Running pattern detector against the corpus
The dataset shows entities the upstream entity-resolution pipeline adjudicated as RELATED to other funded entities — distinct legal registrations that share a name, structure, or governance lineage — where related-party relationships are more likely to materialise.
Glassbox surfaces correlations and observations, not findings of misconduct. Severity is a calibrated read on how strongly the dataset matches the pattern definition — a starting point for review, not a verdict.
The match meets the pattern threshold but the signal is the lower band. Normal review queue.
The match exceeds the routine threshold. Worth a closer look — pull the source rows, walk the citation.
The match is in the highest band of the calibrated scale. Recommend: prioritise for active review.
Severity is computed from the detection signal · general.entity_merge_candidates.llm_verdict = 'RELATED' (Splink candidate → LLM adjudication) edges joined to general.entity_golden_records on both sides · ≥ 2 distinct RELATED counterparties · confidence ≥ 0.7 · signal = related-counterparty count · severity weights cross-dataset presence and source-record volume
The dataset shows Best Buy Canada Ltd (company) adjudicated as RELATED to 2 other entities by the entity-resolution pipeline (mean verdict confidence 0.72) — BBYMARKETPLA BEST BUY; BEST BUY FOR BUSINESS; it appears in 2 datasets (ab, fed) with 342 source records. Glassbox surfaces the network; the funder verifies governance overlap.
Aggregated from the matches below. Severity bands are calibrated per pattern; counts and dollar totals reflect this snapshot only.