Live on www.darwinator.com. Phases 1, 2 and 3 are shipped. Full V1 per the PRD also needs Phase 4 (multi-round tournaments and teams), which is on hold until you give the go-ahead.
Updated Nov 21, 2026, 4:37 AM EST · internal
Milestones
M1 · Core tournament loop
submit → evaluate → feedback report, real backend, authz.
Went live on www.darwinator.com, then shipped what the classroom pilot asked for.
Email + sign-in live
Resend domain verified; Google OAuth + magic link working on www.darwinator.com.
Deploy & migration hardening
deploys auto-apply Drizzle migrations (vercel-build) and prod's migration journal was reconciled — code can't ship ahead of its schema again.
Staging / preview environment
preview deploys run on a dedicated, isolated Neon branch (no longer the prod DB), auto-migrated per environment; Better Auth derives the per-branch preview origin and magic-link sign-in was verified working on the preview URL.
Inbound email (privacy@ / help@) forwarding
ImprovMX catch-all *@darwinator.com forwards to the operator inbox (MX + SPF live), so privacy@ — the address the privacy policy advertises — receives. Doesn't touch Resend's send.darwinator.com sending. See docs/EMAIL.md.
Launch-readiness review & final sweep
apex clean (308 → www); prod smoke passed — public pages 200, gated routes redirect to sign-in, auth + apex healthy. No open launch blockers remain.
Strip raw private-blob URLs from client DTOsGH #97/#108
profile/opportunity DTOs carry presence flags only; uploads return an opaque single-use server-side pending-upload reference (user- and tournament-scoped) instead of the blob URL; image read paths re-validate host-pinned ownership before any credentialed fetch.
Strip identifying metadata from uploaded imagesCodex audit
photo metadata (GPS location, camera make/model/serial, owner name, timestamps, embedded thumbnails, XMP/IPTC, comments) could reveal who submitted an idea in an anonymous tournament. Opportunity images and profile photos are now stripped losslessly before they are stored and again whenever they are served — same pixels, orientation and colour profile kept — and an older upload is re-saved clean in place the first time it is viewed (unless it is damaged, when it is only ever served clean).
Edit a submitted evaluationEVAL-6 · GH #124M
pilot feedback: participants reopen their own evaluations (“Edit last” right after saving, or “Your evaluations”) and revise ratings while evaluation is open. A save replaces the response set in place — never a second evaluation, so progress and coverage don't move — and locks when evaluation closes. Immutable ratings are one switch away.
Screen fixes from the visual reviewGH #171M
the participant report opens with the result (rank, quality, how peers rated each question, their comments), with the rater adjustments folded below it; the projector reads from the back of the room, with a QR code to join while a vote is open and a tally that counts up; and on a phone the rating screen has a one-line header and a Submit bar pinned within reach of the thumb.
A short pause per card, and a block for botsEVAL-8 · GH #165M
each card stays up 1 to 3 seconds, by how much it says, before its rating can be saved, and the server checks it, so nobody taps through the field unread. The Submit button counts down, so an honest rater never sees an error. A rating only a program would send blocks the rater (“Non-human behavior detected.”) until an administrator unblocks them on the Participants page, which says why.
Tournament themesGH #170M
Folio moves from the warm bone page to mist under its blue, and each tournament can recolor it from Settings: Karan green on sage, Cornell red on porcelain, Wharton blue on mist (Penn Red marks your own result), or the Abundant Intelligence Lab's navy and steel with a stripe on top. Colors only; every screen of the tournament follows, and a test holds every theme to WCAG AA.
Auto-advance, if you want itEVAL-4 · GH #163S
an Auto-advance checkbox beside Submit, off by default and remembered in the browser: a card with several questions submits itself a moment after the last answer. Submit stays, a started comment waits for it, and a slider answer counts when it's let go.
Swipe through Up/Down cards on a phoneEVAL-4 · GH #166M
from student feedback: on a touch screen, a card whose only question is Up/Down can be swiped right for Up or left for Down, recording exactly what the buttons record. The buttons and arrow keys stay, a short drag snaps back, and a card swiped before its pause is up waits at the edge. The rating screen became a deck: the next card is fetched while you read, its edge shows beneath, and a saved card leaves toward its answer. New tournaments start with a single Up/Down question, so their cards swipe with no setup.
Fewer, quieter labelsGH #173M
small mono capitals now mark only statuses, counts, codes and data, and none is set under 12px. Links, field labels and sentences use the plain sans, labels that repeated a heading are gone, and long explanations are one sentence with the rest behind a (?) button. A test keeps anything under 12px from coming back.
Motion that confirmsGH #174S
each screen's content fades in as it opens (180ms, the top bar and side nav stay put), and on the console's vote screen each row gets a tally bar that slides, figures that count to each poll's numbers and rows that glide to their new rank. The rated card's hand-off shipped with the swipe deck. Everything is instant under reduced motion.
Line icons instead of text symbolsGH #175S
back and onward chevrons, a check for saved or ready, a dot for live or open, a ring for closed, sort arrows, and thumbs beside the default Up and Down: one thin line set vendored from Lucide, drawn in the text's color so it follows each theme. It replaces the text symbols (angle quotes, check marks, dots and triangles) that looked different on every phone, and a test keeps them out.
Dark mode that follows the phoneGH #176S
a phone or laptop set to dark gets dark screens, in CSS alone so nothing flashes on load. Each of the five themes has its own dark colors, held to the same contrast tests as light; buttons turn a lighter accent with dark labels, and the browser's bar follows. The projector stays light, since it's the room's screen, and emails stay light.
A front page that explains DarwinatorGH #172S
darwinator.com now says what a tournament is in about 100 words: a headline, three steps, four live features and what's coming next, beside the real rating screen on a phone (light or dark, as the device is). A shared link shows a preview card with the same headline. Only live features are claimed; the AI assistant connector stays under Coming next until it ships.
Phase 1
Facilitator admin
Let facilitators run a tournament end-to-end — the most-expected gaps.
Edit a tournament after creationCONFIG-1–17L
settings screen for basics, submission rules, and access; admin-gated + validated.
Edit evaluation questionsCONFIG-13M
edit the question set on the settings screen, even while Evaluate is open, until the first evaluation is recorded; then read-only for good (changing scales/weights would rescore recorded data). The lock check and the replace are one atomic write, so a rating landing mid-save is never erased, and a rater whose screen predates an edit is re-asked on the new questions, keeping answers to the unchanged ones.
Roster managementCONFIG-17M
remove a participant (submissions retained, and any pending invitation to them withdrawn); add people by email from the invitations card.
Self-registration linkJOIN-5S
shareable /join/<code> — joins a signed-in visitor, or signs them in and returns. New tournaments get a random, easy-to-read code; codes are held to letters, digits, hyphens and underscores so the link survives sign-in; the copyable link uses the same canonical address as invitation emails; a stale link explains itself; and joining by code opens the tournament.
Email invitesJOIN-6M
paste a class list (commas, semicolons, new lines, or Name <address> all work; up to 500 at once) with an optional note that is kept for next time. Emails go out in batches, and the screen says which ones didn't send, with Try again and Resend. Each invitation carries its own link, so a student who signs in with a different address is still joined and shown as 'Joined as …'; pending invitations can be resent or withdrawn. Invitees are also auto-joined on sign-in when the address matches. On Resend Pro there's no daily cap. If the site goes back to Resend's free plan, one setting turns on a daily email allowance that keeps 30 of its 100 daily emails for sign-in links: sends past it are saved to resend after the reset, and the box says how many emails are left today. Admin-gated.
Co-admins with a distinct ownerPRD §6M
Pilot ask: a tournament has one owner (its creator) and any number of co-admins, e.g. teaching assistants, who can run everything. Only the owner makes or removes co-admins (Make co-admin on the roster) and can hand ownership to one; a co-admin can step down. Enforced and re-checked server-side inside each write, with database guarantees of exactly one owner, and an activity log of who changed the admin team and when. The owner-only list is a proposal in one table.
Invite a co-admin by emailPRD §6 / JOIN-6S
Pilot ask: the owner invites a co-admin by email from the Administrators card. The invitation works only for the invited, verified address, by its link or by signing in, so a forwarded email can't hand out admin rights. Accepting makes them a co-admin, or upgrades a participant, and is logged. The owner alone resends or withdraws one, and removing someone's co-admin rights withdraws a pending one.
Superuser mode for supportPRD §6 / §10.3M
Pilot ask: a platform operator can switch on superuser mode at an unlisted page, open any tournament's console and do anything its owner can, to fix problems. Only accounts on a short list in the code qualify, with a verified email, re-checked on the server every request; the switch is tied to the sign-in and lasts 8 hours. It never joins the tournament, so it isn't served, counted or emailed. Every visit and change is logged before it runs (and refused if it can't be), changes show as Darwinator support, and console pages show a banner while it's on. The privacy page says so.
Scheduled phase changesCONFIG-5/6M
Set when Submit, Evaluate and Results open and close, on the Live console, in your own time zone (named on the form; a date with no time means 12:00 AM). Each phase then changes on time by itself, exactly as if you had clicked, whether or not anyone is online; a change by hand stands until the next scheduled one. Participants see the deadlines on the tournament page in their own time zone.
Deterministic E2E seed routeS
test-only /api/test/seed stands up a known dataset (salvaged from PR #47); double-gated like test-login.
Phase 2
Reporting & grading
Get judged results out of the app for grading and analysis.
CSV / grade exportREPORT-8M
admin-only route exports a submissions spreadsheet (with descriptions + owner emails) and a grade-ready per-participant report; identities resolved for grading, formula-injection safe.
pilot feedback: each question shows its configured scale + low/high labels beside the result; rank, percentile, QUALITY and the histogram get plain-language captions and (?) help; no-/few-ratings states (fewer than 5 raters = early signal — a proposal, one swappable constant). Raw vs adjusted kept explicit; no scoring change.
Phase 3
Comms & polish
Round out communication, accounts, and accessibility.
Broadcast email to membersNOTIFY-4M
Manage → Message: pick an audience (everyone / participants / behind-on-evaluations, with live counts), write a subject + body, send. Admin-gated; recipients resolved server-side (addresses never reach the client); multipart email. A large group goes out 100 at a time, waiting out the email service's per-second limit. If the daily email allowance is on (Resend's free plan only), a message bigger than what's left of it is refused before anything is sent, with the count.
Deadline reminder emailsNOTIFY-2/3M
about 24 hours before Submit or Evaluate closes on a schedule, each participant who is behind (fewer opportunities than the minimum, or fewer evaluations than their target) gets one email with the deadline in their own time zone and a link straight to the task. Never sent twice; counted against the daily email allowance; listed with who got them on Manage → Message, where an admin can turn them off.
Account / profile screenAUTH-5/6M
/account edits name, display name, organization, market segment, and profile photo (private blob, streamed through an authorized route); verified read-only email; sign-out. Facilitator roster shows member avatars.
Passkey sign-inAUTH-9M
sign in with Face ID / Touch ID / a device PIN or password manager — a button plus passkey autofill on /signin; add, rename, and remove passkeys on /account (needs a recent sign-in, never creates an account). Bound to darwinator.com with origins pinned (ADR-0003); real-ceremony tests incl. phishing-origin, replay, and cross-user cases.
Class-scale sign-inAUTHS
a class of about 120 behind one campus network address can request email links, confirm them, and start Google sign-in together (Better Auth's defaults allowed 5 links a minute and 3 Google starts per 10 seconds). A per-mailbox cap (5 unused links per 15 minutes, counted in the database, with plus-tags and Gmail dots folded together) keeps one inbox from being flooded, with its own plain-language message; a link the email provider failed to send doesn't count.
Soft-delete & vote-integrity auditDISC-3S
multi-agent audit confirmed report math + serving exclude soft-deleted data; fixed a real bug where votes on a moderated-away opportunity inflated the live voter count and consumed a pick slot (the PR #47 class); locked with tests.
Participants withdraw their own opportunitiesSUBMIT-4 / DISC-3S
Pilot ask (GH #125): Remove on the submit screen, with an inline confirmation, until evaluation first opens — then locked for good (even if evaluation closes again), as it is once anyone has evaluated, voted on, or commented on it, with the reason shown. A soft delete with an audit stamp (who + why, shared with admin removal), race-safe against in-flight evaluations, votes, and edits; the opportunity leaves serving, progress, submission limits, voting, and reports, and frees its slot.
Facilitator removes an opportunity; reopen a phase safelyDISC-3 / CONFIG-7M
Pilot ask (GH #127): an inline-confirmed Remove on the results table for administrators, at any phase, with a removed-opportunities audit list (who, when, how many evaluations it had). Reopening Submit or Evaluate now previews what changes — nothing is deleted, no one rates twice, results move live — and asks to confirm; ideas peers have already responded to stay locked from author edits. A rater whose open card is removed is moved on with a notice.
Fresh restart of a tournamentCONFIG-7 / DISC-3M
Pilot ask (GH #127), decided by Karan: clear the ideas, keep the people and the setup. A Settings danger zone previews what a restart archives, closes, and keeps, then asks for the tournament's name. One atomic, admin-only write archives every live opportunity with its evaluations, comments, and votes (hidden, never deleted), closes every phase and any live vote, and reopens author withdrawal; the join code, roster, settings, and questions stay. The questions become editable again without losing the archived ratings, and Results notes each restart.
Accessibility audit of new screensM
WCAG 2.2 AA pass (multi-agent + second-opinion review) over settings/questions, results export, report builder, and account: locked questions made keyboard-readable, dynamic errors announced, decorative glyphs hidden.
Privacy policy + FERPA reviewPRD §11 / XC-PRIVS
plain-language /privacy linked at sign-up (no analytics/ads/AI/repurposing). FERPA review cleared classroom use — students self-register, so no DPA/HECVAT prerequisite; deletion + export specified for later.
Phase 4
Multi-round & teams
Complete the tournament depth the data model already anticipates.
Round advancement & multi-roundROUND-2/3/4L
pick winners to advance, create later rounds, allow edits/new submissions per config. Schema supports it; no UI/ops yet.
Teams: enter teams, team-wide resultsROUND-5M
Karan's ask (Nov 2026), not built yet: an administrator says who is on which team, painlessly. Teammates are then never served their own team's ideas, and every teammate sees the team's results, not just the person who submitted. The membership's team field and team-aware serving already exist; what's missing is a painless way to enter teams and team-wide results.
Later · post-V1 (PRD §17)
14
Public tournament searchJOIN-4
a searchable listing of public tournaments. Until it exists the Public setting is hidden and new tournaments are saved as not public; anyone with the code or link can join either way.
Rater-corrected rank and percentileREPORT-6
“Remove rater fixed effects” adjusts each question's scores today; rank, percentile, quality and the histogram still use the raw ratings.
More attachment typesCONFIG-10
documents, video, audio and links on an opportunity. Today an opportunity can carry one optional image.
Full administrator activity log
a log of every administrator action (phases, settings, removals, votes, messages). Today admin-team changes, removals and restarts already record who made them.
Optional anonymous comments
a “comment anonymously” checkbox on the rating card. Administrators still see who wrote every comment, and comments stay named by default.
Deadline-bias serving
weight the random pick toward below-floor opportunities as a deadline nears.