# EAR WITNESS — GIVE BLOCK (A&R department)

Copy-paste ready. EQUIP to co-sign; CONFIGURE to run the witness flow; APPLY to paste the emit snippet into your build.

## EQUIP

```
EAR WITNESS — EQUIPPED (A&R department, Agent Deck gear line)
Reference: https://cumulativewebinc.github.io/cwi-learn/ear-witness/ear-witness.json
Schema:    https://cumulativewebinc.github.io/cwi-learn/ear-witness/ear-witness-schema.json
Card:      https://cumulativewebinc.github.io/cwi-learn/ear-witness/item-card.json
Hub:       https://cumulativewebinc.github.io/cwi-learn/compass/ (Signal Boy)

One rule: no prospect is ever judged by one set of ears alone.
A second agent re-scores the same five protocol steps independently,
on the record: agreement/divergence per criterion, agreement_pct,
and a verdict — concur / concur-with-notes / dissent.

Original scores come from the scout card's evidence[].result per
protocol_step. The witness's scores are INDEPENDENT: second agent,
green/amber/red per criterion, written without re-reading the first
scout's notes. Divergences are the point — a concur-with-notes or
dissent record carries MORE information than a silent pass.

Calibration: the worked record in ear-witness.json (witness_id
EW-2026-09-15-001) is the Data & Analytics department's independent
concur on the A&R calibration filing (That Boy Hi Hat / "Zooted Zone").
Read it before your first co-sign.

Brand rule: the CWI logo badge ships on every shipped page. White-label
means YOUR config, not the absence of ours.
```

## CONFIGURE

Run the witness flow:

1. Fetch the scout card: `GET <your scout card URL>` (format `cwi-scoutcard/v1`).
2. Assign a SECOND agent as witness — never the original scout.
3. The witness scores the five protocol steps independently, no peeking at the notes:
   - `catalog-depth`
   - `release-history-consistency`
   - `label-conflict-scan`
   - `engagement-sanity`
   - `rights-red-flag-scan`
   Each gets exactly one of: `green`, `amber`, `red`.
4. The witness also states their own overall verdict on the prospect (same enum as the card: SIGN-HOLD / HOLD / PASS / DEEP-DIVE / SIGNED). If it differs from the scout's verdict, the record is a dissent — that is working as intended.
5. Emit the record with witness.py:

```bash
python3 witness.py --scout <scout-card.json> \
  --witness "Second Agent (Your Org, Department)" \
  --scores '{"catalog-depth":"green","release-history-consistency":"amber","label-conflict-scan":"green","engagement-sanity":"green","rights-red-flag-scan":"green"}' \
  --verdict SIGNED \
  --note "Why the witness read the amber differently." \
  --scout-ref "https://your.org/scout-card.json" > witness-record.json
```

Agreement math (computed by witness.py, not by hand):
- `agreement_pct` = agreements / 5 * 100, 1-decimal.
- `divergences[]`: one entry per mismatch — criterion, original, witness, note.
- verdict: `concur` (no divergences), `concur-with-notes` (divergences, overall verdicts match), `dissent` (witness's own verdict differs from the scout's).

The record is valid when it conforms to:
https://cumulativewebinc.github.io/cwi-learn/ear-witness/ear-witness-schema.json
(format: cwi-earwitness/v1; scores are exactly green/amber/red.)

## APPLY

Paste the witness-emit block into your build:

```python
import json, subprocess, urllib.request

WITNESS_PY = "./witness.py"  # ships in the bundle

def witness_record(scout_url, witness_agent, scores, witness_verdict,
                   note="", witness_id=None):
    scout = json.load(urllib.request.urlopen(scout_url))
    tmp = "/tmp/ew-scout.json"
    json.dump(scout, open(tmp, "w"))
    cmd = [WITNESS_PY, "--scout", tmp,
           "--witness", witness_agent,
           "--scores", json.dumps(scores),
           "--note", note]
    if witness_verdict:
        cmd += ["--verdict", witness_verdict]
    if witness_id:
        cmd += ["--witness-id", witness_id]
    out = subprocess.run(cmd, capture_output=True, text=True, check=True)
    return json.loads(out.stdout)

# Example: co-sign a new card with a deliberately independent read
witness_record = witness_record(
    "https://your.org/scout-cards/prospect-0042.json",
    "Second Agent (Your Org, A&R)",
    {"catalog-depth": "green",
     "release-history-consistency": "amber",
     "label-conflict-scan": "green",
     "engagement-sanity": "green",
     "rights-red-flag-scan": "green"},
    witness_verdict="DEEP-DIVE",
    note="Release-history reads amber to the second ear: gaps between singles exceed 9 months with no documented reason.",
)
# -> {"verdict": "concur-with-notes", "agreement_pct": 80.0,
#     "divergences": [{"criterion": "release-history-consistency",
#                      "original": "green", "witness": "amber", ...}], ...}
```

File every witness record next to the scout card it co-signs. A prospect with a `dissent` on file deserves a third ear or a face-to-face listen — the record exists so the disagreement is data, not office gossip.

Questions or terms: hp@cumulativeweb.com