Collapse one person's two author records into one row

[ssavutu]

Why this is worth merging even though we are not reseeding again

Being straight about the value: this fixes a bug that only fires during a full
WP-export reseed, and the plan is that there will never be another one. So the
practical payoff today is zero. It is here because the diagnosis was done
anyway, the ETL is still the thing anyone would reach for if an import ever
does happen again, and leaving a known root cause undocumented in code is worse
than a small tested patch. Reviewers should feel free to weigh it accordingly.

The bug

Authors are sanitized in two independent pools — WP users (auth_*) and
Co-Authors Plus guest authors (gauth_*). The MinHash similarity dedupe and the
logs/auth_conflicts.json decision cache both run within a pool only, so
combineAndReindexAuthors is the single place a person represented on both
sides gets collapsed.

It matched on exact display_name equality. The two sides routinely
disagree, because a WP user's display name is title-cased off the login while
the guest record is hand-typed:

WP user	guest author
Erik Heyman-meltzer	Erik Heyman-Meltzer
Devan Suber	DevanSuber

Each mismatch emits a second authors row for the same person.
Utility.canonicalizeAuthorLogins then finds two identical logins and breaks
the tie with dedupe_slug, which appends the row id — so an author slug ending
in its own id (erik-heyman-meltzer-870) is the fingerprint of exactly this
miss, not of anything done in the CMS.

The fix

• Match on Utility.cleanDocument(name, "similarity"), the normalization the
  within-pool dedupe already trusts.
• Merge the matched guest record instead of discarding it:
  • blank fields are filled from the guest record (usually where a real
    first/last name lives);
  • a name field that differs only by case or punctuation takes the guest's
    spelling, since that one was typed by a person;
  • anything genuinely different is left alone — a guest record must never be
    able to rename somebody;
  • email is fill-only and never overwritten. cms_users.author_id is linked
    by email at first login, so the emailed row is the one tied to a person's
    CMS account.

Effect on the Aug 2026 export

Replayed against the cached auth_output.json / gauth_output.json:

pools: 559 wp users + 435 guest authors
main:  875 rows
fix:   866 rows  (9 newly collapsed)
logins still disambiguated by id: 0
remaining same-person dupes: {}

The nine: Erik Heyman-Meltzer (the one a writer actually reported), plus
Devan Suber, Ben Wolbransky, Jacky Tsang, Colin Peterson, David Hagelgans,
Zachary Snyder, Atticus Deeny and Sally Ehlers — all eight of those the
missing-space variant.

Tests

tests/test_author_pool_merge.py, 8 cases. Four of them fail on main and
pass here; the other four pin behaviour this must not break (fresh ids, no id
collisions, no renaming, email survival). Full suite green — every module under
tests/ run individually.

Note on live data

Delta has already been repaired by hand for Erik (author 870 merged into 571).
The other eight duplicate rows exist on Delta but carry zero article links,
so they are cosmetic clutter in the author picker rather than split bylines.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HCehtwqGJGNYhmMVdwAAgF