Pure function reused by upcoming mentor.getCandidates + AI fallback path.
Refactors getAlgorithmicMatches to call it. No behavior change.
Plan: docs/superpowers/plans/2026-04-28-pr4-mentor-assignment-ux.md
Adds generateMentorObjectKey helper producing
<projectName>/mentorship/<timestamp>-<file>. Replaces the
client-supplied bucket/objectKey on workspaceUploadFile with an
HMAC-signed upload token that binds bucket, objectKey, uploader,
and a 1h expiry — paths can no longer be forged from the client.
Adds workspaceGetUploadUrl, workspaceGetFiles,
workspaceGetFileDownloadUrl, workspaceDeleteFile procedures with
mentor-or-team-member auth. Builds <WorkspaceFilesPanel> and
wires it into the mentor workspace Files tab and the applicant
/applicant/mentor page. Replaces the file-promotion-panel mock
array with a real workspaceGetFiles query.
Tests cover token sign/verify (5), key construction (5), and
end-to-end procedure flow including auth + tampered tokens (7).
Spec: docs/superpowers/specs/2026-04-28-mentor-round-readiness-design.md §F.1
Plan: docs/superpowers/plans/2026-04-28-pr2-mentor-workspace-files.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The dashboard now computes its own composite ranking score on the
client, blending (balanced-or-raw) average score with (balanced-or-raw)
advance pass rate via the existing scoreWeight / passRateWeight
sliders. Both inputs are toggled independently:
- 'Balance juror grading style (score)' — existing useBalancedRanking
- 'Balance juror approval rate (advance vote)' — new useBalancedPassRate
Both default to true and persist per-round. The pass rate is balanced
the same way scores are: each juror's personal yes-rate gives them a
Bernoulli stddev, each vote is z-normalized against that, and the
project's mean z is rescaled to the round's overall yes rate. A 'yes'
from a juror who rarely says yes counts more than a 'yes' from a
lenient juror.
List rows now show two chips — score (Bal/Raw X.XX) and pass rate
(Bal Yes% / Yes% N%) — so admins can see what's driving the order.
The threshold cutoff and live re-sort effect both use the same
composite formula.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Previously the edition-level branch of analytics.getProjectRankings
(programId mode) pooled every juror's evaluations across every round
into a single z-normalization context. A juror's mean and stddev are
not stable across round types — quick intake screening produces a
very different grading profile than a deep evaluation round, and
mixing them yields a meaningless personal calibration.
The rollup now groups points by roundId, computes one balance context
per round, and aggregates per-project as the unweighted mean of the
per-round balanced averages. roundId mode is unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a shared juror-balancing utility (z-score normalization per juror,
rescaled back onto the raw 1-10 scale) and wires it into:
- Admin reports page: Top-10 project table now shows "Raw Avg" and
"Balanced" columns side by side, and the summary stats row shows a
balanced-average tile. Sort defaults to balanced so harsh and lenient
graders no longer skew the ranking.
- Ranking dashboard: each project row shows a green/amber balanced-score
chip next to the raw average when the two differ by ≥0.05, making it
obvious when juror calibration moved a project's effective ranking.
Also adds AI Juror Calibration Advisory — a mutation that takes
anonymized per-juror stats, calls OpenAI, and produces a plain-language
explanation of the cohort's grading patterns plus per-juror severity
(normal / notable / outlier) with a one-sentence narrative. The advisory
describes the statistical balance that already runs; it does not
introduce a new weighting layer. Rendered as a panel in the Juror
Consistency tab when a specific round is selected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The ranking dashboard showed 0/X Yes for every project in rounds using
the 'advance' criterion type because the matcher only looked for
type === 'boolean' with a "move to the next stage" label.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Files uploaded by admins with roundId but no requirementId were not
counted in the finalization page Docs column.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Project names now link to their detail page on all finalization tabs.
Submission/intake rounds show a docs submitted/required column.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add ability to define completely different evaluation criteria for each
competition category. Admins toggle "Separate Criteria per Category" in
round config, then configure criteria independently via tabbed editor.
- Schema: add nullable `category` to EvaluationForm with updated constraints
- Config: add `perCategoryCriteria` boolean to EvaluationConfigSchema
- Helper: new `findActiveForm()` with category-aware resolution + fallback
- Backend: getForm, upsertForm, getStageForm, startStage all category-aware
- AI services: use project category for form lookup in summaries + ranking
- Export/ranking: merge criteria from all active forms for cross-category reports
- Admin UI: toggle switch + tabbed criteria editor with per-category builders
- Jury UI: auto-selects correct form based on project category (invisible to juror)
- Fully backwards compatible: toggle defaults OFF, existing forms unchanged
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Schema:
- Drop 4 dead models: OverrideAction, NotificationPolicy, AssignmentException, AdvancementRule
- Drop 2 dead enums: OverrideReasonCode, AdvancementRuleType
- Drop 3 stale columns: Project.roundId, ConflictOfInterest.roundId, Evaluation.version
- Remove 3 back-relation fields from User, Assignment, Round
Code:
- Fix 6 COI queries in assignment.ts + 1 in juror-reassignment.ts
(roundId filter → assignment.roundId after column drop)
- Remove orphaned Project.roundId write in project.ts createProject
- Remove advancementRules include from round.ts getById
- Remove AdvancementRule from RoundWithRelations type
- Clean up seed.ts (remove advancement rule seeding)
- Clean up tests/helpers.ts (remove dead model cleanup)
- Add TODO comments on user delete mutations (FK violation risk)
Migration: 20260308000000_drop_dead_models_and_stale_columns
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Webhook HMAC: added consumer verification JSDoc with Node.js example using crypto.timingSafeEqual
- CSRF rate limiting: 20 requests/15min per IP on NextAuth /csrf endpoint
- Renamed withRateLimit to withPostRateLimit/withGetRateLimit for clarity
- 429 responses include Retry-After header
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add escapeHtml() helper and apply to all user-supplied variables in 20+ HTML email templates
- Auto-escape in sectionTitle() and statCard() helpers for defense-in-depth
- Replace 5 instances of incomplete manual escaping with escapeHtml()
- Refactor bulkInviteTeamMembers: batch all DB writes in $transaction, then send emails via Promise.allSettled with concurrency pool of 10
- Fix inner catch block in award-eligibility-job.ts to capture its own error variable
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- MinIO: use separate public client for presigned URLs so AWS V4 signature
matches the browser's Host header (fixes SignatureDoesNotMatch on all uploads)
- Consolidate applicant/partner uploads to mopc-files bucket (removes
non-existent mopc-submissions and mopc-partners buckets)
- Auth: allow magic links for any non-SUSPENDED user (was ACTIVE-only,
blocking first-time CSV-seeded applicants)
- Auth: accept invite tokens for any non-SUSPENDED user (was INVITED-only)
- Ensure all 14 invite token locations set status to INVITED
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Custom body support for advancement/rejection notification emails, evaluation
config toggle fix, user actions improvements, round finalization with reorder
support, project detail page enhancements, award pool duplicate prevention.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- processRoundClose now applies reordersJson drag-reorder overrides
when building the evaluation pass set (was ignoring admin reorders)
- Finalization tab groups proposed outcomes by category (Startup/Concept)
with per-group pass/reject/total counts
- Added category filter dropdown alongside the existing outcome filter
- Removed legacy "Advance Top N" button and dialog from ranking page
(replaced by the finalization workflow)
- Fix project edit status defaultValue showing empty placeholder
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- processRoundClose EVALUATION uses ranking scores + advanceMode config
(threshold vs count) to auto-set proposedOutcome instead of defaulting all to PASSED
- Advancement emails generate invite tokens for passwordless users with
"Create Your Account" CTA; rejection emails have no link
- Finalization UI shows account stats (invite vs dashboard link counts)
- Fixed getFinalizationSummary ranking query (was using non-existent rankingsJson)
- New award pool notification system: getAwardSelectionNotificationTemplate email,
notifyEligibleProjects mutation with invite token generation,
"Notify Pool" button on award detail page with custom message dialog
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add scoreWeight and passRateWeight (0-10) to evaluation config for
configurable composite score formula. When ranking criteria text is
empty, triggerAutoRank uses pure formula ranking (no LLM calls).
When criteria text is present, AI-assisted ranking runs as before.
- Add FORMULA to RankingMode enum with migration
- Extract fetchCategoryProjects helper, add formulaRank service
- Update computeCompositeScore to accept configurable weights
- Add score/pass-rate weight sliders to ranking dashboard UI
- Mode-aware button labels (Calculator/formula vs Sparkles/AI)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Sort all ranked projects by compositeScore descending so highest-rated
projects always appear first (instead of relying on AI's inconsistent rank order)
- Deduplicate AI ranking response (AI sometimes returns same project multiple times)
- Deduplicate ranking entries and reorder IDs on dashboard load as defensive measure
- Show advancement cutoff line only once (precompute last advancing index)
- Override badge only shown when admin has actually drag-reordered (not on fresh rankings)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When a round is reopened (ROUND_CLOSED → ROUND_ACTIVE), the old
windowCloseAt was still in the past, causing jury submissions to fail
with "Voting window has closed" even though the round status was active.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- AI ranking now includes ALL projects (never filters/excludes any)
- Updated system prompt: filter criteria inform priority, not exclusion
- Dynamic maxTokens scaling for large project pools (80 tokens/project)
- Fallback: projects AI omits are appended sorted by composite score
- Override badge uses snapshotOrder state (synced with localOrder in same
useEffect) instead of rankingMap.originalIndex to prevent stale-render
mismatch where all items incorrectly showed as overridden
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Backfilled 166 evaluations' binaryDecision from criterionScoresJson on production DB
- Fixed roundEvaluationScores and ai-ranking to look in EvaluationForm.criteriaJson
instead of round.configJson for the boolean "Move to the Next Stage?" criterion
- Added advanceMode (count/threshold) toggle to round config Advancement Targets
- Added "Assign to Jurors" button on Unassigned Projects section and Projects tab bulk bar
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add criteriaWeights to EvaluationConfig for per-criterion weight assignment (0-10)
- Rewrite ai-ranking service: fetch eval form criteria, compute per-criterion averages,
z-score normalize juror scores to correct grading bias, send weighted criteria to AI
- Update AI prompts with criteria_definitions and per-project criteria_scores
- compositeScore uses weighted criteria when configured, falls back to globalScore
- Add collapsible ranking config section to dashboard (criteria text + weight sliders)
- Move rankingCriteria textarea from eval config tab to ranking dashboard
- Store criteriaWeights in ranking snapshot parsedRulesJson for audit
- Enhance projectScores CSV export with per-criterion averages, category, country
- Add Export CSV button to ranking dashboard header
- Add threshold-based advancement mode (decimal score threshold, e.g. 6.5)
alongside existing top-N mode in advance dialog
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add adminEditEvaluation mutation and getJurorEvaluations query
- Create shared EvaluationEditSheet component with inline feedback editing
- Add Evaluations tab to member detail page (grouped by round)
- Make jury group member names clickable (link to member detail)
- Replace inline EvaluationDetailSheet on project page with shared component
- Fix project status transition validation (skip when status unchanged)
- Fix frontend to not send status when unchanged on project edit
- Ranking dashboard improvements and boolean decision converter fixes
- Backfill script updates for binary decisions
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Added AI_RANKING_COMPLETE and AI_RANKING_FAILED to NotificationTypes const
- Added BarChart3 / AlertTriangle icons in NotificationIcons
- Added normal / high priorities in NotificationPriorities
- Add roles UserRole[] to User model with migration + backfill from existing role column
- Update auth JWT/session to propagate roles array with [role] fallback for stale tokens
- Update tRPC hasRole() middleware and add userHasRole() helper for inline role checks
- Update ~15 router inline checks and ~13 DB queries to use roles array
- Add updateRoles admin mutation with SUPER_ADMIN guard and priority-based primary role
- Add role switcher UI in admin sidebar and role-nav for multi-role users
- Remove redundant stats cards from round detail, add window dates to header banner
- Merge Members section into JuryProgressTable with inline cap editor and remove buttons
- Reorder round detail assignments tab: Progress > Score Dist > Assignments > Coverage > Jury Group
- Make score distribution fill full vertical height, reassignment history always open
- Add per-juror progress bars to admin dashboard ActiveRoundPanel for EVALUATION rounds
- Fix evaluation submit bug: use isSubmitting state instead of startMutation.isPending
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Backend: getProjectRoundStates now includes assignment counts and submitted
evaluation counts per project. Frontend: new Reviews column shows X/Y
(submitted/total) with green highlight when all reviews are complete.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add mail/transfer/reshuffle/redistribute icons to each juror row in Members card
- New redistributeJurorAssignments procedure: reassign all pending projects without dropping juror from group
- New DROPOUT_REASSIGNED email template with project names, deadline, and dropped juror context
- Update reassignDroppedJuror to send per-juror DROPOUT_REASSIGNED emails instead of generic BATCH_ASSIGNED
- Transfer dialog now shows all candidates with "Already assigned" / "At cap" labels instead of hiding them
- SQL script for prod DB insertion of new notification setting without seeding
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add COI_REASSIGNED and MANUAL_REASSIGNED notification types with distinct
email templates, icons, and priorities
- COI declaration dialog now shows a confirmation step warning that the
project will be reassigned before submitting
- reassignAfterCOI now checks historical assignments (all rounds, audit logs)
to never assign the same project to a juror twice, and prefers jurors with
incomplete evaluations over those who have finished all their work
- Admin transfer (transferAssignments) sends per-juror MANUAL_REASSIGNED
notifications with actual project names instead of generic batch emails
- docker-entrypoint syncs notification settings on every deploy via upsert
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rewrite ctaButton to use td-background pattern (works in all clients
including Outlook, Gmail, Yahoo, Apple Mail) instead of VML/conditional
comments that broke link clicking in Outlook desktop
- Add plaintext fallback URL below every CTA button so users always have
a working link even if the button fails
- Add getBaseUrl() and ensureAbsoluteUrl() helpers in email.ts to
guarantee all email links are absolute https:// URLs
- Apply ensureAbsoluteUrl safety net in sendStyledNotificationEmail and
sendNotificationEmail so relative paths can never reach email templates
- Standardize all NEXTAUTH_URL fallbacks to https://portal.monaco-opc.com
(was inconsistently http://localhost:3000 or https://monaco-opc.com)
- Fix legacy notification.ts: wrong argument order in
sendJuryInvitationEmail (URL was passed as name parameter)
- Fix legacy notification.ts: missing NEXTAUTH_URL fallback for
evaluation reminder URL construction
- Change tooltip styling from red bg to white bg with black text
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Relative linkUrl paths (e.g. /jury/competitions) were passed as-is to
email templates, causing email clients to interpret them as local file
protocols (x-webdoc:// on macOS). Now prepends NEXTAUTH_URL to any
relative path before sending.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Send Reminders button now works: added sendManualReminders() that bypasses
cron-specific window/deadline/dedup guards so admin can send immediately
- Added Notify Jurors button that sends direct BATCH_ASSIGNED emails to all
jurors with assignments (not dependent on NotificationEmailSetting config)
- Fixed checkbox component: default border is now neutral grey (border-input),
red border (border-primary) only applied when checked
- Widened Add Assignment dialog from max-w-2xl to max-w-3xl to prevent overflow
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- specialAward.setEligibility: add ensureUserExists() guard and use Prisma
connect syntax to prevent FK violation on stale session user IDs
- specialAward.confirmShortlist: same ensureUserExists() guard for confirmedBy
- Round projects table: add Country column showing project origin
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instead of 10 sequential GPT calls (which timeout with GPT-5.1 on 99
projects), use a two-phase approach:
Phase 1 - AI Scoring: ONE API call asks GPT to score each juror's
affinity for all projects, returning a compact preference matrix with
expertise match scores and reasoning.
Phase 2 - Algorithm: Uses AI scores as the preference input to a
balanced assignment algorithm that assigns N reviewers per project,
enforcing even workload distribution, respecting per-juror caps, and
filling coverage gaps.
Benefits:
- Single API call eliminates timeout issues
- AI provides expertise-aware scoring, algorithm ensures balance
- Truncated response handling (JSON repair) for resilience
- Falls back to tag-based algorithm if AI fails
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Cap maxTokens at 12000 (was unlimited dynamic calc that could exceed model limits)
- Replace massive EXISTING array with compact CURRENT_JUROR_LOAD counts and
ALREADY_ASSIGNED per-project map (keeps prompt small across batches)
- Add coverage gap-filler: algorithmically fills projects below required reviews
- Show error state inline on page when AI fails (red banner with message)
- Add server-side logging for debugging assignment flow
- Reduce batch size to 10 projects
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Problems:
- GPT only generated 1 reviewer per project despite N being required
- maxTokens (4000) too small for N×projects assignment objects
- No fallback when GPT under-assigned
Fixes:
- System prompt now explicitly explains multiple reviewers per project
with concrete example showing 3 different juror_ids per project
- User prompt includes REVIEWS_PER_PROJECT, EXPECTED_OUTPUT_SIZE
- maxTokens dynamically calculated: expectedAssignments × 200 + 500
- Reduced batch size from 15 to 10 (fewer projects per GPT call)
- Added fillCoverageGaps() post-processor: algorithmically assigns
least-loaded jurors to any project below required coverage
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Root cause: batches of 15 projects were processed independently -
GPT didn't see assignments from previous batches, so expert jurors
got assigned 18-22 projects while others got 4-5.
Fixes:
- Track cumulative assignments across batches (feed to each batch)
- Calculate ideal target per juror and communicate to GPT
- Add post-processing rebalancer that enforces hard caps and
redistributes excess assignments to least-loaded jurors
- Calculate sensible default max cap when not configured
- Reweight prompt: workload balance 50%, expertise 35%, diversity 15%
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix project documents not displaying on jury project page (rewrote MultiWindowDocViewer to use file.listByProject)
- Add working download/preview for project files via presigned URLs
- Display project tags on jury project detail page
- Add autosave for evaluation drafts (debounced 3s + save on unmount/beforeunload)
- Support mixed criterion types: numeric scores, yes/no booleans, text responses, section headers
- Replace inline criteria editor with rich EvaluationFormBuilder on admin round page
- Remove COI dialog from evaluation page
- Update AI summary service to handle boolean/text criteria (yes/no counts, text synthesis)
- Update EvaluationSummaryCard to show boolean criteria bars and text responses
- Add evaluation detail sheet on admin project page (click juror row to view full scores + feedback)
- Add Recent Evaluations dashboard widget showing latest jury reviews
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Round open/close scheduling now runs as a 60s setInterval inside the
app process (via instrumentation.ts register hook) instead of needing
an external crontab. Removed the /api/cron/round-scheduler endpoint.
- DRAFT rounds auto-activate when windowOpenAt arrives
- ACTIVE rounds auto-close when windowCloseAt passes
- Uses existing activateRound/closeRound from round-engine
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
