ECR-000003 Automated Analysis Pipeline

Quick Start

Verify only:

npm run verify

Dry-run everything:

npm run pipeline:dry

Normalize response files:

npm run normalize

Run comparisons only:

npm run compare:all

Run explainability only:

npm run explain:all
npm run reports:explainability

Generate summary reports only:

npm run reports:all

Run complete pipeline:

npm run pipeline

Run one experiment:

node scripts/run-ecr-000003.js --experiment EXP-002 --compare-all

Safe Recommended Sequence

1. npm run collect:exp003
2. npm run normalize:dry
3. npm run normalize
4. npm run normalize:verify
5. npm run normalize:certify
6. npm run pipeline:dry
7. Review normalization and completeness reports.
8. npm run pipeline
9. npm run explain:all
10. npm run reports:explainability
11. Review final readiness report.
12. Complete EDRs manually.

Phone Upload Workflow

1. Put phone-saved responses under: incoming-responses/gpt/ incoming-responses/claude/ incoming-responses/gemini/
2. Response metadata must include packet_id.
3. Run npm run normalize.
4. Run npm run pipeline.

Collection First

Use response collection before normalization:

1. Collect responses.
2. Normalize responses.
3. Verify dataset.
4. Generate normalization certificates.
5. Run Comparator v3.1.
6. Run Comparator 3.2 explainability.
7. Generate ECR-level explainability reports.
8. Review EDRs.

Important Guardrails

• Raw responses are never rewritten.
• Comparators do not update hypotheses.
• EDRs require human completion.
• Pipeline success does not equal hypothesis validation.
• Comparator v3.1 remains frozen for ECR-000003.
• Comparator 3.2 explanations do not rescore or replace v3.1.