AutoResearch
StaleKeptMedium bandFamily Manager
Flyer OCR pre-pass before extraction
- Baseline
- 71%
- Final
- 89%
- Delta
- +18 pts
- Variants
- 3
Objective
What we set out to improve
Reduce missed dates on scanned camp flyers by adding an OCR pre-pass before structured extraction.
KeptPromoted to a templateWrote to a KB
Kept. The high-DPI OCR pre-pass lifted date recall from 0.71 to 0.89 at a medium resource cost, and the heuristic was promoted to the family-documents knowledge base.
Iterations
Variants we tried
Each variant and its coarse objective metric. The kept variant is marked; bars are relative to the best run.
- 1Baseline — extraction onlyLow71%
- 2Variant A — OCR pre-pass, default DPIMedium83%
- 3Variant B — OCR pre-pass, high DPI + deskewWinnerMedium89%
Run
Stages
baseline
Succeeded · 4.2s
variant run
Succeeded · 9.8s
eval
Succeeded · 1.5s
promote
Succeeded · 300ms
Output
Artifacts and what shipped
Redaction-safe artifact previews, diffs, metric tables, and prompt variants with sensitive text removed.
- Metric table
Date-recall by variant (baseline → 0.89)
- Diff summary
Pipeline diff: insert OCR pre-pass stage
- KB write
Promoted OCR heuristic to the family-docs KB