Initial objective and assumption
Dataset qualification sought to establish provenance and split discipline for bounded research material. An earlier identity approach allowed meaning to be coupled to row, campaign or generation metadata, risking an invalid measure of semantic separation.
Original assumption, test and root cause
The original assumption was that row-linked identity could stand in for semantic identity. The contradiction was found through qualification analysis; the root cause was coupling meaning to incidental metadata.
Failure or contradiction and design response
Qualification analysis identified the coupling. The design response used a corrected semantic envelope, regenerated material, balanced cross-split grammar representation and independent comparison. The distinction between Dataset Foundry R0 and later R1.1 research material remains explicit.
Validation after correction
Defined regeneration, provenance, disjointness and holdout checks supported qualification of bounded R1.1 corpus and holdout material. This is qualification evidence only.
- Claim
- Semantic identity was corrected to represent meaning rather than row metadata.
- Method
- Regeneration, split, leakage, provenance and independent comparison checks.
- Result
- Qualified bounded research material was produced under the corrected identity model.
- Limitation
- The result does not establish successful model training, generalisation, production readiness or promotion.
- Source period
- 2026 qualification records.
Validation after correction, current state and lesson
The correction was the semantic-envelope approach and regenerated qualification material. Validation after correction covered corpus and holdout qualification only. Remaining limitation: no generalisation, successful final training or model promotion is established.
Current canonical state and transferable engineering lesson
The candidate remains RESEARCH ONLY with a terminal REVISE CANDIDATE result below frozen thresholds. Transferable lesson: semantic identity must describe meaning rather than incidental generation metadata.
Related evidence: KEJ-EV-010.
Source and redaction note: Public summary derived from canonical qualification records. Dataset contents and experimental identifiers are withheld.