Skip to content

feat(review): the week-in-review — arjev review, a pure re-read of the journal - #84

Merged
aarontrowbridge merged 1 commit into
mainfrom
amico/issue-83-arjev-review
Oct 11, 2026
Merged

aarontrowbridge merged 1 commit into
mainfrom
amico/issue-83-arjev-review

Conversation

@aarontrowbridge

Copy link
Copy Markdown
Member

Closes #83.

The week-in-review, shipped

arjev review re-reads the journal the digests already wrote — the same three sources calibrate joins (journal, receipts, label ledger), zero fetches, zero Jev calls, deterministic — and renders the across-days view the capped daily digest can't see:

  • The Buried — the cap's casualties: never-posted candidates re-ranked by Jev confidence, deduped across runs at their best judgment (top 10 of the window's total, both numbers stated)
  • Your People — the author-lane matches this window, posted or not
  • The Divergences — the two rank-gap tails, biggest divergence first (a regression guard pins the sort: ascending-by-gap once surfaced gap-1 noise as the top)
  • The Week's Receipts — kept/skipped counts with the calibration headline (precision@5, Brier n), and when there are no labels it says so plainly instead of reading hollow

--days N (default 7); an empty window is an honest report, a missing journal fails loud (digest first). --post slack rides the digest's client seam — same posting path, testable offline. Read-only throughout: the review proposes, never stages; the only fail-open Jev rows ({fail_reason} with no confidence) carry no mass and never rank.

What the first live run surfaced

On the real journal (6 runs, 3,350 never-posted candidates this window) the report's tails are informative in opposite directions: the jev-hot/lexical-buried tail is old cs.AI survey spam carrying suspicious confidence: 1.0 — the same pattern the v0.3 stopwords work chased — while the lexical-hot/jev-cold tail is real physics (trapped-ion gates, bosonic-code protection, Kitaev-chain readout) that Jev ran cold on. That's the review doing its job; the calibration will say more once reaction labels exist.

Verification

  • pytest: 226 passed (16 new — window filtering, each section on fixture data with exact expected lines, fail-open shapes, sort-direction regression, empty-window honesty, read-only byte-snapshot, no-Jev/no-fetch tripwire, CLI smoke, Slack seam, docs coverage)
  • ruff check .: clean; docs-command lint green (README + agent-setup document the verb verbatim as shipped)

…e journal

The daily digest is capped at top-5 and sees one day; the review looks across
the window (issue #83). Four read-only, deterministic sections over the same
three sources calibrate joins — no fetches, no Jev calls:

- The Buried — the cap's casualties, never-posted candidates ranked by Jev
  confidence, deduped across runs at their best judgment
- Your People — the author-lane matches, posted or not
- The Divergences — the two rank-gap tails, biggest divergence first
- The Week's Receipts — kept/skipped with the calibration headline, n-stated

--days N (default 7); --post slack rides the digest's client seam. Read-only:
the review proposes, never stages.
@aarontrowbridge
aarontrowbridge merged commit 903f0e1 into main Oct 11, 2026
1 check passed
This was referenced Oct 12, 2026
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.

feat(review): the week-in-review — arjev review over the journal

1 participant