Skip to content

Weekly screener: candidate ranking is undifferentiated (flat score-70, alphabetical tiebreak + 20-cap) #1782

Description

@dayfine

Observed (2026-06-28, reviewing the backfilled weekly pick series): every long candidate in generate_weekly_snapshot scores 70 / grade A with identical rationale. Ties break alphabetically, then capped at 20 → output is the first 20 A-ticker grade-A breakouts (ACA..AXTA), never reaching B+. 'Top 3' is alphabetical, not quality-ranked.

Weinstein-faithful rationale: in a broad bull many stocks qualify; the book says be selective and buy the strongest by relative strength (RS-for-selection is a spine item). Current screener can't — no discriminating score among grade-A candidates.

Direction: continuous score so top-N = best, not alphabetically-first. Faithful factors: RS rank vs market (spine); earliness/non-extension (book warns vs extended Stage-2 — also explains 'same symbol week after week, identical entry/stop'); volume strength; sector RS. Ship live-screener ranking first; if it changes backtest selection, default-off per experiment-flag-discipline. Cross-ref project_cascade_selection_inversion (validate, don't assume).

Files: trading/trading/weinstein/snapshot/gen/lib/weekly_snapshot_generator.ml, analysis/weinstein/screener/.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions