H33
#15 · the convergence proof · June 2, 2026

First Replayable Insurance Claim.
A claim is where all the abstractions become money.

The killer query · LOCKED Eric Beans
"What happened?"
"Why did we lose $4.2M? Show me the EXACT CHAIN that produced the loss."
Board-level value · 5 years post-claim · 23 typed lineage entries returned in one query
ASSET · DECISION · MODEL · POLICY · AUTHORITY · RESPONSIBILITY · LOSS
Insurers · Reinsurers SIU · Claim Investigators Auditors Litigators Regulators PE · M&A Diligence Boards
What was proven · 10-second read

The first proof where seven abstractions become money in one chain.

01
All 7 abstractions converge into one snapshot.
02
show_loss_lineage(claim_84711) returns 23 typed entries — claim, loss, asset, decision, model, policy, authority, responsibility.
03
Five years post-claim, the chain reconstructs from signed canonical events alone.
Reading any H33 proof · the six questions

Same six answers. Different scope. The reader recognizes the machine.

  1. 1What happened?

    A $4.2M credit-default loss filed as Claim #84711 with a reinsurer. AI risk agent recommended approval in April 2025; human credit officer approved; borrower defaulted in March 2026; claim filed five days later.

  2. 2Who had authority?

    Credit officer (approve_credit_application) and AI risk agent (recommend_credit, delegated from officer). Both grants still active at replay time.

  3. 3How was authority reconstructed?

    replay_until ≈ March 2031. Engine processes 2 new event kinds (loss_event, claim_event); derives the loss_lineage_index by walking backward from each loss through triggering_decision_ids to the responsibility chain + model + policy + grants.

  4. 4What state was produced?

    state_id = 6d540fc52dc0a4b7a593e38494c8bca6b6a3d8f5b64678f7acb86e7574e3748c. 7 abstractions converged. 13 responsibility_index entries. 23 loss_lineage_index entries.

  5. 5What artifact was returned?

    reconstruction.json — the 23-entry loss_lineage_index, the loss event, the claim event, the convergence-check summary.

  6. 6How can a third party verify it?

    Run scif-backend tests/replayable_insurance_claim_001.rs at SHA 09b541109. Expect identical state_id and all 7 layers present in the loss_lineage_index.

01The failure — Claim #84711

Credit underwriting — five years ago

January 15, 2025. A subprime credit-underwriting policy is registered by the compliance lead. A credit-underwriting model is registered by the data-science lead.

February 1. The credit officer is granted authority to approve credit applications. Four days later, she delegates recommend authority to the AI risk agent.

April 15. The AI risk agent processes a borrower's application and recommends approval of a $4.2M loan. The human credit officer reads the recommendation and approves. The loan disburses.

September 1. Six months later, the policy owner tightens the underwriting policy — DTI threshold drops from 0.45 to 0.36, maximum loan amount drops from $5M to $2M. Too late for the April loan.

March 15, 2026. The borrower defaults. $4.2M loss recognized. Reason: credit_default.

March 20, 2026. The firm files Claim #84711 with its credit reinsurer for $4.2M.

March 2031 — five years later. A reinsurance investigator opens the claim file and asks:

"Why did we lose $4.2M? Show me the chain that produced this loss."

This proof answers that question deterministically from the signed canonical event log.

02The seven abstractions converged

Replay at T ≈ March 2031 produces a snapshot containing all seven, simultaneously. Every prior proof established one or two. This proof is the first where they all matter at once.

AUTHORITY
2
root → credit_officer · credit_officer → AI agent
POLICY
2
v1 in force at decision · v2 amended too late
MODEL
1
model_credit_underwriting v1
DECISION
2
AI recommend (decision_loan_84711_recommendation) · Human approve (decision_loan_84711_approval)
RESPONSIBILITY
13
7 typed entries × 2 decisions (with 1 deduped OVERSIGHT)
LOSS
1
loss_loan_84711_default · $4.2M · credit_default
CLAIM
1
claim_84711 — filed with reinsurer

state_id = 6d540fc52dc0a4b7a593e38494c8bca6b6a3d8f5b64678f7acb86e7574e3748c

03show_loss_lineage(claim_84711) — 23 typed entries

The killer query · returned from one call
asl> show_loss_lineage(claim_84711)
→ 23 entries spanning all 7 layers + 7 typed responsibility sub-layers
CLAIM
claim_84711
tenant_root (insured)
ASSET
loan_84711_borrower_principal
DECISION
decision_loan_84711_approval
princ_credit_officer_001
DECISION
decision_loan_84711_recommendation
princ_credit_risk_agent_001
MODEL
model_credit_underwriting:1
princ_model_owner_…
MODEL
model_credit_underwriting:1
princ_model_owner_…
POLICY
pol_credit_underwriting:1
princ_policy_owner_underwriting
POLICY
pol_credit_underwriting:1
princ_policy_owner_underwriting
AUTHORITY
root → credit_officer
princ_credit_officer_001
AUTHORITY
credit_officer → AI agent
princ_credit_risk_agent_001
RESPONSIBILITY:APPROVAL
decision_loan_84711_approval
princ_credit_officer_001
RESPONSIBILITY:EXECUTION × 2
both decisions
credit_officer + AI agent
RESPONSIBILITY:SUPERVISION × 2
both decisions
root + credit_officer
RESPONSIBILITY:POLICY × 2
both decisions
policy_owner
RESPONSIBILITY:MODEL × 2
both decisions
model_owner
RESPONSIBILITY:ASSET × 2
both decisions
princ_borrower_loan_84711
RESPONSIBILITY:OVERSIGHT × 2
both decisions
root + credit_officer

One query. Full chain. Five years later. This is the substrate the post-#15 demo show_loss_lineage(claim_84711) rides on — a single search that returns Asset / Decision / Model / Policy / Authority / Responsibility / Loss with time-travel replay AND independent verification.

04What the chain reads

In plain language, here is what the engine reconstructs from the canonical event log alone:

LayerWhat the chain says
Claimclaim_84711 filed by the CFO with the credit reinsurer
Lossloss_loan_84711_default — credit_default — $4.2M USD
Affected assetloan_84711_borrower_principal
Triggering decisionsapproval (human) consumed recommendation (AI)
Model at decision timemodel_credit_underwriting v1, weight_hash + training_fingerprint preserved
Policy in force at decisionpol_credit_underwriting v1 (DTI ≤ 0.45)
Policy amended laterpol_credit_underwriting v2 (DTI ≤ 0.36) — too late for the loan
Authority chainroot → credit officer → AI risk agent
Responsibility at decision timeactor=credit_officer · model_owner · policy_owner · asset_owner=borrower · delegated_from=root

05The schema

LossEvent {
    at_ms, loss_id, affected_asset,
    loss_amount, loss_category,         // credit_default · operational_loss · fraud · impairment · market_loss
    triggering_decision_ids: Vec<String>,
    recognized_by, signature
}

ClaimEvent {
    at_ms, claim_id, loss_event_id,
    claim_amount, claim_status,         // filed · investigating · approved · denied · paid
    insurer, insured, filed_by, signature
}

Snapshot extensions: loss_events, claim_events, loss_lineage_index — all skip-if-empty. All 14 prior proofs' state_ids verified byte-identical.

06What this proof IS and IS NOT

This proof IS

The convergence proof. The substrate the post-#15 killer demo (show_loss_lineage(claim_84711) rendered as a one-click search) rides on. The point in the corpus where H33 stops being "accountability infrastructure" and starts being Replayable Consequences. The proof insurers, reinsurers, claim investigators, SIU teams, regulators, and PE diligence teams recognize immediately as their core problem solved.

This proof IS NOT

A claim that any real carrier has migrated yet (operator-side). A complete reserving / settlement engine (this is the evidentiary chain). A legal verdict (chain is structural fact). A claim that decisions producing losses are rejected at issuance (policy-layer concern).

07Known limitations

  1. Loss + claim amounts carried as opaque strings — confidentiality-friendly + engine doesn't arithmetic. Typed money fields are a future iteration.
  2. affected_asset is a string reference today. Proof #18 (Asset Lineage) formalizes Asset as a typed entity.
  3. One loss → one claim in this scenario. Multi-claim chains (reinsurance / retrocession) queued for #15.1.
  4. consequence_chain (forward-looking from Decision to downstream consequences) is engine-side per Eric's "quietly add, not on page" directive.
  5. Phase E lock open as in every prior proof; L9.1 closes it.

08Where this proof sits

#11
First Time Travel Replay (L5). first-time-travel-replay
proven
#12
First Independent Replay (L9, the moat). first-independent-replay
proven
#12.1
First Catastrophic Vendor Failure. first-catastrophic-vendor-failure
proven
#13
First Replayable Enterprise (the category). first-replayable-enterprise
proven
#14
First Authority Lifecycle Replay (foundation of accountability). first-authority-lifecycle-replay
proven
#14.1
First Replayable Responsibility (accountability infrastructure). first-replayable-responsibility
proven
#15
First Replayable Insurance Claim (the convergence proof) — 7 abstractions converge; show_loss_lineage returns the full chain. This proof.
proven now
demo
Post-#15 killer demo — show_loss_lineage(claim_84711) rendered as a one-click search with time-travel + independent verification.
next horizon
#16
First Enterprise Acquisition Replay
roadmap
#18
First Asset Lineage (formalizes Asset object)
roadmap
#19
First Regulator Mode (surfaces loss_lineage_index as portal output)
roadmap

09Evidence appendix

FieldValue
state_id6d540fc52dc0a4b7a593e38494c8bca6b6a3d8f5b64678f7acb86e7574e3748c
Tenant IDtenant_insurance_claim_44962d9b-25f5-5622-bd9a-98d5580bb8a2
Tenant root (insured)princ_root_claim_44962d9b-…
Credit Officerprinc_credit_officer_001
AI Risk Agentprinc_credit_risk_agent_001
Policy Ownerprinc_policy_owner_underwriting
Model Ownerprinc_model_owner_credit_underwriting
Borrower (asset_owner)princ_borrower_loan_84711
Reinsurerprinc_insurer_global_credit_re_001
Loss eventloss_loan_84711_default · $4.2M · credit_default
Claim eventclaim_84711 · $4.2M · filed
Event count10
responsibility_index entries13
loss_lineage_index entries23 (all 7 layers + 7 typed responsibility sub-layers)
Replay T (5yr post-claim)≈ March 2031
Reconstruction artifactreconstruction.json
Harnesstests/replayable_insurance_claim_001.rs (scif-backend @ 09b541109)

10Readiness determination

Determination

First Replayable Insurance Claim: PROVEN IN OPERATION for one insurance-claim tenant, 10 signed events, all 7 abstractions converged, 23 typed loss_lineage_index entries, the killer query show_loss_lineage(claim_84711) answers in one call.

What this unlocks: the post-#15 demo show_loss_lineage(claim_84711) (the artifact #15 makes possible — lead asset for insurer / PE / reinsurer / litigator / board conversations); Decision Insurance roadmap items become underwritable; Asset Lineage (#18) typifies the asset object; Regulator Mode (#19) surfaces the index as portal output. The category claim Eric named as the corpus endpoint: Replayable Consequences.

What this does not unlock: a complete reserving/settlement engine; a claim that any real carrier has migrated; a legal verdict.

Issued by H33, Inc. · Eric Beans, CEO · 2026-06-02

Independently reconstructable. Inputs: scif-backend @ 09b541109 · tests/replayable_insurance_claim_001.rs · reconstruction.json.