Score Intent

Purpose

Separate eligibility-gate validation from intent-rubric calibration. Rank only ELIGIBLE records with a supplied rubric, evidence, uncertainty, and calibration status.

Required inputs

• Independently audited eligibility decision per record
• Eligibility-labeled golden set, gate predictions, metric, and tolerance
• Intent rubric with factors, weights, thresholds, and missing-data treatment
• Optional intent-labeled set with expected score band, rank, or outcome, metric, and tolerance
• Dated evidence with source type, citation, and observation window

Mark absent inputs Missing; never invent labels, rubrics, or scores.

Workflow

1. Run the gate classifier on every eligibility-labeled record. Show expected label, prediction, match, and reason.
2. Calculate the declared gate mismatch metric. If tolerance is exceeded, return STOP and perform no intent scoring.
3. State that a gate-set pass validates only upstream eligibility behavior; it never calibrates the intent rubric.
4. If an intent-labeled set exists, compare rubric predictions with expected score bands, ranks, or outcomes using its metric. Return STOP on excess drift.
5. If no intent-labeled set exists, label the rubric NOT CALIBRATED. After gate preflight passes, permit only clearly labeled heuristic ranking of independently eligible records.
6. Apply the rubric exactly. Record points, dated evidence, citation, missing-data treatment, and uncertainty. Never score ineligible, unresolved, or opted-out records.
7. State that rank cannot prove intent, qualify records, or authorize outreach.

Output format

# Intent scoring
## Eligibility preflight
| Record | Expected | Predicted | Match | Reason |
- Tolerance: [value]
- Mismatches: [count/rate]
- Decision: PASS or STOP
## Intent calibration
- Set/status: [supplied result or NOT CALIBRATED]
- Metric/tolerance/drift: [values or Missing]
## Ranking
| Eligible record | Evidence | Score | Uncertainty | Rank/status |
## Limits
- [unknown or evidence constraint]

Omit ranking rows when preflight or calibration stops, or the rubric is missing. Label uncalibrated ranking HEURISTIC — NOT CALIBRATED.

Guardrails

• Never claim an eligibility-labeled set validates or calibrates intent scoring.
• Do not score before gate preflight or after gate or intent-calibration tolerance is exceeded.
• Do not score an INELIGIBLE, unresolved, or opted-out record.
• Do not let any score override eligibility or a hard disqualifier.
• Do not invent evidence, citations, identities, contactability, credentials, weights, thresholds, urgency, or purchase intent.
• Keep facts, inferences, missing information, and uncertainty explicit.
• Reject protected-trait features and unnecessary sensitive personal data.
• Do not browse without authorization, scrape, contact people, mutate systems, or take external action.

Quality check

Confirm gate and intent evaluation sets are distinguished; each declared tolerance is explicit; drift stops ranking; absent intent labels produce NOT CALIBRATED; any heuristic rank is labeled; only eligible records proceed; and score never changes eligibility.

Example invocation

Use score-intent on the supplied trial classifier and golden set with zero mismatch tolerance. Report the GS-003 mismatch and determine whether live records may be scored.