Backfilling an old week used to sweep in undigested papers published later
(stamping them into the wrong week and removing them from the next Monday's
digest); regular weekly runs are unaffected since the collection window
already ends at week_end.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
OpenAlex indexes TSE ~2-3 months late (nothing after 2026-06-26 as of Sep 2),
so TSE never appeared in weekly digests. Sources can now be flagged
crossref_fresh: their recent papers are discovered via Crossref (created-date
window, paper-level CC licence = OA) and listed without abstracts; a digested
paper's missing abstract is filled once OpenAlex indexes the DOI, so a later
--force re-run can add its per-paper summary.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
MiniMax occasionally mixes hanja/kana into Korean text (実態, 软件, かな).
Parsed outputs are now scanned (verbatim fields exempt: title/doi/claim) and
offending replies flow through the existing repair retry; prompts spell the
rule out with examples.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Large chunked responses (the 210KB digest detail) can be cut off by some
browser/middlebox combinations (Safari 'Load failed'); send every body with
an explicit length, gzipped when accepted, and log finishes/aborts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A transient failure (e.g. container swap during redeploy) used to leave the
body panel silently empty; show a retry hint in the body instead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Per-paper summary batches are fact-checked against abstracts (verify ->
revise once on major issues -> re-verify), controlled by SEBRIEF_VERIFY_ENABLED
- The paper cap never drops papers already published in this week's digest
(--force re-runs could otherwise erase them from every digest)
- SEBRIEF_MAX_PAPERS default 40 -> 60 (observed weekly volume ~47)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
