Running pattern detector against the corpus
The dataset shows contracts that started small and grew at least threefold through amendments — surfacing procurement relationships that may have outgrown their original justification.
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 · 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
The dataset shows record 172-2021-2022-Q4-987525 (Sanotize Research and Development Corp. · National Research Council Canada | Conseil national de recherches Canada) growing from $500,000 initial commitment to $5,500,000 current commitment across 2 amendments — a 11.0× expansion, above the 90th percentile (1.7×) of amended agreements in "Industrial Research Assistance Program – Contributions to Firms".
Once a match is identified, the dataset suggests these review actions. Glassbox does not prescribe outcomes — these are calibrated options for funders, auditors, and program officers, framed as correlations to investigate.
Pull the amendment 0 record. If procurement_method indicates competitive, the growth ratio is the signal. If it was sole-source from the start, the pattern is different (and may be normal).
Within the same department + program, what is the median growth ratio? Glassbox can produce that distribution. If this match sits in the top decile of growth, it warrants extra scrutiny.
Each amendment row is a published decision. Read the descriptions for purpose drift (also flagged separately as amendment-purpose-drift). Sustained scope expansion across multiple amendments without re-procurement is the review-worthy signal.
Departments with policies on amendment caps (e.g. 2× original within 5 years) can use this match to surface agreements that crossed the threshold and were not refreshed.
These are calibrated suggestions, not directives. The dataset shows the pattern; the funder decides the response.
Aggregated from the matches below. Severity bands are calibrated per pattern; counts and dollar totals reflect this snapshot only.
The dataset shows record 033-2018-2019-Q3-511845 (Computers for Success - Canada Incorporated · Innovation, Science and Economic Development Canada | Innovation, Sciences et Développement économique Canada) growing from $432,000 initial commitment to $4,674,109 current commitment across 7 amendments — a 10.8× expansion, above the 90th percentile (2.3×) of amended agreements in "Computers for Schools Intern Program".
The dataset shows record 003-2023-2024-Q1-00021 (European Space Agency|Agence spatiale européenne · Canadian Space Agency | Agence spatiale canadienne) growing from $5,678,697 initial commitment to $59,993,933 current commitment across 1 amendment — a 10.6× expansion, above the 90th percentile (10.0×) of amended agreements in "Contributions under the Canada/European Space Agency (ESA) Cooperation Agreement".
Why this may be benign · The recipient business number is missing or a placeholder ('0', all-zeros, '-', 'n/a'): rows were grouped by legal name only, so distinct entities with the same name may be merged (KNOWN-DATA-ISSUES F-6/F-7).
The dataset shows record 236-2018-2019-Q3-00197 (Holloway, Alison C · Canadian Institutes of Health Research | Instituts de recherche en santé du Canada) growing from $158,355 initial commitment to $1,583,550 current commitment across 1 amendment — a 10.0× expansion, above the 90th percentile (1.8×) of amended agreements in the same program.
Why this may be benign · The recipient business number is missing or a placeholder ('0', all-zeros, '-', 'n/a'): rows were grouped by legal name only, so distinct entities with the same name may be merged (KNOWN-DATA-ISSUES F-6/F-7).