Inside Flare.

The front door of the app: every open task in your scope, visible to everyone. Nothing is allocated in private — if work exists, it's here. Each card carries the skills it needs and any prerequisite; the recommendation engine reads your skill levels and history and surfaces the tasks where you're most likely to win.

▸ Try it — toggle ★ For you to see your skill-matched tasks (your skills: Prompting Lv 2, Python Lv 2, Data viz Lv 1), or filter by scope.

⚙ Under the hood: a recommendation service scores every open task against your skill levels, XP history and past wins — match % is explained, never a black box.

A task is only postable if it's fully specified — objective, template, steps, acceptance criteria, required skills, prerequisite. Complete enough that no meeting is ever needed: read it at 4 a.m., deliver by morning. Questions and answers happen asynchronously on the task itself, and the poster's answers are visible to every competitor — one answer, everyone informed.

▸ Try it — open the steps, click I'm on it, then ask a question in Q&A and answer it wearing the poster's hat.

Where demand enters the market — and where the AI earns its keep. Before anything goes live, the brief is reviewed by a validation service: is it complete? is it clear? which skills does it actually need? what's a fair bounty, given effort and skill scarcity? Posting spends from your Lumens balance — the same Lumens you win from accepted work; no balance, no posting. That's the anti-flood economics: nobody can AI-split a project into a thousand free micro-tasks.

▸ Try it — write a title, pick skills, run the AI review, apply its suggested bounty, publish. Then publish again and watch your balance run out.

⚙ Under the hood: POST /briefs/validate — an LLM call scores completeness and clarity against the template library and 1,200+ past postings, proposes missing criteria, infers required skills from the brief text, and prices the bounty from estimated effort × skill scarcity (how few players hold the skill at level).

The poster's side of the loop — and the moment value changes hands. Accept the best submission and three things happen on the button-press: the escrowed bounty transfers, the winner gains XP in every skill the task was tagged with, and the leaderboards re-rank. Spam gets filtered reputationally.

▸ Try it — Accept J. Fortin and watch the Leaderboard (Stop 5) re-rank and his Synthesis skill level up on the Player Card (Stop 6). Or flag the bulk submission.

The scoreboard that makes the game real — and the firm's talent radar. Two axes: time period (week, month) and skill — "best Prompting player this month" is a real, staffable answer. Rising names here get noticed long before any staffing request exists.

▸ Try it — switch periods, then filter by skill. If you accepted at Stop 4, J. Fortin has already moved.

Every practitioner maintains a living CV — but the skills on it behave like a game character sheet. Each accepted task pays XP into the skills it was tagged with (10 × bounty per skill); XP accumulates into levels; levels unlock harder, better-paid tasks and feed the per-skill leaderboards. A self-declared skill starts unvalidated at Lv 0 — only wins level it up. The profile is a track record, not a self-assessment.

▸ Try it — add a declared skill (starts Lv 0, unvalidated). If you accepted at Stop 4, Synthesis just hit Lv 3.

One balance, one currency. The Lumens you win from accepted work are the same ones you post with (Stop 3) — earning gives you real power in the market rather than a score you cannot use. Spending never costs standing: leaderboards and review outcomes read acceptance events, not what you hold. Sponsors can top up a balance, but nothing converts to cash — that's the wage-theft firewall (Risk R-1).

▸ Try it — this balance is the same one the Composer spends.

The prerequisite system — qualification management, not just courses. A prerequisite is created from a project's primary material (SOW, repo, transcripts): an AI agent generates a ~3-hour course and acts as coach. Task creators attach prerequisites in the Composer and see the qualified population before posting; practitioners pass the gate once and compete on everything behind it. Two views below: the learner taking a gate, and the manager creating and tracking them.

▸ Try it — complete the three learner modules, then switch to Manage prerequisites and generate a new one with AI.

⚙ Under the hood: POST /prerequisites/generate — an agent ingests the selected source material, builds the course and quiz, then serves as the per-project coach; qualification counts sync live to the Composer and the Board's cards.

The operator's console — because the market is moderated, not a free-for-all. Curators approve postings before they go live (the AI review from Stop 3 pre-screens the queue), market-health metrics expose stalls and price failures, and the talent radar surfaces rising players per skill. Five people attempting one problem down three paths is intelligence, not waste.

▸ Try it — approve or return the pending postings; dismiss the duplicate-effort alert.

⚙ Under the hood: clustering agents group similar submissions to detect duplicate effort and harvest distinct solution paths; the same signals feed the talent radar and the price-health metrics.