Glassbox · pattern catalog

Follow the money.

9 named patterns across federal and Alberta provincial spending. Each surfaces specific records that match. Each output cites every source row.

1 TRACE-derived · 8 Glassbox-native

Zombie Recipients

Live

The dataset shows entities that received substantial federal funding then ceased appearing in the corpus — flagging recipients that went silent after the money flowed.

Signal · recipient total (original commitments, is_amendment = false) ≥ p90 of recipient totals within the recipient's primary department AND ≥ $500K floor · last agreement_start_date < as-of − 36 months, as-of = MAX(agreement_start_date) ≤ CURRENT_DATE · signal = years silent

17,233 records match

Challenge #1

Ghost Capacity

Live

The dataset shows entities receiving substantial federal funding with no recorded business identity — recipients the federal government cannot independently identify.

Signal · recipient_business_number IS NULL or a placeholder ('0', all-zeros, '-', 'n/a', 'none') · is_amendment = false · total ≥ p90 of ALL recipient totals within the recipient's primary department AND ≥ $500K floor · publisher batch-report rows excluded · signal = total $

20,000 records match

Challenge #2

Funding Loops

Live

The dataset shows circular money flows between charities — reciprocal pairs, triangular cycles, and longer chains. Most loops are structurally normal; the signal is the deviation from norm.

Signal · app.gift_loops: simple cycles ≤ 4 hops in the ≥ $5K qualified-donee gift graph, hop fiscal periods non-decreasing and the whole loop within 12 months, hubs (degree ≥ 50 or DAF/foundation/federated/diocese names) never interior; edge weight = gift ÷ donor's gifts that year; score = min edge weight × hops penalty (2:1.0, 3:0.8, 4:0.6); aggregated per BN, best loop ≥ 0.05 · fallback: cra.loop_universe score ≥ 12

1,391 records match

Pattern based on Alberta TRACE methodology

Challenge #3

Sole-Source & Amendment Creep

Live

The dataset shows contracts that started small and grew at least threefold through amendments — surfacing procurement relationships that may have outgrown their original justification.

Signal · agreement_value is cumulative (F-3): original = amendment-0 row ≥ $100K · final = highest-amendment row · final/original ≥ 3.0 AND ≥ p90 of growth ratios among amended agreements in the same program · ≥ 1 amendment row · chains keyed by (ref_number, BN|legal name) per F-1 · signal = growth ratio

333 records match

Challenge #4

Vendor Concentration

Live

The dataset shows programs where a single supplier or small group receives a disproportionate share of spend — incumbency replacing competition.

Signal · per (department, program) over the rolling 36 months to the corpus as-of date · program total ≥ $10M · ≥ 3 recipients · HHI = Σ recipient share² × 100² ≥ p90 of program HHIs within the department AND ≥ 1500 floor · bands: ≥5000 extreme, ≥2500 highly, ≥1500 moderately concentrated · signal = HHI

45 records match

Challenge #5

Related Parties & Governance Networks

Live

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.

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

17,501 records match

Challenge #6

Policy Misalignment

Live

The dataset shows the gap between named policy priorities (emissions, housing, reconciliation, healthcare) and the actual flow of funds — concrete spend versus stated plan.

Signal · ILIKE keyword match on prog_purpose_en / prog_name_en, summed over the rolling 5 years to the corpus as-of date, compared to a calibrated stated annual commitment per priority · signal = actual ÷ stated

4 records match

Challenge #7

Duplicative Funding & Gaps

Live

The dataset shows recipients receiving funding from both the federal grants & contributions stream and Alberta provincial grants — entities sitting at the intersection of two funder pools where program duplication is a real risk that funders can verify against.

Signal · general.entity_golden_records with both 'fed' and 'ab' in dataset_sources · ≥ 5 federal and ≥ 5 Alberta source records · combined record count ≥ p90 within the entity's entity_type · confidence ≥ 0.7 · signal = combined record count

281 records match

Challenge #8

Amendment Purpose Drift

Live

The dataset shows agreements whose current-amendment description shares few keywords with the original commitment — the contract has drifted from its initial purpose.

Signal · Jaccard token similarity (initial.description, current.description) < 0.30 with ≥ 3 amendments · chains keyed per F-1 · signal = 1 − similarity

307 records match

Match counts · totals from the pattern store (see each pattern page footer for live-table vs snapshot)

TRACE attribution

Six patterns are derived from Alberta's TRACE program (Targeted Review of Alberta's Contracts and Expenditures), Ministry of Technology and Innovation. Glassbox extends those pattern definitions to the federal corpus and surfaces them in a public-facing UI.

Read the data lineage →

Calibrated language

Glassbox surfaces correlations and observations, not causal claims. Every match carries an audit token, an evidence array citing source rows, and language reviewed against the calibration sweep (no “fraud”, no “clearly shows”, no “should have”).

Read the methodology →