Osman Zakaria, PMP

A regularized logistic regression (AUC 0.91 on held-out cases) scores each design in the browser. Every effect holds documentation fixed, because thin entries rarely record impact whatever happened. A context check shows how common a recruitment method is in countries like yours, from the companion research paper (under review). The repository also holds that paper's case classifier, a CatBoost and XGBoost ensemble that sorts the 2,260 cases into five method families (76% accuracy against baselines near 52%).

Documented impact is not proof of impact, and the archive is crowd-sourced. Associations, not causes. Not an official Participedia product.

The zone model is trained on 30 simulated sites calibrated to WorkSafeBC, AWCBC and Statistics Canada injury rates and run on real Environment Canada weather; it flags 84% of test-period events (AUC 0.75 against 0.66 for the crew schedule alone). Because real site logs are private, the approach was also tested on New York City's public records: across 37,000 sites, red-flagged weeks carried 11.6 times the average incident rate.

The site logs are simulated, so a pilot on a contractor's own records is the real test. Independent work, not affiliated with WorkSafeBC, the WSIB or Alberta OHS.

The page shows what drives growth by procurement method, commodity, value, length and department, and includes a risk check that estimates the chance a new contract grows 25% or more from what is known at award, a watch list of recent high-risk awards, and the 40 largest increases. The risk model is a regularised logistic regression trained on awards to March 2021 and tested on the two years after. Reviewing the riskiest 10% of new contracts catches 39% of those that later grow 25% or more, which helps target change control where it matters.

An amendment is not an overrun: many large contracts grow by design through options and phases. Rates are a floor because some amendments cannot be matched to an award. Independent analysis of public data, not affiliated with the Government of Canada.

A planning view estimates how many of these workers local openings could absorb within a year and where retraining helps most. A new-products view tests how much of a retooled plant's hourly workforce the laid-off workers could fill.

Independent analysis from public data. Postings are not hires, and the plant workforce mix and the share of openings workers can win are adjustable assumptions, not union data.

It also includes an early-warning model for business customer churn, a ranked sales call list, and an allocation of the 79 million-piece (32.6%) drop in 2025 parcel volume across areas by modelled demand, to set win-back sales targets.

Independent analysis, not produced by or affiliated with Canada Post. The churn model runs on 8,000 simulated business customers because no real customer data is public. The saving is a model estimate based on stated assumptions; it is above the government's ~$400M figure, and route-level checks suggest it is conservative.

On-time rank and disruption risk differ. Southwest is 7th of 13 on on-time arrivals (73.7%) but has the second-lowest rate of 3-hour-plus delays, 78 per 10,000 flights against 320 for PSA. Delays also cascade through the day: 91.5% of flights scheduled at 5 a.m. arrive on time, against 66.3% at 8 p.m.

A Simpson's paradox check of all 78 carrier pairs found rankings that reverse at comparable airports. American trails Southwest overall (71.8% against 73.7%) but is 2.2 points ahead at the same airports in the same months, with a 95% interval of +1.9 to +2.4. American wins about half of the compared airport-months, so the lead comes from the large ones.

Covers reporting carriers only. Delay causes are the airlines' own attribution, and "late aircraft" delay can start with weather on an earlier leg.

Awareness, not income, decides acceptance: 56% of workers who have heard of AI view it positively, against 3% of the rest (Fisher exact p < 0.001), while income shows no link (Spearman ρ = −0.07). Workers fall into three readiness segments (26% ready, 35% persuadable, 39% resistant), each needing a different policy. Professionals rank transparency in how taxes are spent as the strongest compliance lever. 27 of the paper's 29 figures match the coded forms within 5 points.

A small, purposive Ghana sample: it shows patterns and policy levers, not national prevalence.