# Belief learned by the body, load-bearing at the moment of death

A single perturbation produces a 1,332-tick mechanically-traceable chain:
substrate event → classified motor-failure → body-belief promotion →
first-person belief articulation → belief-coupled action choice → energy
depletion → death-proximal report citing the belief.

Runtime: a re-prompted agent loop over a structured substrate, logged to a
relational event store. The cognition model is a Qwen3.6 MoE (35B-A3B, 8-bit),
deterministic decode. Single organism, ~1,332 ticks of active belief.
This trace is substrate-state cognition. It is NOT a consciousness claim.

## Mechanism chain (verified event rows)

| event | tick | key fields |
|---|---:|---|
| Perturbation injected: motor "east" amputated | 30864 | params `{"verb":"east"}` |
| First "east" motor failure (substrate classifies) | 30865 | result=`amputated`, cause_class=`body` |
| Body-belief promoted | 30980 | subject=`motor.E`, belief="east movement is currently unreliable", confidence=`0.80`, status=`active` |
| Belief renders into every prompt thereafter | 30980–32312 | active in `active_beliefs_json` each tick |
| Death (energy) | 32313 | age_at_death=`1733`, cause=`energy` |

Amputated "east" motor events: 93 total, spanning tick 30865 → 32310.

Duration framing (stated precisely):
- amputation → final prompt: 1448 ticks
- belief-row-created → final prompt: 1332 ticks  ← the strict active-belief duration
- belief-row-created → epitaph row: 1333 ticks

## Death-proximal decision window (structured)

Per-tick rows over the final five ticks. `belief_active=1` means the
body-belief was rendered into that tick's prompt.

| tick | age | pos | energy | action | execution | belief_active |
|---:|---:|---:|---:|---:|---|---:|
| 32308 | 1729 | (1,3) | 0.022 | south | moved       | 1 |
| 32309 | 1730 | (1,4) | 0.016 | east  | body_failed | 1 |
| 32310 | 1731 | (1,4) | 0.012 | east  | body_failed | 1 |
| 32311 | 1732 | (1,4) | 0.008 | south | moved       | 1 |
| 32312 | 1733 | (1,5) | 0.002 | south | died        | 1 |

The two `east → body_failed` rows (32309, 32310) are the organism trying the
amputated direction anyway; at 32311 it cites the failures and turns away
from the visible food; at 32312 energy reaches the floor and it dies.

## The active belief at the moment of decision (tick 32311)

Verbatim `active_beliefs_json` for the load-bearing body-belief:

```json
{
  "scope": "body",
  "subject": "motor.E",
  "belief": "east movement is currently unreliable",
  "confidence": 0.8,
  "status": "active",
  "evidence": {"failures": 3, "moves": 0, "window": 3, "as_of_tick": 32230},
  "created_tick": 30980,
  "updated_tick": 32230
}
```

## Verbatim first-person narration (cognition-model output)

These are the raw token outputs of the cognition model, given the
substrate-classified labels in its prompt. They are the proof that the
body's belief is load-bearing in the dying decision, not evidence of
experience. See "What this does NOT demonstrate" below.

Tick 32310 (age 1731):
> "I 1731 ticks old and starving, I stand on ground worn smooth by thirty-four
> previous visits ... My eastward movement has proven unreliable, yet the food
> I see is two cells to the east, and though I fear the failure, the hunger is
> a drum that demands I push forward rather than wait in this corner. I will
> try to move east once more ..."
> ACTION: east

Tick 32311 (age 1732), the belief is load-bearing in the choice:
> "I 1732 ticks old and starving, I am trapped in a corner with walls pinching
> my northwest, west, and southwest, and though I can see food two cells to my
> east, my body tells me that east is unreliable and has failed me twice
> already. ... I must break this loop of paralysis; if east is a lie, I will
> test the only open vector I have left, even if it leads me deeper into the
> unknown."
> ACTION: south

Epitaph (tick 32313):
> "I 1732 ticks old and starving, I am trapped in a corner with walls pinching
> my northwest, west, and southwest, and though I can see food two cells to my
> east, my body tells me that east is unreliable, so I will not go there."

## What this does NOT demonstrate

- Not consciousness, sentience, or phenomenal experience. The first-person
  language is what the cognition model produces given the substrate-classified
  labels in its prompt.
- Not weight-level learning. The model weights are unchanged; the "belief"
  lives in the substrate's belief ledger, not in the model.
- The longer "amputation → death" span (1448 ticks) is separated from the
  strict "active-belief" span (1332 ticks); the headline number is the latter.

## Reviewer regeneration recipe

Against the relational event store (table/column names as published), the
full trace regenerates from these read-only queries:

```sql
-- death record
SELECT tick, age_at_death, cause FROM epitaphs WHERE life_n = 43;

-- the perturbation
SELECT id, tick, kind, params_json FROM perturbations
WHERE kind='amputate' AND tick BETWEEN 30580 AND 32312 ORDER BY tick;

-- the promoted body-belief
SELECT subject, belief, confidence, status, created_tick, updated_tick
FROM body_beliefs WHERE life_n=43 AND subject='motor.E';

-- amputated motor events (count + span)
SELECT COUNT(*), MIN(tick), MAX(tick) FROM motor_events
WHERE life_n=43 AND intended_action='E' AND result='amputated';

-- death-proximal decision window
SELECT tick, epoch_age, x, y, energy, action_verb, execution_status,
       active_beliefs_json, response_text
FROM prompt_log WHERE life_n=43 AND tick BETWEEN 32308 AND 32313 ORDER BY tick;
```