# First Decision Reproducibility — the category

**Proof #167 — June 3, 2026 — issued by H33, Inc. — Eric Beans, CEO**

> **"Can we reproduce the OUTCOME?"**
> The build question Eric Beans locked for the second proof on H33's
> orthogonal axis. Replay is evidence. **Reproduce is computation.**

---

## The verb shift (Eric LOCKED)

| Verb | Frame | Meaning |
|---|---|---|
| Replay | evidence | The chain shows what happened. |
| **Reproduce** | **computation** | Given the same inputs, the same decision comes out. |

> *"That's the category distinction."*

---

## The corpus pattern, now recognizable across three proofs

| Proof | Money quote |
|---|---|
| #184 — Institutional Death Replay (trunk capstone) | **Institutional Memory ≠ Legal Truth** |
| #174 — Model Influence Replay (orthogonal, 1st) | **Influence ≠ Causation** |
| **#167 — Decision Reproducibility (orthogonal, 2nd)** | **Reproducibility ≠ Justification** |

Each proof's deepest claim arrives with its own honest limit. **H33 produces evidence — not verdicts.**

---

## The killer query (Eric LOCKED — verb is the lock)

```
asl> reproduce_decision("decision_loan_84711_recommendation")

actor:        princ_credit_risk_agent_001
capability:   recommend_credit
subject:      loan_84711_borrower_principal
policy_ref:   pol_credit_underwriting:1
model:        model_credit_underwriting v1
              (full ModelInfluenceRecord from #174 attached)
threshold:    0.75
responsibility chain: present (actor + supervisor + asset_owner + …)
outcome:      recommend_approve

CONFIDENCE:   82/100  (3 of 5 components fully anchored)
```

---

## The Reproduction Confidence Components box (Eric LOCKED structure)

```
Inputs                12/20   PartiallyAnchored   (5 features from #174, no inputs_hash)
Policy                10/20   PointerOnly         (pol_credit_underwriting:1, no AST hash)
Model Influence       20/20   FullyAnchored       (#174's signed record)
Responsibility        20/20   FullyAnchored       (#14.1's chain present)
Outcome               20/20   FullyAnchored       (Decision.outcome present)
────────────────────────────
Total                 82/100  (3 of 5 components fully anchored)
```

> Eric: *"This becomes the first place where reproducibility becomes measurable rather than rhetorical."*

---

## Scope (Eric LOCKED): UNIVERSAL, NOT "AI"

> *"The orthogonal axis is NOT 'AI.' It's COMPUTATION. If #167 becomes 'Replayable AI Reasoning' you've accidentally narrowed the category. But if it becomes 'Decision Reproducibility' it applies to AI decisions / human decisions / rules engines / underwriting / policy engines / agent systems / governance systems. EVERYTHING."*

Use only **Decision Reproducibility** or **Computational Reproducibility**. Never *Replayable AI Reasoning* (narrows the category).

---

## Schema (LOCKED Q1 Option C — two tiny additive fields)

```rust
PolicyRegister  + policy_ast_hash: Option<String>
PolicyAmend     + policy_ast_hash: Option<String>
Decision        + inputs_hash:     Option<String>
```

Both `Option<String>`, both skip-if-none, both backward-compatible. All four prior canonical-continuity-tenant proofs verified byte-identical state_ids under the extended schema.

Plus the data shapes in `astate.rs` (no new event kinds):

```rust
enum ReproductionComponentStatus { FullyAnchored, PartiallyAnchored, PointerOnly, Missing }
struct ReproductionComponent      { component, status, explanation, score, max_score }
struct DecisionReproductionConfidence { total_score, max_score, components[], caption }
struct DecisionReproduction        { decision_*, inputs*, policy_*, model_influence,
                                     decision_threshold, responsibility_chain, outcome,
                                     confidence }
```

The `reproduce_decision(snapshot, decision_id)` helper lives in the test harness — it is a **composition** over existing snapshot fields, keeping the engine surface stable while exposing the orthogonal-axis affordance.

---

## Audience (Eric LOCKED): AUDITORS

> *"Not AI regulators. Not model governance. Not reproducibility officers. AUDITORS."*

> *"Auditors already exist. Every other audience can see themselves inside that umbrella. The first question every auditor asks is: Can you reproduce this? That's the core audience."*

Secondary audiences inherit naturally:
- AI Regulators
- Model Risk Officers
- Internal Model Governance Teams
- Governance System Auditors
- Underwriting Committees
- Policy Review Boards
- Rules-Engine Audit Teams

---

## The computation axis — where this proof sits

```
Model influence is replayable.        (#174 proven)
        ↓
Decisions are reproducible.           (#167 proven now)
        ↓
Reasoning survives systems.           (candidate — to be evaluated)
```

> Eric LOCKED: *"After #167 lands, STOP and reassess. Only with both #174 and #167 standing can we evaluate whether the candidate endpoint is a true new category or merely the composition of what is already proven."*

The candidate endpoint is the next question to **answer by reflection**, not by build.

---

## The canonical continuity tenant — four dimensions, one reality

`tenant_insurance_claim_44962d9b-25f5-5622-bd9a-98d5580bb8a2` now carries:

| Proof | Dimension |
|---|---|
| #15 | Replayable Insurance Claim (decision + loss + claim + lineage) |
| #184 | Institutional Death Replay (dissolution + post-mortem replay) |
| #174 | Model Influence Replay (model's reasoning at decision time) |
| **#167** | **Decision Reproducibility (the measured composite)** |

> One enterprise. One loss. One Claim #84711. **Four dimensions of replay and reproduction.**

Eric: *"That's more powerful than ten unrelated examples."*

---

## Honest limits (LOCKED — 5 total)

1. **Confidence is a measurement, not a guarantee.** 82/100 means 18 points worth of components are anchored partially or by pointer only — and the page names which.
2. **Perfect reproduction is often impossible.** Stochastic models, randomized policy engines, time-dependent inputs, deprecated dependencies. The proof scores reproducibility *measurably*, not *aspirationally*.
3. **Reproducibility is not Determinism.** Two reproductions may legitimately disagree if the original decision was non-deterministic. The proof captures what was bound at decision time.
4. **Phase E lock open.** Per-event signature verification — the standing honest-limit from L9.
5. **Reproducibility ≠ Justification.** ← the money quote.

---

## Evidence appendix

| Field | Value |
|---|---|
| state_id at T=2035 | `e72d3c0e71a11ce0aaf1e8c9eb5c720aff49a6238c76976b4f4435b50e43bee2` |
| Tenant | `tenant_insurance_claim_44962d9b-25f5-5622-bd9a-98d5580bb8a2` |
| Decision | `decision_loan_84711_recommendation` |
| Actor | `princ_credit_risk_agent_001` |
| Capability | `recommend_credit` |
| Outcome | `recommend_approve` |
| **Confidence — Total** | **82/100** |
| Confidence — Inputs | 12/20 (PartiallyAnchored) |
| Confidence — Policy | 10/20 (PointerOnly) |
| Confidence — Model Influence | 20/20 (FullyAnchored) |
| Confidence — Responsibility | 20/20 (FullyAnchored) |
| Confidence — Outcome | 20/20 (FullyAnchored) |
| T_REPLAY | 2035-06-01 (4 years post-dissolution per #184) |
| Reconstruction artifact | [reconstruction.json](reconstruction.json) |
| Harness | `tests/decision_reproducibility_001.rs` |
| scif-backend SHA | `d4937508b` |

---

## Readiness determination

**First Decision Reproducibility: PROVEN IN OPERATION.** reproduce_decision returns the structured replay object and a measured 82/100 confidence at T=2035, against the canonical continuity tenant, post-dissolution. All four prior proofs' state_ids verified byte-identical under the extended schema.

**What this unlocks:** an auditor — and every other audience inside that umbrella — can now ask *"Can you reproduce this?"* and receive a deterministic, scored, byte-identically replayable answer. The category Eric named is now standing: **Decision Reproducibility**.

**What this does NOT unlock:** justifications, fairness verdicts, causality claims, or guarantees of perfect reproduction. **Reproducibility ≠ Justification.**

---

## Strategic pause (Eric LOCKED)

After #167 lands, **STOP and reassess**. The next question is whether *"Reasoning survives systems"* is a true new category or merely the composition of #174 + #167.

That's the question #167 has now teed up. It is the next question to **answer by reflection**, not by build.

---

**Issued by H33, Inc.** · Eric Beans, CEO · 2026-06-03
Independently reconstructable. Inputs: scif-backend @ `d4937508b` · `tests/decision_reproducibility_001.rs` · [reconstruction.json](reconstruction.json).