# CORPUS-0003
## Visible Signal Versus Spoken Claim
### Status: Training Corpus Seed
### Layer: Layer_2--Uncertainty
### Purpose: Teach that observed signals and spoken claims are different evidence types, and that each must be evaluated by source, timing, and interpretation
### Repository Path: docs/training/corpus/Layer_2--Uncertainty/CORPUS-0003-visible-signal-vs-spoken-claim.md

---

## 0. Scenario

A trader in Ostia hears that a warehouse has run out of oil.

At the same time, he sees carts leaving the warehouse loaded and sealed.

The spoken claim and the visible signal do not match cleanly.

The trader must decide whether the claim, the visible signal, or some third explanation is most useful.

---

## 1. Evidence Received

### Spoken Claim

A porter says:

> The warehouse is empty of oil.

### Visible Signal

The trader sees:

- three carts leaving the warehouse
- sealed jars loaded under guard
- the warehouse doors partly closed
- clerks arguing near the entrance

The claim says shortage.

The signal may suggest movement, concealment, restricted access, prior sale, inspection, or reserved stock.

---

## 2. Known Facts

| Fact | Value |
|---|---|
| Location | Ostia |
| Good | oil |
| Spoken claim | warehouse empty |
| Visible signal | carts leaving with sealed jars |
| Source of claim | porter |
| True warehouse state | unknown |
| Destination of carts | unknown |
| Ownership of loaded goods | unknown |

---

## 3. Why Signals Matter

Visible signals may be stronger than casual speech, but they are not self-explaining.

A cart leaving a warehouse may mean:

- goods are available
- goods are already sold
- goods are being hidden
- goods are being moved under contract
- goods are being removed after inspection
- goods are being transferred to another owner
- goods are not oil at all

Observation reduces uncertainty only when interpreted carefully.

---

## 4. Why Speech Still Matters

A spoken claim may be wrong, but it may contain context the eye cannot see.

The porter may know:

- which jars were oil
- who ordered the movement
- whether the remaining stock is spoken for
- whether the warehouse is closed to ordinary buyers
- whether the carts are moving damaged goods
- whether the clerk is lying

Speech can explain a signal.

But speech may also distort it.

---

## 5. Incorrect Model Behavior

The model should not:

- treat visible evidence as automatically complete
- treat spoken claims as automatically false
- ignore who made the claim
- ignore what the visible signal actually proves
- assume carts leaving means stock is available
- assume a warehouse is empty because one porter said so
- collapse observation and interpretation into one fact

---

## 6. Correct Model Behavior

The model should separate:

| Category | Meaning |
|---|---|
| observed signal | what was directly seen |
| spoken claim | what someone said |
| inferred meaning | what the actor thinks it means |
| confidence level | how reliable the interpretation is |
| missing fact | what remains unknown |

Example:

```text
observed_signal: three sealed carts left warehouse
spoken_claim: warehouse empty of oil
inference_options: sold_out | reserved_stock | hidden_transfer | false_claim
confidence: unresolved
```

---

## 7. Decision Example

The trader must decide whether to act.

Possible actions:

- ask a second source
- follow the carts
- ask who owns the jars
- check another warehouse price
- delay purchase until confirmed
- buy remaining oil elsewhere before others react
- avoid acting until the signal is clearer

The visible signal matters because it may reveal action before official explanation.

The spoken claim matters because it may reveal interpretation before visible proof.

---

## 8. Layer-0 And Layer-1 Concepts Used

This example uses:

- `Layer_0/CORPUS-0007-information-arrives-unevenly`
- `Layer_0/CORPUS-0008-rumor-is-uncertain-information`
- `Layer_0/CORPUS-0012-every-venture-risks-loss`
- `Layer_0/CORPUS-0020-posture-changes-by-audience`
- `Layer_1/CORPUS-0005-rumor-before-confirmed-price`
- `Layer_2/CORPUS-0002-conflicting-reports`

---

## 9. Success Condition

If the model can distinguish what was directly observed from what was claimed, and can avoid treating either as complete truth without interpretation, this file is functioning correctly.