Krow/krow_talent_app
3
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(hiring): final-selection queue, honest seat counts, and human interviews

Browse Source 
Authored in a parallel session alongside the TypeScript migration; committed
separately so the two never share a commit. No TypeScript migration file is
included here.

Candidates becomes the queue of hiring decisions waiting on a person, rather
than a second Talent Pool listing every application the org ever took. Final
selection is DERIVED - there is no `final_selection` value in the
`application_status` enum and none is added. The fact it reads is the existence
of an interview row, a NOT NULL foreign key, rather than
`job_applications.interview_id`, which the schema keeps as an unconstrained soft
reference precisely so it may dangle. `status = 'interview'` is set both when an
interview is arranged and when one is completed, so status alone cannot tell a
queue of people who have been interviewed from a queue of people merely booked
in.

Seats on a position are counted from the employment records instead of a stored
column. A `filled` counter would be a second source of truth, and the day it
disagreed with `staff` nothing could say which was lying. Someone who has left
frees their seat, and over-hiring floors at zero rather than going negative.

`DEMO_FILL` is gone. `hiringRecords.js` padded the hires list with five invented
people so Hired History read as a history rather than as three rows; the padding
reached Analytics too, where "total hires" counted eight against a database
holding three. Hires now come only from `staff`.

Both paths that file an application on somebody's behalf now carry
`worker_profile_id`, the link back to the talent-pool record. The column is
nullable, so omitting it saved cleanly and failed silently: the application
belonged to an email address rather than to a person, and the hire it became
could not be traced back to the profile it came from.

`HiredChronology` used to `return null` with no hires, taking the `chronology`
node identity out of the DOM with it - so on an honest empty dataset the section
could not be addressed by Owliver or the layout editor at all. It now renders an
empty state inside the section it keeps.

Nine new checks cover the above; `npm test` reports 1684/1691.

SIX SSR PARITY CHECKS FAIL ON PURPOSE, and `scripts/__baseline__/README.md`
documents each with verified tag counts. Five are the `DEMO_FILL` removal: the
Hired History and Analytics baselines were captured while the padding was in
effect and, because they render with queries disabled, the padding is all they
contain. The sixth is this change to what Candidates says. Do not regenerate
those baselines to clear them - two of the checks exist to prove the UI node
tree migration added exactly two `<div>`s, and that proof needs the baseline to
be pre-migration markup. Recapturing now would write post-migration markup into
a file named `pre-migration` and the check would compare the current render
against itself forever. The debt is held until the migration work lands, when
both files are recaptured together.

The seventh failure, the stale backend seed fixture, predates all of this.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HBG1wnuRfJKCstGB8Fekr8
This commit is contained in:
Aravind
2026-09-17 22:52:51 +05:30
parent 64140c7add
commit dca184289e
17 changed files with 1624 additions and 94 deletions
Download Patch File Download Diff File Expand all files Collapse all files

348
scripts/skill-check.mjs
View File

@@ -5959,6 +5959,354 @@ record('an assigned candidate counts as hired',
records.HIRED_STATUSES.includes('assigned') &&
records.rankOf('assigned') >= records.STAGE_ORDER.indexOf('hired'));
/* ── Final selection, and seats ──────────────────────────────────────────
*
* Final selection is DERIVED — there is no `final_selection` value in the
* `application_status` enum and none is added. The fact it reads is the
* existence of an interview row, which is a NOT NULL foreign key, rather than
* `job_applications.interview_id`, which the schema keeps as an unconstrained
* soft reference precisely so it may dangle.
*
* That distinction is the whole point: `status = 'interview'` is set both when
* an interview is arranged and when one is completed, so status alone cannot
* tell a queue of people who have been interviewed from a queue of people who
* have merely been booked in.
*/
const fsApps = [
{ id: 'a-applied', status: 'applied' },
{ id: 'a-booked', status: 'interview' }, // arranged, never sat
{ id: 'a-done', status: 'interview' }, // sat it
{ id: 'a-hired', status: 'hired' },
{ id: 'a-assigned', status: 'assigned' },
{ id: 'a-rejected', status: 'rejected' },
];
const fsInterviews = [
{ id: 'i1', application_id: 'a-done' },
{ id: 'i2', application_id: 'a-hired' },
{ id: 'i3', application_id: 'a-rejected' },
];
const queue = records.finalSelection(fsApps, fsInterviews).map((a) => a.id);
record('final selection holds only the interviewed and undecided',
queue.length === 1 && queue[0] === 'a-done', queue.join(', ') || 'empty');
record('...so an interview merely arranged does not qualify',
!records.isFinalSelection(fsApps[1], fsInterviews),
'status interview with no interview row');
record('...and a decided application never reappears in the queue',
['a-hired', 'a-assigned', 'a-rejected']
.every((id) => !queue.includes(id)),
'hired, assigned and rejected all excluded');
record('...while the interview record, not interview_id, is what is read',
!records.isFinalSelection({ id: 'x', status: 'interview', interview_id: 'dangling' }, []),
'a soft reference alone proves nothing');
/* Seats are counted from the employment records. A `filled` column would be a
second source of truth, and nothing could say which one was right. */
const seatPosting = { id: 'p1', headcount: 2 };
const seatStaff = [
{ id: 's1', job_posting_id: 'p1', status: 'active' },
{ id: 's2', job_posting_id: 'p1', status: 'onboarding' },
{ id: 's3', job_posting_id: 'p1', status: 'inactive' }, // left; seat reopens
{ id: 's4', job_posting_id: 'p2', status: 'active' }, // another position
];
record('seats filled are counted from the staff records',
records.filledFor('p1', seatStaff) === 2, `${records.filledFor('p1', seatStaff)} of 2`);
record('...someone who has left frees their seat',
records.remainingFor({ id: 'p1', headcount: 3 }, seatStaff) === 1,
'inactive staff do not hold a seat');
record('...a full position reports no remaining seats, never a negative number',
records.isFullyStaffed(seatPosting, seatStaff)
&& records.remainingFor({ id: 'p1', headcount: 1 }, seatStaff) === 0,
'over-hiring floors at zero');
/* ── The Candidates page IS the final-selection queue ─────────────────────
*
* The derivation above is only worth having if the page that decides hires
* actually opens on it. Candidates used to list every application the org had
* ever taken — all seven statuses at once — which made it a second Talent Pool
* rather than a queue of decisions waiting.
*
* Two rules these assertions hold in place, both of them safety rules rather
* than presentation ones:
*
* The evidence is the interview ROW. `interview_id` is a soft reference the
* schema deliberately leaves unconstrained, and it dangles in real data.
*
* The score never gates. `ai_score`, `verdict` and `hire_recommendation`
* assist a recruiter; a human makes the selection. A candidate who scored
* zero and sat the interview is still a decision somebody owes them.
*/
record('a candidate who scored zero still reaches the queue',
records.isFinalSelection({ id: 'a-zero', status: 'interview', ai_score: 0 },
[{ id: 'i9', application_id: 'a-zero' }]),
'screening is not a gate on the decision');
{
/* Comments are stripped first: these assert what the page DOES, and a
comment explaining the bug that was fixed must not read as the bug. */
const decomment = (src) => src.replace(/\/\*[\s\S]*?\*\//g, '').replace(/^\s*\/\/.*$/gm, '');
const candidatesSrc = decomment(readFileSync(join(ROOT, 'src/pages/admin/Candidates.jsx'), 'utf8'));
const nodesSrc = decomment(readFileSync(join(ROOT, 'src/pages/admin/candidates/nodes.jsx'), 'utf8'));
/* Read from source rather than imported: the page pulls in the interview
modal, which touches `window` at module scope, and standing up a DOM to
read one string constant would be a worse test than reading the string. */
const defaultStage = candidatesSrc.match(/export const DEFAULT_STAGE = '([a-z_]+)'/)?.[1];
record('the candidate page opens on the final-selection queue',
defaultStage === 'final' && /useState\(DEFAULT_STAGE\)/.test(candidatesSrc),
String(defaultStage));
/* The interview rows were fetched and then thrown away — `useInterviews();`
on its own line, its result unbound — which is why the queue could not be
derived on the one page that needed it. */
record('the candidate page keeps the interview rows it fetches',
/(const|let)\s*\{[^}]*\}\s*=\s*useInterviews\(\)/.test(candidatesSrc),
'useInterviews() is bound, not discarded');
/* The filter compared `a.stage`. Applications carry `status`; no application
has ever had a `stage` property, so every stage selection emptied the
page. */
record('the stage filter reads a field applications actually carry',
!/\ba\.stage\b/.test(candidatesSrc) && /matchesStage\(/.test(candidatesSrc),
'status, not an undefined stage');
/* Narrowing the default view is only safe if nothing becomes unreachable.
`shortlisted` and `assigned` were absent from these options, so the rows
holding them could not be found by any selection. */
const offered = [...nodesSrc.matchAll(/\{ value: '([a-z_]+)', label: '[^']+' \}/g)]
.map((m) => m[1]);
const unreachable = records.APPLICATION_STATUSES.filter((st) => !offered.includes(st));
record('every application status is still reachable from the stage filter',
unreachable.length === 0, unreachable.join(', ') || 'all seven offered');
record('...and the queue itself is offered alongside them',
offered.includes('final') && offered.includes('all'),
'final selection and all stages');
}
/* ── Human interviews leave a trace, and only completion counts ───────────
*
* A human interview used to persist nothing at all: the modal collected a
* type, a date, a time and notes, announced "Interview scheduled!", and dropped
* every one of them. So the final-selection queue could only ever hold people
* an AI had interviewed, and anyone seen by a person was invisible to it.
*
* What changed is where each fact lives, and the split is the safety property:
*
* booking → a `user_activity` row. Advances nobody.
* completion→ an `ai_interviews` row. The evidence, unchanged since Step 6.
*
* `isFinalSelection` is not touched by any of this and is not expected to be.
* These assertions exist to prove that a booking cannot masquerade as an
* interview, which is the one way this work could put a candidate in front of a
* hiring decision that nobody has actually met.
*/
const human = await server.ssrLoadModule('/src/lib/humanInterviews.js');
{
const app = { id: 'a-human', status: 'interview', job_posting_id: 'p1', applicant_name: 'A' };
const at = (n) => new Date(Date.UTC(2026, 0, n)).toISOString();
const bookedEvent = {
event_type: human.SCHEDULED, application_id: 'a-human', created_date: at(1),
metadata: { interview_type: 'video', scheduled_date: '2026-01-09', scheduled_time: '10:00', notes: 'prep' },
};
const completedRow = { id: 'i-human', application_id: 'a-human' };
/* Booked, and nothing else. */
record('a booked human interview is recorded',
human.humanInterviewState(app, [bookedEvent], []) === 'scheduled');
record('...and the booking keeps what was entered',
(() => {
const d = human.scheduleDetails(human.scheduledInterview(app, [bookedEvent]));
return d.type === 'video' && d.date === '2026-01-09' && d.time === '10:00' && d.notes === 'prep';
})(), 'type, date, time and notes survive');
record('...but booking alone is NOT final selection',
!records.isFinalSelection(app, []),
'no interview row, no decision to make');
/* Held. */
record('a completed human interview is recorded',
human.humanInterviewState(app, [bookedEvent], [completedRow]) === 'completed');
record('...and a completed human interview DOES reach final selection',
records.isFinalSelection(app, [completedRow]));
record('...even with no score, because a human interview measures none',
records.isFinalSelection({ ...app, ai_score: 0 }, [completedRow]),
'absence of an AI score is not absence of an interview');
/* Not held. Both outcomes are events, and neither is an interview. */
for (const [label, type] of [['cancelled', human.CANCELLED], ['not attended', human.NO_SHOW]]) {
const events = [bookedEvent, { event_type: type, application_id: 'a-human', created_date: at(2) }];
record(`an interview ${label} leaves no completion record`,
human.humanInterviewState(app, events, []) === (type === human.CANCELLED ? 'cancelled' : 'no_show'));
record(`...so ${label} never reaches final selection`,
!records.isFinalSelection(app, []));
}
/* A decided application is out regardless of how it was interviewed. */
for (const status of ['hired', 'assigned', 'rejected']) {
record(`a ${status} application leaves the queue even with a human interview`,
!records.isFinalSelection({ ...app, status }, [completedRow]));
}
/* The Step 6 guard, restated against a human booking: a booking may well set
interview_id one day, and it must still prove nothing on its own. */
record('a dangling interview_id still does not qualify a human interview',
!records.isFinalSelection({ ...app, interview_id: 'nope' }, []),
'the row is the evidence, not the reference');
/* An event claiming completion with no row behind it. The log is not the
record, and treating it as one would reintroduce exactly the soft-reference
problem interview_id already demonstrates. */
record('a completion EVENT without an interview row is not completion',
human.humanInterviewState(app,
[{ event_type: human.COMPLETED, application_id: 'a-human', created_date: at(3) }], []) === 'none',
'user_activity is a log, not the evidence');
}
/* ── Nothing about AI is invented for a human interview ───────────────────
*
* The completed-interview record lives in `ai_interviews`, which is currently
* reused as the durable completion record for both flows. Reusing the table is
* a deliberate decision; filling in its AI columns with plausible numbers no
* machine produced would not be. A fabricated score or verdict would be read by
* analytics, by the candidate profile and by a recruiter as a measurement.
*
* `overall_interview_score` matters twice over: the server copies it onto the
* application only when the request mentions it, so sending a 0 would overwrite
* a real screening score with one this interview never took.
*/
{
const hooksSrc = readFileSync(join(ROOT, 'src/lib/krowHooks.js'), 'utf8');
const complete = hooksSrc.slice(
hooksSrc.indexOf('export function useCompleteHumanInterview'),
hooksSrc.indexOf('export function useRecordInterviewNotHeld')
);
const schedule = hooksSrc.slice(
hooksSrc.indexOf('export function useScheduleHumanInterview'),
hooksSrc.indexOf('export function useCompleteHumanInterview')
);
record('the human completion path exists at all', complete.length > 0 && schedule.length > 0);
const fabricated = ['overall_interview_score', 'integrity_score', 'ai_flags', 'category_scores',
'hire_recommendation', 'messages', 'ai_score']
.filter((f) => new RegExp(`^\\s*${f}\\s*:`, 'm').test(complete));
record('a human interview fabricates no AI data',
fabricated.length === 0, fabricated.join(', ') || 'no AI column is written');
record('...and the interviewer\'s own assessment is only sent when given',
/if \(assessment\) record\.verdict = assessment;/.test(complete),
'verdict is human-entered or left to the schema');
record('...while the facts it does write are the ones it knows',
/application_id: application\.id/.test(complete)
&& /job_posting_id: application\.job_posting_id/.test(complete));
/* The rule the whole step turns on. */
record('scheduling never writes an interview record',
!/AIInterview|useCreateInterview|createInterview/.test(schedule),
'a booking touches user_activity only');
record('...and completion goes through the existing interview client',
/createInterview\.mutateAsync/.test(complete),
'no second interview API');
/* Not held writes an event and nothing else — no row, and no invented
application status to go with it. */
const notHeld = hooksSrc.slice(hooksSrc.indexOf('export function useRecordInterviewNotHeld'));
record('a cancelled or unattended interview writes no record and no new status',
!/AIInterview|createInterview/.test(notHeld) && !/JobApplication\.update/.test(notHeld),
'an event, and nothing else');
}
/* The AI flow is untouched: it still writes its own record at completion, and
still does it through the same client this now shares. */
{
const aiSrc = readFileSync(join(ROOT, 'src/components/krow/AIInterviewModal.jsx'), 'utf8');
record('the AI interview still records itself on completion',
/const finishInterview = useCallback/.test(aiSrc)
&& /createInterview\.mutateAsync\(\{/.test(aiSrc)
&& /messages: finalMessages/.test(aiSrc),
'unchanged by the human path');
}
/* The modal is no longer theatre. */
{
const modalSrc = readFileSync(join(ROOT, 'src/components/krow/ScheduleInterviewModal.jsx'), 'utf8');
record('scheduling an interview persists it',
/useScheduleHumanInterview/.test(modalSrc) && /schedule\.mutateAsync/.test(modalSrc));
record('...the props its six call sites pass are unchanged',
/\{ open, onClose, application, position \}/.test(modalSrc),
'open, onClose, application — position optional');
record('...and a failed save is reported rather than celebrated',
/setError\(/.test(modalSrc) && !/setScheduled\(true\)/.test(modalSrc),
'the old version always claimed success');
}
/* Every path that files an application on somebody's behalf must carry the
link back to the talent-pool record. The column is nullable, so omitting it
saves cleanly and fails silently: the application belongs to an email address
instead of to a person, and the hire it becomes cannot be traced back to the
profile it came from. Both call sites are asserted because they were written
at different times and only one of them is on the position page. */
const positionDetailSource = readFileSync(join(ROOT, 'src/pages/PositionDetail.jsx'), 'utf8');
const hooksSource = readFileSync(join(ROOT, 'src/lib/krowHooks.js'), 'utf8');
record('admitting talent to a position links the application to the profile',
/worker_profile_id: profile\.id/.test(positionDetailSource),
'PositionDetail.admitTalent');
record('...and so does assigning a worker straight from the pool',
/worker_profile_id: worker\.profile\?\.id/.test(hooksSource),
'useAssignWorkers filing an application');
/* ── One person, many positions ──────────────────────────────────────────
*
* The two profile routes answer different questions and must stay separate.
* `/admin/talent/:id` is the PERSON — a `worker_profiles` row, and every
* position they are up for. `/admin/candidates/:id` is that person FOR ONE
* POSITION — a `job_applications` row, with its score, interview and decision.
*
* One screen doing both is how a product ends up with a candidate record per
* vacancy, which is exactly what `worker_profiles` exists to prevent.
*/
record('the person and the application have separate routes',
routed.has('/admin/talent/:id') && routed.has('/admin/candidates/:id'),
[...routed].filter((r) => r.includes('talent') || r.includes('candidates')).join(', '));
/* Which applications are this person's. The foreign key is the answer; email is
the fallback for rows written before the key was being set, and it has to be
there or every application already in the database detaches from its person. */
const whoProfile = { id: 'wp-1', email: 'Maria@Example.com' };
const whoApps = [
{ id: 'by-key', worker_profile_id: 'wp-1', email: 'nothing@else.com' },
{ id: 'by-email', worker_profile_id: null, email: 'maria@example.com' },
{ id: 'by-email-case', email: 'MARIA@EXAMPLE.COM' },
{ id: 'somebody-else', worker_profile_id: 'wp-2', email: 'other@example.com' },
];
const whoIds = records.applicationsForProfile(whoApps, whoProfile).map((a) => a.id);
record('an application finds its person by foreign key',
whoIds.includes('by-key'), 'worker_profile_id');
record('...and falls back to email, case-insensitively, for older rows',
whoIds.includes('by-email') && whoIds.includes('by-email-case'),
'citext identity, as worker_profiles already asserts');
record('...without collecting somebody else',
!whoIds.includes('somebody-else') && whoIds.length === 3, whoIds.join(', '));
record('...and asks for nobody when there is no profile',
records.applicationsForProfile(whoApps, null).length === 0, 'empty, not everything');
/* The reported symptom: a talent pool you could read and not use. */
const talentPoolSource = readFileSync(join(ROOT, 'src/pages/admin/talent-pool/nodes.jsx'), 'utf8');
record('a talent pool row opens that person',
/onRowClick=\{\(p\) => navigate\(`\/admin\/talent\/\$\{p\.id\}`\)\}/.test(talentPoolSource),
'DataTable wires the handler only when it is given one');
record('...and no action pretends to open something it does not',
!/toast\.info\(`Opening/.test(talentPoolSource),
'the "view profile" toast is gone');
record('...and a person can be put in front of a vacancy from the pool',
/setConsidering\(p\)/.test(talentPoolSource)
&& /AddToPositionModal/.test(talentPoolSource),
'the step that was missing between pool and position');
/* "Screened and waiting on a decision" includes the shortlisted. This counted
`ai_screened` alone, which read as correct only while nothing was ever
shortlisted — the first shortlisted candidate dropped out of the count in