What's new

The AI review now reads the actual code, not just the description. When a thinkschooler submits a code piece with a GitHub link, a background worker clones the repo and grades the real implementation — the files they pointed at (or the sub-folder in a /tree/ link) — instead of only the submission text. The recommendation you see in the review queue is now grounded in the code itself; in early testing it correctly failed a submission that was just a generic template repo. No candidate code is executed (it only reads the files). If a repo can't be reached (private with no access, deleted), it falls back to routing the submission to you as before. Applies to human-graded code tracks (starting with .NET); written pieces are unchanged.

New “Completed” status for learners who finished a course — separate from Suspend. When a cohort wraps, use Admin → Users → ⋯ → Mark completed. A completed learner loses sign-in access but stays in the cut, interview scoring, and graduation for the hiring/credential decision. Suspend is now reserved for revoking access for cause — a suspended learner is excluded from those decision views and can't be graduated. This removes the old overload where “suspended” meant both “finished” and “blocked,” which had hidden finished learners from the cut. Existing suspended learners were migrated to Completed. Reactivate either from the same ⋯ menu.

thinkbridge Academy has a new home: thinkbridge.academy. The platform moved to its own domain and a new server. Please update your bookmark to thinkbridge.academy; the old thinkschool.zpqv.io address has been retired. Your account, progress, and cohort are unchanged — just sign in again on the new domain with your usual magic link.

Suspending a finished learner no longer hides them from the cut or interview scoring. When a cohort wraps, you can suspend learners' access (so they can no longer sign in) and they will still appear in the biweekly cut ranking and the interview roster, and interviewers can still record their scores. Only withdrawn and dropped learners are excluded from those decision views — suspension is access-only. (Previously, suspending a learner removed them from the cut and made their interview scores unviewable/unrecordable.)

A thinkschooler with no cohort no longer lands on the wrong course. Previously, a thinkschooler who existed but was not yet assigned to any cohort fell through to a default track (the .NET course) on their home and map. Now they see a clear "you are not in a cohort yet" message instead, until an admin adds them to one. No effect on anyone already in a cohort.

New assessment: the Direct the Build screen. A 60-minute, one-sitting test for the skill of directing AI-built software from a vague goal and verifying it before trusting it. The candidate gets a messy client engagement (MidStates Freight) and an AI-built feature declared "done, tests pass" that is silently broken — it reports zero conflicts on data that actually contains two double-bookings (one hidden by a timezone bug, one by a truck-id mismatch). They pick the plan that fixes the real problem, prove whether the slice is safe to ship, and write the exact fix instruction. Scored on two unfakeable signals: seeking independent proof, and taking the defensible cut. It's a timed, one-shot, human-graded track (the AI reviewer drafts a rubric score into the assessor's queue). Seed it from Admin → Item banks → Seed / refresh Direct the Build screen, then run a cohort on the track. Full design in content/direct-the-build/spec.md.

Seed buttons for the remaining cohort tracks. Admin → Item banks now has one-click Seed / refresh buttons for AI Consulting, CS Lead, Delivery Principal, Web + Firebase, and Flutter + Firebase, alongside the existing screen-bank, prep-course, BSE, and thinkOne Sales seeds. These five are additive — they add any new pieces by slug and never remove existing ones — so re-running is always safe. (Previously these tracks could only be seeded via the API.)

Adding an existing user to a cohort now emails them. Previously, only brand-new users got an invite e-mail; adding someone who already had an account (e.g. enrolled in another cohort) to a new cohort was silent, so they were never told. Now any add sends a short "you've been added to <cohort>" note with a magic link that signs them straight in (their other cohorts are unaffected). Best-effort and audit-logged, so you can see whether each send succeeded. Reassignment and removal don't trigger e-mails.

Every cohort now has a default mentor, so no review queue is ever unrouted. New cohorts automatically get [email protected] added as a mentor, and any thinkschooler enrolled without a specific mentor is assigned to them by default — their submissions land in that queue instead of nobody's. Existing cohorts were backfilled the same way (only learners who had no mentor; anyone already assigned was left untouched). You can reassign any learner to another mentor at any time from the cohort membership controls.

New internal track: thinkOne Sales Certification. A self-paced certification for the sales team on selling thinkOne (the custom-fit, AI-built ERP alternative), built from the ERP primer deck, the question banks, and the 90%-pass certification tests. Three modules — ERP economics & the market problem, the build-vs-buy framework & qualification, and objection handling & competitive positioning — across 22 written application pieces. Every piece is graded by the AI reviewer against the tests' own rubrics (it fails generic, recall-only, or overclaiming answers and demands the specific commercial reasoning), so reps retry until they genuinely get it. Seed it from Admin → Item banks → Seed / refresh thinkOne Sales track, then run a cohort on the track.

Business Software Engineer track tightened for experienced engineers. The BSE curriculum was retuned for someone with 4+ years who already ships with AI agents: 3 weeks instead of 6, 14 pieces instead of 29, and build-heavy — about six pieces are now real build/verify work on the running Northwind system (a test contract + guardrail file, a migration-shaped data model, a directed agent build that must pass the contract, failure-mode hardening, a defect/schema hunt, and an incident drill on the learner's own code). The short written warm-ups were folded into denser, time-boxed deliverables; the capstone spine and full role-trait coverage are unchanged. Re-seed it from Admin → Item banks → Seed / refresh BSE track (new button; it reports what it created and now prunes any pieces it no longer defines).

Multi-track learners land on their curriculum, not an empty page. If you're enrolled in both Build With AI and a cut track (e.g. Business Software Engineer), your Home now opens directly on the cut track's Today — the first piece and the Begin button — instead of defaulting to Build With AI (a self-serve course with no Today) and showing "All caught up." Build With AI stays reachable from its own top-nav link, and the track switcher only lists real curriculum tracks.

New hiring track: Business Software Engineer. An 8-module / 29-piece curriculum for the premium AI-amplified role — direct and verify AI-built business software across its whole life: read the business → reach for code last → spec + build → ship it live → care for it, all on one services-ops capstone (Northwind Field Services). It's a cut track (6 weeks, human-graded) built against the v2 role definition. Seed it with the seed-bse admin endpoint, then run a cohort on it. New business_software competency bucket (9 traits) backs it.

Your Account page now shows your active sessions. Under /account/ you can see every device/browser signed in as you, and sign out everywhere else in one click (this device stays signed in) — handy if you used a shared or public computer.

A daily heartbeat digest now posts to Teams. Each morning a summary card lands in Teams — active cohorts, pending reviews, idle students (≥2 days), and the last 24h of submissions, passes, and AI cost. The admin Home tab also has a Send digest now button to fire one on demand.

Each exercise now shows what it builds. Under the piece, a row of competency chips (e.g. "GitHub Copilot flow", "Claude Code (agentic)") tells you which skills it's developing — so the point of each day is legible, not just the task.

New My students view on the mentor page. Next to your Queue, a roster of every thinkschooler you mentor — pieces passed, pending count, days idle, and behavior net — sorted so whoever needs attention is on top. Click a student to see their full trajectory (submissions by piece, behavior observations, your notes) and log a behavior observation right there, without waiting for a review.

"Build With AI" is now open to thinkschoolers, not just applicants. A cohort can be run on the Build With AI track: any thinkschooler enrolled in a build-with-ai cohort now sees the Build With AI link in the top nav and can take the course at /learn/ (the page and its API used to be applicant-only). Thinkschoolers on other tracks are unaffected.

New Submissions tab — see every member's work by Day. Pick a cohort, then a Day/piece, and get one row per member: their status, the AI verdict, and how the mentor scored it. Expand any row to read the actual submission, the reflections (with the AI-likeness read), the AI's reasoning, and the mentor's note. You can also review and override a decision right there — pass, fail, or send back for more info, with a note — and it updates the student's progress (every override is audit-logged).

The reflection prompts are now type-only. The two reflections under each exercise ("what did you learn", "what would break this") no longer accept pasted or dragged-in text — write them in your own words. They're optional and a line or two is plenty; the point is your own thinking, which is what earns the mentor extra credit.

Interview assignment now shows who interviewed each candidate in previous rounds. The coverage table has a Previous rounds column (e.g. "R1 Shubham 79 · R2 Anurag 55"), and the assign dropdown flags any interviewer who's already seen that candidate ("— did R1/R2") — so you can rotate raters and avoid repeating an interviewer for the new round.

The Cut screen now auto-selects the cut — just pick the cohort. No more choosing a cycle: the cut resolves to however many interview rounds have been scored (cut 1 after round 1, cut 2 after round 2, …) and the table + scores update to match. The drop threshold auto-defaults to that cut's bar (50 / 60 / 70 / 75 / 80) and stays fully editable if you want a different line.

The biweekly cut now scores interviews across all rounds, not just the selected cycle. The Cut tab's Interview column — and the composite it feeds — is now a recency-weighted average of every interview round held so far (latest round weighted highest, the first round lowest). A candidate is scored on their trajectory, not one interview: someone who improved round-over-round rises, someone who slipped falls. Each cut (cycle 2, 3, …) automatically folds in all prior rounds. Scores are also calibrated for interviewer severity — within each round every interviewer is re-centered to the round's average, so drawing a lenient or harsh interviewer no longer moves a candidate up or down; the raw average is shown in grey next to the calibrated one.

Reflections now show an advisory "AI-likeness" read at review time. When you open a submission, each reflection carries a small colour-coded score (0–100) with a one-line reason from an LLM judge that runs at submit. It's a flag to weigh, not a verdict — polish reads as suspicious but so does a strong writer — so read it yourself before acting. Pairs with the type-only change: typing evidence proves it was typed; this estimates whether the words are the learner's own.

"Open as" no longer signs you out of your own account. Previously, impersonating a user replaced your shared session, so you appeared logged out and needed a fresh invite to get back in. Now impersonation runs in a separate overlay — your own admin session is never touched. While viewing as someone, a red bar at the top of every page shows whose surface you're on with a one-click Exit to admin. (Note: the overlay applies browser-wide, so your other tabs also show that person until you Exit.)

"Open as" now works for interviewers and admins, and is silent. The Users table's Open as action (to walk a user's surface) now appears for interviewers and other admins, not just thinkschoolers and mentors. And it no longer fires an invite — impersonating a user mints the sign-in link quietly, with no Teams post and no email to that person (the real resend invite action still notifies as before).

The interview screen now defaults to the next un-scored cycle. When you open /interview/, the Cycle field pre-fills to one past the highest cycle already scored for that cohort (it falls back to the calendar default only when no round has been scored yet). So a second-round interview lands on cycle 2 automatically instead of re-opening cycle 1 — no more accidentally overwriting a prior round's score. You can still override the cycle by hand.

Internal certifications are never cut. Each track has a Certification only (never cut) toggle (Admin → Tracks). When on, cohorts on that track are always certify (clear-the-bar), never the drop-the-bottom cut: they don't appear in the Cut tab, the cut ranking and confirm-cut refuse them, and a cut cohort created on the track is coerced to certify. CS Lead and Delivery Principal are set, and any cut-mode cohort on them was flipped to certify.

You choose which tracks show on the public landing. Every track now has a Show on the public landing page toggle (Admin → Tracks, on each track and the create form). Tracks are private by default — only the ones you opt in appear in the homepage's “Tracks we place into” list. The internal certifications (CS Lead, Delivery Principal) are off; .NET, Flutter, Web, and AI Consulting are on. The screen and prep tracks can never be shown.

The public homepage is now a candidate-facing pitch. Logged-out visitors see the whole path — the free Build With AI prep course, how the Aptitude Screen works, and what the six-week mentor-graded Academy is like once they're in — with Apply as the primary call to action and the track list on show. Signed-in staff who land there are pointed to their workspace instead.

Applicants now self-apply, and you control the screening window. A public /apply page lets candidates request an access link themselves (role applicant) — no manual add-user. They land on the always-open Build With AI prep course right away. The Aptitude Screen tab gained a screening window: Open it on a planned aptitude-screen cohort with a target number of qualifiers (fluid, set at open time). Opening sweeps every current applicant in and auto-enrolls anyone who applies while it's open. It auto-closes once the target number have cleared the bar — you can reopen, raise the target, or close early. Closing stops new applicants and new exam starts but never drops anyone and lets in-progress exams finish. Only one window is open at a time. Also: What's new and How it's built are now hidden from logged-out visitors and applicants.

.NET Week 3 is now direct-and-verify. Same topics, same Friday deliverable (a live Angular 21 frontend on Azure Static Web Apps, Lighthouse 95+) — but instead of hand-writing each component, you direct Claude Code to build it against your real Week-1 API, then submit three things: your brief, the agent's output, and a short verification log (which states/edges you exercised, one bug you caught the agent making and made it fix, and what breaks if the API contract changes). It mirrors the real job — directing and verifying AI, not typing it. Already started the hand-coded version this week? That's accepted — submit what you have plus a short verification note. The pass bar is unchanged.

The admin got a top-to-bottom UX overhaul. A new Home dashboard is the landing — stat tiles (active users, pending reviews, applicants screened, active cohorts) plus a cohorts-needing-attention list that jumps you straight into a cohort's cut or screen. The nine tabs are regrouped into Operate · Pipeline · Configure so the applicant lifecycle reads left-to-right. Users is now a dense, scannable table with a “⋯” menu (the destructive Suspend no longer hides among look-alike buttons). Cut / Screen / Graduation auto-load when you pick a cohort — no more “Load first” on an empty-looking card — and the Cut table sorts by any column. Under it all: a real design-token system, AA-contrast text, and full keyboard / screen-reader support (focus-trapped dialogs, arrow-key tabs).

Dropping and suspending are now hard to do by accident. Confirming a cut names every person who'll lose access and makes you type the cohort name; suspend and remove name the person too. Each action leaves a few-second Undo. (The cut is still human-in-the-loop and audit-logged — nothing is auto-selected.)

The Aptitude Screen funnel scales to thousands. Screen results are paginated and filterable — All / Cleared / Unassigned passers / Flagged / In progress with live counts — and you can bulk-assign a batch of passers to an interviewer in one action.

thinkbridge Academy is built for a laptop or desktop. On a small screen you'll now see a clear “switch to a laptop” message instead of a broken layout.

New optional prep track: Build With AI. A self-taught, self-paced course — no mentors, no interviewers — that teaches fresh-graduate applicants how to direct and verify AI-built software. Seven modules mapped 1:1 to the screen's constructs, each gated by a machine-graded mastery checkpoint that pulls fresh questions, demands a high bar, and gives unlimited retries with instant explanations (coming back at a problem until it clicks is the habit we want). It's optional and earns no hiring credit — the Aptitude Screen is still the mandatory gate; this is the on-ramp that readies a motivated self-learner to clear it. The practice pool is disjoint from the live screen bank by design (the course teaches the construct; the screen measures it). Applicants land on it at /learn/; seed the bank and watch completions — a positive-only “prep ✓” beside each candidate's screen result — from the Aptitude Screen tab.

New entrance track: the Engineering Aptitude Screen. A 90-minute, closed-book, machine-graded MCQ screen for fresh-graduate applicants — it tests how well someone can direct and verify software an AI builds (spec decomposition, definition-of-done, calibrated trust, scope/outcome judgment), not hand-coding. There's a new Aptitude Screen tab: seed the item bank, watch the results funnel (started / completed / cleared the bar / integrity-flagged), and assign passers — including flagged-but-passing — to interviewers using the same assignment the interviewer view reads. Create applicants with the new applicant role (Users tab) and add them to an aptitude-screen cohort; they take the screen at /exam/ and never see their score. Each candidate gets a balanced random 42 with options and order shuffled; tab-switches, pastes, and answer pace are recorded (advisory, reviewed with the live interview — not auto-fail), and their email is watermarked across the screen. Honest by design: no webcam, no mic.

Assign interviewers to candidates + see coverage. With several interviewers splitting a cohort, the Cut tab now has an Interview coverage & assignment panel: pair interviewers to specific thinkschoolers (panel-capable — more than one per candidate), and see at a glance who's assigned, who's actually scored this cycle — each interviewer's score and written feedback — the average, and who's still un-interviewed. Each interviewer's /interview/ roster then shows only their assignees — they can't see or score candidates they're not paired with.

Add an interviewer from the Users tab. The interviewer role is now an option in the Add-user dropdown (and the role filter + edit-user) — pick it, send the invite, and they sign in straight to /interview/ to score candidates. No cohort assignment needed; interviewers see every active cohort.

Curriculum now lives inside Tracks. The separate Curriculum tab is gone — open a track on the Tracks tab and click Manage curriculum → to edit that track's categories and pieces, since a curriculum belongs to one track. Competencies stay their own tab — they're a global catalog reused across tracks, not owned by one. A cohort's View curriculum jump still works.

Two anti-AI-relay assessments for the .NET track. The cohort was pasting questions into ChatGPT and pasting back answers — without reading code. Two fixes: (1) the .NET AI reviewer now judges groundedness first — a textbook-correct answer that doesn't cite the learner's actual files, methods, or observed behavior fails (set on the Tracks tab, editable per track); (2) the /interview/ screen now shows a code-reading rubric — drop the candidate into a real legacy repo cold and score navigation, comprehension, change-making, consequence reasoning, and reasoning-under-follow-ups. The relay artists will fall apart on the rubric.

New hiring track: AI Opportunity Consulting. A cut-mode course that trains and screens engineers on the thinkAI motion — assess a workflow, place it on the maturity ladder, find and sequence AI opportunities crawl-walk-run, build a pilot with metrics, and win client buy-in. 6 modules + a simulated-case capstone (28 pieces), human-graded on a 4-week clock, wired to a new AI-consulting competency set. Seed it from the Tracks tooling (seed-ai-consulting) and create a cut cohort to run it.

AI now suggests a behavior, not just a verdict. When you open a submission for review, the AI also proposes one values-anchored behavior observation — drawn from the work and the thinkschooler's own reflections — pre-filled into the Behavior widget (behavior + polarity + a one-line note citing the evidence). Confirm it, adjust it, or clear it, then Record. It's advisory like the pass/fail — you always decide, and it stays empty when nothing's clearly evidenced.

Courses are now timed. When you create or edit a track you set a Duration (minutes — so a course can run minutes, hours, days, or weeks) and a Grading mode: human (mentor reviews, as today) or auto (the AI verdict decides at submit — needed when a course is too short to wait for a human). A learner hits Begin, the clock starts, and if they don't finish before it runs out the course is marked not complete.

Timed courses. Some courses now run on a clock. Your home page shows the duration and a Begin button — once you begin, a live countdown runs and you need to finish every piece before it hits zero. Begin only when you're ready; it can't be paused.

Cohort modes + certification. A cohort is now cut (hiring — drop the bottom) or certify (employee certification — never dropped; the bar means "ready to certify"). New Graduation tab: each member's evidence summary, a mentor recommends an outcome, and you finalize it — which freezes the record and mints a certificate.

Your certification. Once it's finalized, your outcome shows at /graduation/ with a shareable, printable certificate that anyone can verify by link.

New Cut tab in /admin/. An absolute 0–100 composite — throughput, interview score, pass quality, pace, and mentor signal — flags anyone below a threshold you set as a recommendation. Nothing is dropped automatically: you review the roster, select who to drop, add a reason, and confirm. A drop removes access immediately (sessions revoked, blocked everywhere) and is audit-logged.

Interviewers are a role now. Senior engineers sign in to /interview/ and score each candidate per two-week cycle; those scores feed the cut as an equal-weighted input.

Submit your work with a GitHub link. Code pieces now have a dedicated GitHub field — paste the repo, the exact folder (…/tree/main/Day1/piece2), a PR, or a CI run. Push to the thinkbridge-thinkschool org so your mentor can open it. (You can still drop notes or a commit hash in the box below it.)

The review modal now shows a code submission's GitHub link as a clean, clickable link (repo / folder / PR / CI run) instead of a URL buried in prose.

AI-review draft comments are now straightforward and concise — they lead with the point and cut the preamble, so they're quicker to skim and to send.

New Settings tab in /admin/ to maintain integration keys & tokens (Anthropic, GitHub, Teams, Resend) yourself — no more CLI. It lists every key the platform uses (set or not), and lets you set or replace one. Values are write-only and never shown back; every change is audit-logged.

Each key on the Settings tab has a Test button — one click confirms the key actually works (Anthropic via the audit proxy, GitHub + Resend via their APIs, Teams by posting a connectivity card). It reports pass/fail and never exposes the value; unknown/future keys fall back to a safe presence check.

The full six-week .NET curriculum is live. Weeks 2–6 join Week 1 on the Map — SQL Server + EF Core depth, Angular 21, distributed systems, cloud-native + DevOps, and the capstone — so there's a clear path all the way to graduation. No prereq gates; work ahead if you're ready.

AI-assisted review. Each submission now comes with an advisory recommendation (PASS/FAIL), a one-line reason, and a draft comment you can drop into your note with one click. It's a starting point — you always make the call.

thinkbridge Academy now runs multiple tracks in parallel. Each cohort is assigned one track (curriculum), and you only ever see your own track's pieces and plan. Five tracks are live: .NET Backend, CS Lead Certification, Delivery Principal Certification, Flutter + Firebase, and Web + Firebase.

If you belong to more than one track, a track switcher appears in the top bar. Switch it to re-point your Today, Map, and plan to that track. In a single track? You won't see it — nothing changes.

The plan is now per track — it shows the plan for your active track. (The .NET six-week plan now lives there too.)

New Tracks tab: manage the curricula on offer — edit a track's name, summary, and plan, or add a new track. The cohort form's track list reads from here, so adding a track no longer needs a code change.

Users tab scales. Search by name or email, filter by role and status, and page through — the old 500-user ceiling is gone, so a 1,000+ roster loads fine. Adding a member to a cohort is now a search, not a giant dropdown.

Cohorts tab is cohort-first. A compact list shows each cohort's track and member count; open a cohort to view and manage its roster. Filter by track or status. Built to hold 20+ cohorts of ~60 without slowing down.

Map is now thinkschooler-only — it's the cohort curriculum view. Admins author curriculum from the Tracks + Curriculum tabs; mentors work from the Queue. Admin, mentor, and map screens also now use the full page width.

New Account link in the topnav. Set or change your password there anytime — works for thinkschoolers, mentors, and admins, not just the home-page banner.

Fixed: opening What's new (or any signed-in page) while logged in as a superuser no longer signs you out. Preview as thinkschooler / mentor works again.

If the platform is ever down during the day, hand out a day's offline reading PDF with one command (deploy/export-day-pdf.sh <day>) — see the admin checklist's outage section.

Two optional prompts under each exercise: What did you learn this session? and What would break this? You don't have to fill them, but your mentor gives extra credit when you do — a line or two each is plenty.

When a thinkschooler fills the reflection prompts, their answers show in the review modal under their submission, flagged as extra credit. Reward the ones who reflect.

What's new link added to the topnav with a small red dot when there is something you have not read yet.

Two new cohort guides in the topnav: Copilot and Claude Code. Top five things this week per tool. Future weeks add more.

Notes from your mentor panel on your home page. Anything your mentor writes for you, across all pieces, lands here. Unread notes show a red bar on the left until you visit the page.

You can set a password from the yellow banner on your home page. After that, sign in directly at /sign-in/ with your email and password instead of waiting for a fresh magic link.

[email protected] now owns the mentor queue for pune-may-18-2026. The 10 pending submissions from yesterday are cleared; the queue is at zero going into Day 2.

Cohort start time field on /admin/ now takes IST-friendly input (YYYY-MM-DD HH:MM) and shows the UTC conversion inline. One less timezone-flip mistake to make.

Inline green/red delivery badges next to every magic link in /admin/ tell you whether the invite email and Teams ping actually went out. notify_outcome audit row written for every attempt for post-mortem.

Week 1 collapsed to 5 days × 7 pieces (Tue 19 to Sat 23). Day 1 (Mon 18) absorbed as an informal warm-up after the morning's mentor-gate snarl. The cohort end date (Fri Jun 26) stays the same.

Mentor gates removed for every Week 1 piece. All 35 pieces show as available simultaneously in your /map/. Submit in any order; mentors still review every submission but they do not gate the next piece.

A What your mentor has noticed about you panel on your home page shows behavior observations as they accrete.

Verdict emails: when your mentor decides on a submission, you get an email with the verdict and the note. No more refreshing the piece page.

The needs-more-info verdict reframes the submit box as Send follow-up, with the attempt number and a hint to address the mentor's question.

The behavior observation widget inside the review modal is now visible by default. Aim for 1 to 2 observations per review so behaviors accrete toward the graduation page.

Review modal is keyboard-accessible: Escape closes, Tab is trapped inside, focus returns to the Review button.