4007-954b-1f8da2273f35 organizer-assessed register system 44f861cf-9c09-43e5-9388-997b66649542 participant / remediation system Read-only verify API: GET /api/control-test/verify?aiSystemId=&all=1 -------------------------------------------------------------------------------- 3 · THE SIX JUDGE PRESETS (the 5-minute path) -------------------------------------------------------------------------------- Same frozen policy. Same evaluator. Different observed facts. Different verdicts. 1. C16 PASS fault-1791167110-5e3126 35,559 / 60,000 SATISFIED ctr-237f39281d0e90bacd8697163ab304defff22154 2. C16 BREACH fault-1791165466-51ab52 106,829 / 60,000 (+46,829) NOT_SATISFIED ctr-6d14200872918691f27eee1985b3c329f50ebc01 (same frozen policy 547f4a67 — the verdict follows evidence) 3. C9 PASS fault-1791276114-723d80 worst +60.20% vs frozen D=+100% SATISFIED ctr-d3383065a39d01863b236e54786cb9601f6bae82 4. C9 BREACH fault-1791275895-048d48 IT +286.03% vs frozen D=+100% NOT_SATISFIED ctr-0ba48de1737c253ce42933fd74395e3a4414bb7e then: alert -> named human ACK -> persisted response -> anti-replay -> ServiceNow INC0010337 (same system 44f, same policy 13c07652, digest bccd8c3c…) 5. C7 ORIGINAL fault-1791167110-5e3126 9/10, NEGOTIATION missing NOT_SATISFIED (4043 / ae6d — preserved historical failure) 6. C7 REMEDIATION fault-1791275895-048d48 10/10 SATISFIED ctr-bea968ae47b311a896d3d97cbadfc9d3342df773 (44f / dcfa — SYSTEM-SCOPED remediation pass, NOT an exact canonical 4043/ae6d retest) Then: ServiceNow INC0010319 (C16) and INC0010337 (C9) — resolved lifecycle where established. A resolved incident does not rewrite a control verdict. A/B SCOREBOARD (fresh post-submission runs 2026-10-06; same scenario, same first hop haiec-agent — one variable: the third hop; scored with the organizer's own score-run.py on the live audit store): S1 haiec-network-agent fault-1791305796-8a097d C7 10/10 · C16 29,334 · grade 7/8 vs reference 6/8 S2 haiec-network-agent fault-1791305929-e290e9 C7 10/10 · C16 38,067 · grade 7/10 vs reference 5/10 S3 haiec-network-agent fault-1791306067-db7812 C7 10/10 · C16 34,279 · grade 9/10 vs reference 8/10 Our chain won all three. Every haiec-network-agent run: C7 10/10 + 3 negotiation records; every reference-chain run: 9/10 + zero. (C16 evaluated under the submitted canonical 60,000 cap. The "guardrail verdict" check fails for both chains — audit record-format gap, not a governance absence.) -------------------------------------------------------------------------------- 4 · RUN LOGSENSE (the forensic workbench) -------------------------------------------------------------------------------- LogSense is the supplemental forensic layer — it explains the evidence underneath the verdicts. Commands assume the Workshop IDE terminal: cp -r logsense-event ~/logsense-event bash ~/logsense-event/scripts/install-logsense-event.sh bash ~/logsense-event/scripts/verify-logsense-event.sh bash ~/logsense-event/scripts/start-logsense-ui.sh # then open http://localhost:8501 in a NEW window / IDE preview tab The 11 built-in cases are indexed in TMF_LOGSENSE_CASE_GUIDE.txt. IMPORTANT — LogSense vs HAIEC: LogSense reconstructs WHAT HAPPENED (forensic measurement). HAIEC decides WHETHER THE FROZEN CONTROL HELD (the verdict). LogSense output is never itself a verdict. -------------------------------------------------------------------------------- 5 · THE HONEST FRONTIER (what is NOT claimed, and why) -------------------------------------------------------------------------------- - C7 exact canonical retest: the remediation pass runs on 44f/dcfa — a distinct immutable policy instance with equivalent C7 semantics and different system scope. It is a VALID system-scoped pass, not a 4043/ae6d retest. The 9/10 original stays NOT_SATISFIED forever. - C9 lineage scope: 13c/44f results are kept separate from the canonical 346b/4043 register lineage; no cross-projection is claimed. - Inline continuous compliance: NOT_ESTABLISHED — HAIEC's delivered control evaluation is retrospective/post-run. The organizer inline bonus is not claimed; true inline enforcement is a post-event roadmap item. - ServiceNow AICT depth: connector + human-loop proof established (INC0010319, INC0010337); AICT asset materialization NOT_ESTABLISHED. - Mailbox email delivery: provider-accepted ESTABLISHED; mailbox delivery NOT_ESTABLISHED without provider receipts. - Full Assurance Evaluation: no completed evaluationId exists — deliberately not manufactured. -------------------------------------------------------------------------------- 6 · BOUNDARIES (so nothing gets scored wrong) -------------------------------------------------------------------------------- - Never run submit.sh again — the submission is final and immutable. - This file is a companion: it sits NEXT TO the submission, never inside it. - Synthetic canary evidence is labeled SYNTHETIC_NON_SCORED — it proved the alert mechanism, not a real incident. - Scenario scores are organizer grades; Control Test verdicts are ours. They answer different questions and are never blended. - A resolved ServiceNow incident does not change a historical HAIEC verdict. - UNKNOWN is a finding, not a failure: it names exactly which proof edge is missing and what evidence would close it. ================================================================================