> Data revision: baf7c2f4307d3a989a8fb9363df8000fda3d9310
> Source files:
- [results/claim-ledger.jsonl](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/results/claim-ledger.jsonl)

# CL-WA-kh-compare-2026-09-29-public-capture-identity

**Source:** `results/claim-ledger.jsonl`  
**Ledger status:** INTERIM  
**Evidence status:** reproducible from public data for candidate capture counts; source-reported, not reproducible from public data for the historical design variants

Values below are transcribed from this claim-ledger entry. The site does not recalculate them. A status such as INTERIM, CONFOUNDED, or WITHDRAWN is not a validated advantage.

## Claim fields

- **metric:** `"candidate_delivery_capture"`
- **numerator:** `null`
- **denominator:** `null`
- **unit:** `null`
- **comparison:** `null`
- **analysis_rule:** `"Count exact IDs, non-unique handles, and exact official-grade matches separately. Do not treat candidate deliveries as the historical scored cohort."`
- **evidence_status:** `"reproducible from public data for candidate capture counts; source-reported, not reproducible from public data for the historical design variants"`
- **claim_status:** `"INTERIM"`
- **publication_status:** `"RECONCILIATION_QUEUE"`
- **reported_values:** `{"exact_delivery_ids": 32, "exact_official_grade_matches": 0, "manifest_has_arm_labels": false, "manifest_rows": 33, "non_unique_handles": 1}`
- **interpretation:** `"The captured rows do not reconstruct the historical score cohort; no harness comparison is supported."`
- **denominator_variants:** `[{"evidence_status": "reproducible from public data", "n": 33, "population": "public candidate-manifest rows"}, {"evidence_status": "reproducible from public data", "n": 32, "population": "unique public delivery IDs"}, {"evidence_status": "reproducible from public data", "n": 1, "population": "captured handles not accepted as unique cell IDs"}, {"evidence_status": "source-reported, not reproducible from public data", "n": 3, "population": "source-reported scored design", "unit": "tasks at one repetition each"}]`
- **evidence:** `[{"path": "rounds/kh-compare-2026-09-29/public-capture-manifest.csv", "selector": "all data rows; exactness from record_identity_status"}, {"path": "results/run-records/index.json", "selector": "manifest source_file and matching candidate IDs"}, {"path": "results/cells.jsonl", "selector": "official_grade_match field in candidate manifest"}]`

## Population references

Round: [kh-compare-2026-09-29](/rounds/kh-compare-2026-09-29/)

### Related hypotheses

- [H21](/hypotheses/h21/)
- [H23](/hypotheses/h23/)
- [H26](/hypotheses/h26/)
- [H28](/hypotheses/h28/)

### Related experiments

- [kh-compare-2026-09-29](/rounds/kh-compare-2026-09-29/)
Exact cell IDs are listed in the [source ledger](https://github.com/KogenAI/kogen-bench/blob/baf7c2f4307d3a989a8fb9363df8000fda3d9310/results/claim-ledger.jsonl); the site does not join outcome-only rows by guessed keys.
