ONWSIB Deep-Dive Method Note
This appendix explains the short local follow-up pass used for the ONWSIB archive after the 2020-2026 scrape was reconciled.
Scope
- Dataset: 463 ONWSIB decisions from 2020-2026
- Coverage check: local counts matched observed CanLII browse counts for each year in scope
- Source files: local JSON outputs in
data/tribunal-decisions/
What The Deep-Dive Did
- Ran in sidecar mode against local files only
- Made no additional API calls
- Applied rule-based outcome detection to the reconciled archive
- Used
predMinConfidence = 0.55to keep the threshold conservative
Result
- 12 high-confidence outcome reads in the current archive
- 6 additional cases queued for manual review
- Year-level gains came from 2022 (+2) and 2023 (+1); other years held steady
Year Summary
| Year | Total decisions | Known before | Known after | Net gain |
|---|---|---|---|---|
| 2021 | 49 | 2 | 2 | 0 |
| 2022 | 149 | 0 | 2 | 2 |
| 2023 | 120 | 0 | 1 | 1 |
| 2024 | 73 | 0 | 0 | 0 |
| 2025 | 64 | 3 | 3 | 0 |
| 2026 | 8 | 4 | 4 | 0 |
What This Note Does Not Claim
- It does not create an official ONWSIB success rate.
- It does not replace manual reading of difficult files.
- It does not change the core public-record limitation: most ONWSIB outcomes remain unresolved in published materials.
Primary Files
- Summary JSON: onwsib-deep-dive-summary.json
- Manual-review queue: onwsib-manual-review-queue-top100.json
Use this note as a methodology appendix for ONWSIB-only content. For worker-facing posts, the safest summary remains: the archive is reconciled, the deep-dive is local-only, and public outcome completeness is still limited.