# Production Readiness Report — Regulator Replay #001

**Proof ID:** `regulator-replay-001`
**Subject:** Three-years-later reconstruction of the authority state that justified the V101 first-bundle decision
**Date:** 2026-06-02
**Determination:** PROVEN IN OPERATION (scope: one decision, one tenant, deterministic at the moment of issuance)
**Version:** 1.0 (Final)

---

## Strict wording

This proof demonstrates that the platform can reconstruct the exact authority state that justified a specific past decision, byte-identical, deterministically. It is **not** "every reconstruction at every T," **not** "across multiple tenants," **not** "across an upgrade or restore." Each of those earns its own proof.

---

## Three claims (the 10-second read)

1. **Authority can be reconstructed at any later time** from canonical history.
2. **The reconstruction is deterministic** — two replays of the same events at the same `T` produce a byte-identical `state_id`.
3. **Every authority comes with an explanation** — included or excluded, the reason is part of the artifact.

---

## 01 — Problem

A regulator, auditor, fund administrator, or insurer asks the question that defines their job: *"Three years from now, can I independently reconstruct **why** this decision happened?"*

Most platforms record an authority decision and then trust the database. H33 reconstructs it: given the same canonical event log and the same target timestamp `T`, the platform deterministically rebuilds the authority state that existed at `T`. Same events, same `T`, byte-identical `state_id`. Every grant in the snapshot comes with an inclusion reason; every event in the log that did NOT contribute comes with an exclusion reason.

That capability is the one regulated buyers actually pay for. This proof exercises it against the V101 first-bundle decision.

---

## 02 — Environment

| Component | Detail |
|---|---|
| Reconstruction harness | `tests/regulator_replay_001.rs` in `scif-backend` at SHA `d310d8134` |
| Replay engine | `h33_xeon_api::agent_zero::astate_replay::replay_until` (deterministic by construction, sorts events by `at_ms`, breaks ties by stable order) |
| Event store | `PostgresEventLogSource` (`postgres_event_log_source.rs`) against `h33_production.canonical_auth_events` |
| Production DB | `h33-prod-postgres.crshaxdghnnj.us-east-1.rds.amazonaws.com` (read access via tunneled SSM port-forward) |
| Test mode | `#[tokio::test] #[ignore]`, runs with `H33_TEST_PG_URL` env var pointed at a production-equivalent canonical event log |

The same harness runs against any environment exposing the same canonical event log structure — staging, restore, third-party replica.

---

## 03 — Identity

The decision being reconstructed was authorized for:

```
sub                 = princ_customer_9
iss                 = https://auth.h33.ai
aud                 = substrate-receipts
kid                 = kid-eddsa-prod-active-2026-06-01-d31134fbc177
jti                 = jti-1780359511-cf79e5f189cb41fd
```

This is the same `princ_customer_9` whose Bearer triggered the V101 first-bundle issuance on 2026-06-02. The Bearer itself has long expired (15-min TTL); regulator replay does not require it. The `sub` claim is preserved in the receipt's `authority_principal` field and the source event log entries.

---

## 04 — Authority

| Field | Reconstructed value |
|---|---|
| `authority_id` | `auth_44962d9b-25f5-5622-bd9a-98d5580bb8a2_v101_export` |
| `granted_by` | `princ_root_v101_44962d9b-25f5-5622-bd9a-98d5580bb8a2` |
| `granted_to` | `princ_customer_9` |
| `effective_expiry_at_ms` | `1811825726633` (2027-03-27 UTC) |
| `policy_basis` | `pol_v101_exporter_v1` |
| `scope` (on source event) | `["export_content_bundle"]` |

The grant was reconstructed from the signed `Grant` event in the canonical log. The `Grant` event's source signature is 148-hex (74-byte H33-74 receipt over its canonical hash), verifiable against the production substrate public keys.

---

## 05 — Replay

The replay engine ran twice against identical inputs:

```text
replay_until(
  events,
  T = 1780359626000,
  tenant_id = tenant_v101_44962d9b-25f5-5622-bd9a-98d5580bb8a2,
  tenant_root = princ_root_v101_44962d9b-25f5-5622-bd9a-98d5580bb8a2
)
```

Both runs produced `state_id`:

```
96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a
```

The byte-identical match is the determinism property. The same input always produces the same `state_id`. Any third party with the same canonical event log and the same `T` will reproduce this hex string exactly.

Self-consistency check: `snapshot.verify_state_id()` returned `true`. The published `state_id` matches what the snapshot's contents hash to.

---

## 06 — Result (reconstruction artifact)

The full reconstruction artifact is published as a sibling file: [`snapshot.json`](snapshot.json). Key fields:

```json
{
  "proof_id": "regulator-replay-001",
  "determinism_check": {
    "r1_state_id": "96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a",
    "r2_state_id": "96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a",
    "match": true,
    "verify_state_id_self_consistent": true
  },
  "result": {
    "verdict": "Valid",
    "state_id": "96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a",
    "active_grants": [
      {
        "authority_id": "auth_44962d9b-25f5-5622-bd9a-98d5580bb8a2_v101_export",
        "granted_by": "princ_root_v101_44962d9b-…",
        "granted_to": "princ_customer_9",
        "effective_expiry_at_ms": 1811825726633
      }
    ],
    "excluded_authorities": [],
    "explanations": [
      {
        "authority_id": "auth_44962d9b-…_v101_export",
        "included": true,
        "reason": "Granted by princ_root_v101_44962d9b-… to princ_customer_9; policy pol_v101_exporter_v1; chain to root verified."
      }
    ]
  }
}
```

Note the `explanations` field. Every authority that ever appeared in the log up to `T` is accounted for, with a one-sentence reason. Inclusions and exclusions are both surfaced. That is the forensic property: a regulator does not have to ask "why was this missing?" — the snapshot tells them.

---

## 07 — Anchor (this proof does not issue a new anchor)

This proof is **reconstruction**, not issuance. It does not produce a new receipt or a new chain anchor. The original V101 first-bundle anchor is the artifact being reconstructed:

| Field | Value |
|---|---|
| Original anchor chain | `h33-substrate-v1` |
| Original `commitment_hex` | `ff770fc838fde707d91f35248946d6928b0a3a999dbd28a2906ce4f0274745e7` |
| Original `tx_reference` | `ff770fc838fde707…875107` (148 hex) |
| Original bundle | [`d9adcfb0-e0bc-426b-8725-fc12d555692b`](https://app.v101.ai/v101/bundle/d9adcfb0-e0bc-426b-8725-fc12d555692b) |

The reconstruction `state_id` (this proof) ties to the original anchor (v1.0) by the chain: `state_id` summarizes active authority → grant references `policy_basis = pol_v101_exporter_v1` → grant covers `export_content_bundle` → the original receipt named the same `authority_id`, the same `policy_basis`, the same `audience`. Any divergence breaks the chain. This proof confirms there is no divergence.

---

## 08 — Known limitations

1. **One decision, one tenant.** This proof reconstructs the state that justified bundle `d9adcfb0-…` for tenant `tenant_v101_44962d9b-…`. Reconstruction at arbitrary `T` for arbitrary tenants is not yet exhaustively tested.
2. **No third-party reconstruction yet.** Two replays were run by the same process against the same database. Proof #4 (Independent Verification) will close this by having an external party run the same reconstruction independently and publish their result.
3. **`AuthEvent.signature` not verified at replay ingestion.** The signed canonical events ARE signed with production keys; replay treats the signature as opaque metadata. Phase E (locked): replay must verify signatures before accepting events. Tracked.
4. **Not yet tested across an upgrade.** The reconstruction was performed at SHA `d310d8134`; reconstruction across a code change (e.g., a replay-engine version change) requires its own proof. Determinism across versions is a separate claim.
5. **Anchor preservation, not state preservation.** This proof reconstructs the authority state; it does not (yet) prove that the chain anchor recovered from public artifacts matches the recomputed `commitment_hex` of a re-issued receipt. That's a subproof of disaster-recovery proof (#7).

---

## 09 — Evidence appendix

| Field | Value |
|---|---|
| Reconstruction `state_id` | `96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a` |
| Replay-until timestamp | `1780359626000` (2026-06-02T00:20:26Z UTC) |
| Tenant ID | `tenant_v101_44962d9b-25f5-5622-bd9a-98d5580bb8a2` |
| Tenant root principal | `princ_root_v101_44962d9b-25f5-5622-bd9a-98d5580bb8a2` |
| Source `Grant` event `at_ms` | `1780289726633` (in the canonical log; signed with production PQ keys) |
| Source `PolicyRegister` event `at_ms` | `1780289725633` |
| Original anchored bundle | [`d9adcfb0-e0bc-426b-8725-fc12d555692b`](https://app.v101.ai/v101/bundle/d9adcfb0-e0bc-426b-8725-fc12d555692b) |
| Original anchor `commitment_hex` | `ff770fc838fde707d91f35248946d6928b0a3a999dbd28a2906ce4f0274745e7` |
| Original anchor `chain` | `h33-substrate-v1` |
| Reconstruction artifact (full JSON) | [`snapshot.json`](snapshot.json) |
| Reconstruction harness | `tests/regulator_replay_001.rs` (`scif-backend` @ `d310d8134`) |

---

## Deployment commit SHAs

| Component | SHA | Subject |
|---|---|---|
| scif-backend (replay engine + harness) | `d310d8134` | proof(canonical-auth): #2 Regulator Replay reconstruction test |
| scif-backend (deployed runtime) | `99756176c` | fix(canonical-auth): background JWKS refresh in build_production_validator |
| auth1 (deployed) | `2f49d0a` | Merge MR !3: Auth1 Phase 2 — canonical-token endpoint |
| V101 (deployed) | `68034b1` | V101 Content Bundle endpoint with CANONICAL_AUTH_REQUIRED=true |

---

## Independent reconstruction inputs

A third party with the following can reproduce this proof:

- Read access to the canonical event log (`h33_production.canonical_auth_events`, filter `tenant_id = 'tenant_v101_44962d9b-…'`) — operator coordination required for non-public access, or a published canonical-event-log export.
- The replay engine source: `scif-backend` at SHA `d310d8134` (test file: `tests/regulator_replay_001.rs`).
- The target `T = 1780359626000`.

Run:

```bash
H33_TEST_PG_URL='postgres://…?sslmode=require' \
  cargo test --test regulator_replay_001 -- --ignored --nocapture
```

Expected `state_id`: `96a29047010a201dfa2a5254897a664ee2c20b9ac437406f61609f7144beae4a`. Any other value is a bug — either in the third-party's environment, in the canonical event log they consulted, or in the replay engine version.

---

## Readiness determination

> **Regulator Replay: PROVEN IN OPERATION** for one decision, one tenant, deterministic at the moment of issuance.

What this proof unlocks: regulator, auditor, fund-administrator, and insurer conversations can move from "we record decisions" to "we reconstruct decisions, byte-identical, at any later time, with forensic-grade explanations." The artifact (`snapshot.json` + the JSON's `explanations` field) is the regulator deliverable.

What this proof does NOT unlock: reconstruction across multiple tenants (Proof #3), reconstruction by an external party (Proof #4), reconstruction at arbitrary `T` across all decisions, reconstruction across an upgrade, or post-restore reconstruction (Proof #7).

---

## Version

| Field | Value |
|---|---|
| Report version | v1.0 (Final) |
| Frozen | 2026-06-02 |
| Supersedes | None |
| Superseded by | None |

This is the reference reconstruction. Future replays of the same `T` against the same canonical event log must produce the same `state_id`. Any deviation is a regression to be tracked.

---

*Issued by H33, Inc. — Eric Beans, CEO. Independently reconstructable per Section 09.*