Skip to content

feat(crosspost): H7 taste-cohort shadow + freeze H6/H7 plan - #348

Open
ohld wants to merge 1 commit into
productionfrom
feat/crosspost-h7-taste-shadow
Open

feat(crosspost): H7 taste-cohort shadow + freeze H6/H7 plan#348
ohld wants to merge 1 commit into
productionfrom
feat/crosspost-h7-taste-shadow

Conversation

@ohld

@ohld ohld commented Aug 10, 2026

Copy link
Copy Markdown
Member

Summary

  • H7 shadow only: attach n_taste_likes, taste_boost_shadow, taste_cohort_version to RU crosspost decision_log candidates (does not change pick).
  • Frozen cohort: top-50 users in src/crossposting/data/ru_taste_cohort_v1.json (train time-split, offline lift 1.19).
  • Freeze predictions for future data:
    • H6 (score_version=4): 2026-08-12 smoke / 2026-08-17 keep-kill
    • H7: 2026-08-24 top-half vs bottom-half f1k among taste≥1
  • Experiment note: experiments/active/2026-08-10-crosspost-v4-and-taste-cohort.md
  • Readout SQL: docs/analyst/crossposting-taste-shadow.sql
  • Frequency stays 5/day; no taste-only ranker.

Test plan

  • pytest tests/crossposting/ -q --noconftest (4 passed)
  • ruff check/format on touched modules
  • After merge/deploy: next RU post decision_log has n_taste_likes on candidates
  • 2026-08-12/17: run crossposting-v4-like-volume.sql → fill Metrics after
  • 2026-08-24: run crossposting-taste-shadow.sql query 4 → H7 pass/fail

Log n_taste_likes / taste_boost_shadow on RU ranker decision_log without
re-ranking. Freeze falsifiable H6 (v4 mature 8-12/17) and H7 (shadow 8-24)
gates in HYPOTHESES + experiment note for future readout.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant