Review-queue nav slices render auto-derived columns without the verdict field — declare highlightFields on ats_employer and ats_job - #71

Fixes #33

What changed

Three highlightFields declarations, nothing else — src/objects/employer.object.ts,
job.object.ts, candidate.object.ts, plus docs/evidence/issue-33/.

platform.nav.ts and seeker.nav.ts mount three queues as filters slices on the bare data
surface. objectui's ObjectDataPage calls defaultColumnsFromObject, which takes
highlightFields when the object declares it and otherwise walks the first six business fields.
It never consults an object's authored list — which is why authoring ats_employer.all and
ats_job.all in #3 did not move these queues, exactly as the card predicted.

The card's comment adds a third object: the seeker's My Profile entry is a filters slice
({ user: '{current_user_id}' }) on ats_candidate for the same reason. Re-audited on this
branch: those three are the only filters entries in the app's navigation; every other entry
names a view or lands on a declared list.

Measured, not expected

Booted twice in the same container on the same recipe — once on 62e496d (origin/main, this
branch's base) for the "before", once on this branch for the "after". Reproduced rather than
quoted from the docs/evidence/issue-3/ shots; the reproduction matches them exactly. Persona
admin@platform.example, and candidate01@mail.example for the seeker slice. Boot was
--database-driver memory because #39 empties the Employers Pending queue on sqlite.

The nav slices — the acceptance

The declaration is not a no-op on this surface. Shots 00-… through 05-…; both platform
slices were reached by clicking the nav entry, not by typing the URL.

The other surfaces highlightFields feeds

The job record header is the one place this is not a clean win. At five strip items two badges
truncate to Pendi… and Full-ti…, and Status is now shown twice — once in the strip and once
in the state-machine stage bar immediately below it. The strip divides its width evenly, so this is
content-length behaviour rather than anything the metadata sets. Dropping status from
ats_job.highlightFields would clear it and would also undo the card: a missing status in the
Jobs Pending queue is the defect being repaired. Kept, cost reported. Shot 09-….

The candidate record header loses Phone and Email. Both remain on the Details tab as live
tel: / mailto: links. Net better in our reading — the header now answers "who is this
professionally, and are they looking" instead of showing two fields that field-level security masks
from every recruiter anyway — but it is a real removal, so it is stated rather than buried. Shot
11-….

Non-regression

highlightFields feeds only surfaces that derive their columns. Checked, and each rendered set is
identical to the array in src/views/: ats_employer.all (columns), ats_job.all
(platformColumns), ats_job.published (seekerColumns), ats_inquiry.inbox. Shots 12-…
and 13-….

One deviation from the card's proposal

ats_job takes work_mode where the card proposed published_at.

The job_status_transitions state machine only reaches published through pending_review, so a
row in the Jobs Pending queue has never been published and its published_at is null by
construction. Measured on the seed: 0 of the 6 pending rows carry it, and 0 of the 12 rows in
draft / pending_review / rejected; the 28 that carry it are exactly the
published / paused / closed rows. As proposed it would have been a permanently empty sixth
column in the one surface this card exists to repair. work_mode is set on 40 of 40 jobs and
completes the terms pair a reviewer checks alongside employment_type. published_at keeps its
place on the record's Details tab and in the authored ats_job.all view.

The substitution also lands the set on the authored platform list: ats_job.platformColumns is
title · employer · status · city · employment_type · published_at · expires_at, and the declared
set is now its first five plus work_mode. Same for ats_employer, whose declared set is the
authored columns array minus size. The queues now agree with the lists #3 authored.

Everything else in the proposal measured out and is used as written, in the proposed order. Two
proposed-adjacent fields were checked and deliberately left out: ats_employer.verification_docs
(the card lists it as missing but does not propose it — the seed carries it on 0 of 12 rows and a
multi-file cell cannot be read at a glance; the reviewer opens it on the record), and size, which
the auto-derived set already had and the authored list keeps.

Every field name in all three declarations was confirmed against the object before use.
defaultColumnsFromObject filters the curated array through objectDef.fields[name], so a
misspelling would have silently shrunk the column set rather than failing any gate.

Caveats worth a reviewer's eye

• In each filtered slice the verdict column is constant by construction — the filter pins it, so
  Employers Pending shows Pending on both rows and Jobs Pending shows Pending Review on all six.
  Its value is that the queue now states which verdict state it is showing, and that the same set
  carries real variance on the unfiltered list, the record header and the related list.
• profile_visibility on ats_candidate is now legible to anyone who can already read the row.
  It is still enforced nowhere — that is Employers cannot read any candidate, and profile_visibility is enforced nowhere #13's subject, and nothing here should be read as changing
  it. Called out in the card's comment and repeated in the source comment.

Gates

Exit codes captured before any pipe, on the committed tree.

pnpm validate   VALIDATE_EXIT=0
pnpm lint       LINT_EXIT=0        (objectstack lint --i18n-strict)
pnpm typecheck   TYPECHECK_EXIT=0
npx objectstack build   BUILD_EXIT=0

validate and lint each report the same 3 pre-existing approval-staffing suggestions that
62e496d reports; no new diagnostic. No new translatable string is introduced — the declarations
reference existing field keys, whose labels already ship in both en and zh-CN.

highlightFields in dist/objectstack.json after the build:

ats_candidate  ['full_name', 'current_title', 'city', 'seeking_status', 'experience_years', 'profile_visibility']
ats_employer   ['name', 'industry', 'city', 'verification_status', 'service_tier', 'owner']
ats_inquiry    ['display_name', 'status', 'email', 'submitted_at']       (pre-existing)
ats_job        ['title', 'employer', 'status', 'city', 'employment_type', 'work_mode']

Out of scope, filed separately

#67 — Field.user columns render the raw user id in every grid, so Primary Contact reads
usr_ats_pixelforge_admin. Visible in these shots but not caused here: the authored
ats_employer.all view in shot 12-…, which this branch does not touch, renders it identically.
Unassigned and unlabelled for triage.

Generated by Claude Code