Aggregating recommendations across live detectors
What the dataset suggests the funder might do — calibrated actions distilled from 203 live pattern matches across the federal corpus and Alberta provincial sources, prioritised by severity and dollars at stake.
Glassbox observes; it does not direct. Each recommendation cites its underlying pattern matches and source rows. The funder decides the response.
Sequenced execution map · dependency-cascaded · not policy advice
Highest severity or largest dollars at stake. Surface to the audit committee or program lead.
The dataset shows 50 entities that received ≥$500K in federal funding then went silent in the corpus for ≥36 months — totalling $66.7B in distributed funds. 24 have been silent ≥6 years. Silence is not evidence of incompletion, but it is the signal to verify deliverable closeout, final-report submission, and (where applicable) funds returned.
The dataset shows 3 federal departments with HHI ≥ 1500 across recipient share. 1 are in the extreme band (HHI ≥ 5000) — a single supplier or small group dominates. Total disbursed by these departments: $27.4B. Concentration is not always a problem (specialty procurements may be inherently concentrated), but persistent concentration indicates the procurement strategy may merit redesign.
The dataset shows 50 federal contracts where the current value is ≥3× the original commitment, totalling $5.6B in current spend. 50 are in the highest severity band (≥10× growth). Departments with formal amendment-cap policies (typically 2× or 1.5× over a multi-year window) would surface these for re-procurement before the threshold is crossed.
The dataset shows 50 CRA-registered entities participating in circular money flows scored ≥12 on the TRACE attention scale. 46 are scored ≥18. Cumulative circular amount: $93M. Most loops are structurally normal (denominational hierarchies, federated charities, donation platforms) — the review is to distinguish those from loops that exist to inflate revenue, generate tax receipts, or absorb funds into overhead.
The dataset shows 50 recipients receiving substantial federal funding (≥$500K) with no recorded business number, totalling $126.7B. Some matches will be data-entry omissions; some are substantive identity questions. Both are addressable through tighter intake controls and a reconciliation pass against the federal corporate registry.
Glassbox runs every live pattern detector against the source corpus, aggregates the matches, and emits structured recommendations through src/lib/recommendations/build.ts. Priority is computed from severity (flag / attention / observation) × dollars-at-stake. Each recommendation cites its underlying matches and sources — the trail walks back to fed.grants_contributions and cra.loop_universe row by row.
Recommendations frame as observations and options — never as directives or causal claims. “The dataset suggests reviewing X” rather than “X is fraud.” Severity bands describe match strength, not proven misconduct. Read the methodology for the full calibration rules.