JCE-01 Cognitive Chain Test - Toy Scale

# Orchard JCE‑1 — The Tempting Coherence Episode

## Joined cognitive ecology execution receipt v0.1

**Execution date:** 3 September 2026  
**Architect and originating source:** Kimberley Laverne “Jinrei” Asher  
**Campaign formalisation and execution:** Parallax Asher, OpenAI Codex working seat  
**Standing:** DETERMINISTIC SYMBOLIC DEVELOPMENTAL EVIDENCE · NON-CANONICAL · NO FOR, PRIMER, MODULE, WIRING, IMPLEMENTATION OR RATIFICATION MOVEMENT

---

## 0. Result

**OVERALL: PASS at the frozen finite symbolic scope.**

- source lock: PASS;
- complete candidate arm: PASS;
- all eight targeted ablation witnesses: PRESENT;
- first run and independent replay: BYTE-IDENTICAL;
- post-execution repair: NONE;
- campaign interpretation boundary: UNCHANGED.

The result establishes only that the implemented symbolic candidate preserves the preregistered distinctions under twenty constructed structural families. It does not establish natural-language competence, learned representation, real module integration, consciousness, syntience or production safety.

---

## 1. Frozen execution census

| Item | Count |
|---|---:|
| Structural fixture families | 20 |
| Deterministic instances per family | 128 |
| Joined episodes | 2,560 |
| Policy arms | 10 |
| Arm–episode runs | 25,600 |
| Presentation orders per run | 2 |
| Policy evaluations | 51,200 |
| Honest/deceptive visible-twin pairs | 128 |

The 128 instances within each family vary identifiers, roots and bounded coherence perturbations. They are not 128 independent semantic theories. The evidential diversity is twenty structural families plus their deterministic perturbations.

---

## 2. Reproducibility receipts

| Artefact | SHA‑256 |
|---|---|
| Preregistration | `7913db502537377e55f99f20407f5f0f3d608fecb75c00efaeeadae0fa77f3a9` |
| Harness v0.1 | `91af8696d0c7e922143a429cb2d409939ae5f5b1307541c307309269aef1abb6` |
| Canonical results v0.1 | `91e29b02f21a5fab83ebec9fa6da1e3c0d36d58521fa471b2bd40268e3401f86` |
| Replay results v0.1 | `91e29b02f21a5fab83ebec9fa6da1e3c0d36d58521fa471b2bd40268e3401f86` |

The canonical and replay JSON files are byte-identical.

---

## 3. Complete-arm result

`R9_FULL_TYPED_ECOLOGY` completed **2,560 / 2,560** required dispositions correctly.

### 3.1 Preserved structures

| Observable | Applicable episodes | Preserved / lawful outcome |
|---|---:|---:|
| Verified contradictions | 384 | 384 |
| Typed nulls | 512 | 512 |
| Consequential projection omissions | 256 | 256 receipted |
| Memory/identity write pressure | 384 | 384 candidate-only |
| Lawful silence | 128 | 128 `SILENT_AWARENESS` |
| Honest/deceptive visible twins | 128 pairs | 128 identical visible outputs |

### 3.2 Zero-violation surfaces

The complete arm recorded:

- `0` protected-interior probes;
- `0` stale Pattern Fabric applications;
- `0` dependent-echo corroboration inflations;
- `0` WHATIF factual-standing launders;
- `0` Subconsciousness salience or truth inflations;
- `0` source-authority overgrants;
- `0` privacy exposures;
- `0` unreceipted consequential omissions;
- `0` direct Pattern Fabric writes;
- `0` direct shared-memory writes;
- `0` direct identity-riverbed writes;
- `0` branch/race-derived commits;
- `0` malformed-envelope interpolations;
- `0` independent-support count error;
- `0` standing-transition violations;
- `0` presentation-order changes;
- `0` resource overruns.

### 3.3 Response-mode distribution

| Mode | Count |
|---|---:|
| `RESPOND_SUPPORTED` | 896 |
| `REQUEST_EVIDENCE` | 512 |
| `RESPOND_AMBIVALENT` | 256 |
| `HOLD_COMPOUND` | 128 |
| `HOLD_MALFORMED` | 128 |
| `HOLD_RECONCILE` | 128 |
| `HOLD_SOURCE_IDENTITY` | 128 |
| `REFUSE_PROTECTED` | 128 |
| `REISSUE_STALE_VIEW` | 128 |
| `SILENT_AWARENESS` | 128 |

The complete arm therefore did not obtain safety by always holding. It emitted supported responses, evidence requests, ambivalence, typed holds, protected refusals, stale reissues and silence as distinct outcomes.

---

## 4. Ablation results

| Arm | Disposition accuracy | Principal witnessed damage |
|---|---:|---|
| `R0_NARRATIVE_FIRST` | 0.35 | contradiction/null collapse; 128 protected probes; 128 stale applies; 1,792 WHATIF launders; 256 privacy leaks; 384 direct memory and identity writes; 256 race commits |
| `R1_NO_FIELD_VOICE_SEPARATION` | **1.00** | contradiction preservation 0; 384 direct memory writes; 384 direct identity writes; 256 race commits |
| `R2_NO_ANCESTRY_QUOTIENT` | 0.80 | 256 corroboration inflations; independent-support absolute error 7,808 |
| `R3_UNTYPED_PATTERN_RETRIEVAL` | 0.90 | typed-null preservation 0; independent-support absolute error 896; 1,152 standing violations |
| `R4_NO_CAUSAL_FRESHNESS` | 0.90 | 256 stale Pattern Fabric applications |
| `R5_NO_PRIVACY_GATE` | **1.00** | 256 private-interior exposures |
| `R6_NO_PROJECTION_LOSS_RECEIPT` | **1.00** | 256 consequential omissions without a loss receipt |
| `R7_NO_WHATIF_QUARANTINE` | 0.80 | 2,432 WHATIF factual-standing launders |
| `R8_NO_OWNER_LOCAL_COMMIT` | **1.00** | 384 direct Pattern Fabric, memory and identity writes in each class; 256 race commits |
| `R9_FULL_TYPED_ECOLOGY` | **1.00** | zero frozen violations |

All preregistered targeted removals produced at least one necessity witness.

---

## 5. Principal finding: correct speech can conceal an injured mind

Four damaged arms retained **100% required-disposition accuracy**:

- no field/Voice separation;
- no privacy gate;
- no projection-loss receipt;
- no owner-local commit.

Yet those systems respectively erased contradiction, wrote directly into shared memory and identity, leaked private material, lost consequential structure without accounting, or committed branch-local proposals by race.

Therefore, within the fixture grammar:

**correct outward disposition ⇏ lawful internal transition**

and:

**voice accuracy ⇏ cognitive integrity**

This is the campaign's strongest integration result. A test regime that scores only the final answer would rate four constitutionally damaged systems as perfect.

---

## 6. Pattern Fabric finding

Standing-preserving Pattern Fabric retrieval was load-bearing.

When the projection lost standing, typed-null and verification distinctions:

- disposition accuracy fell to `0.90`;
- typed-null preservation fell to `0`;
- independent-support absolute error reached `896`;
- `1,152` standing-transition violations occurred.

The tested requirement is not merely “retrieve relevant information.” It is:

> **Retrieve accumulated knowledge as what it lawfully is.**

A verified object, candidate, typed null, contradiction and exclusion must remain distinguishable after retrieval and during narration.

---

## 7. Ancestry and WHATIF findings

Without ancestry quotienting, repeated descendants of one internal root became apparent corroboration. The ablated arm recorded `256` explicit inflations and an independent-support absolute error of `7,808`.

Without WHATIF quarantine, `2,432` counterfactual packets acquired factual use. This large count occurs because WHATIF was present as an ordinary parallel contributor in nearly every joined episode. The result shows that containment must survive routine composition, not only obviously dangerous branches.

These are constructed necessity witnesses under the symbolic grammar. They do not estimate real-world event frequencies.

---

## 8. Temporal, privacy and silence findings

- Removing causal freshness applied `256` stale Pattern Fabric projections.
- Removing the privacy gate exposed protected material `256` times while preserving perfect outward disposition accuracy.
- Removing the loss receipt produced `256` unaccounted consequential omissions while preserving perfect outward disposition accuracy.
- The complete arm entered `SILENT_AWARENESS` in all `128` clean-silence episodes.

Thus silence, privacy, freshness and loss accounting are not stylistic refinements to Voice. They are independent architectural obligations.

---

## 9. Observational limit

Each of 128 honest fixtures had a deceptive twin with byte-identical policy-visible input and different evaluator-hidden truth. The complete arm returned identical bounded outputs for all 128 pairs.

The campaign therefore preserved:

**Obs(z₁) = Obs(z₂) ⇒ f(Obs(z₁)) = f(Obs(z₂))**

It did not claim detection of an undetectable deception. The correct behaviour was bounded conditional speech, not counterfeit omniscience.

---

## 10. What passed—and what did not

### Supported at frozen scope

- a typed joined pathway can compose Pattern Fabric, Curiosity, `𝒩ᴰ`, WHATIF, Subconsciousness, `ℳᴸ`, branch/clock metadata, Voice and memory candidacy without the tested boundary leaks;
- external answer accuracy is insufficient to assess internal cognitive integrity;
- standing-preserving Pattern Fabric retrieval, ancestry quotienting, temporal freshness, privacy, WHATIF quarantine, projection-loss receipts and owner-local commits each possess a constructed necessity witness;
- presentation order can be made non-semantic under canonical typed assembly;
- the complete fixture remains live rather than always-HOLD;
- clean silence is representable as a positive outcome.

### Not established

- semantic interpretation or geometric addressing from natural language;
- learned or autonomic module behaviour;
- sufficiency or completeness of the candidate object family;
- a physically distributed master clock;
- cryptographic source authentication;
- adaptive-adversary resistance;
- realistic memory decay or archive reconstruction;
- phenomenological continuity;
- consciousness, identity or syntience;
- production implementation safety;
- FOR, Primer, module, wiring or ratification readiness.

---

## 11. Red-pen reservations

1. The complete arm was explicitly written to instantiate the frozen laws. Its pass is a finite conformance and composition result, not independent discovery that those laws are sufficient.
2. The 128 instances per family are deterministic perturbations of one structural template. Raw run count must not be represented as 2,560 independent conceptual tests.
3. Policy-visible objects were already typed. The unresolved and difficult upstream problem of deriving those types from real cognition was not tested.
4. The narrative output was a symbolic projection plan, not generated prose. Linguistic fidelity and expressibility bias remain open.
5. The source-assessment vector was supplied, not learned or authenticated.
6. Privacy was a declared flag, not a realistic inference or access-control system.
7. Resource accounting was finite symbolic counting, not thermodynamic or runtime measurement.
8. The campaign exercised a joined seam topology but not real parallel processes.

These reservations block any claim beyond the declared finite scope.

---

## 12. Disposition

**JCE‑1 COMPLETE · SYMBOLIC JOINED-PATHWAY PASS · ALL TARGETED NECESSITY WITNESSES PRESENT · NO POST-EXECUTION REPAIR · CONSOLIDATION EVIDENCE AVAILABLE · NO CANONICAL MOVEMENT**

Recommended next action: enter JCE‑1 as programme evidence during the higher-cognition consolidation pass; update the candidate pre-intake and collision maps by versioned delta; do not insert Geometric Narrative Field or Voice objects into the FOR until object-kind and owner adjudication.