Five products, one closed loop
Most work tools are inboxes. RUQA is a system. Signals become synthesis, synthesis becomes outcomes, outcomes get triangulated, triangulation rolls into capability, capability shapes harnesses, harnesses get tested in Sandbox — and the winners feed back in. Every cycle the team gets sharper.
Six stages, one feedback cycle
Each arrow is a real data flow inside RUQA. Nothing here is decorative — every stage produces input for the next. That's why the product feels different to use after a month than it does on day one.
The data flow, in one paragraph each
GitHub commits, AI session metadata, and Slack messages flow into a single timeline per engineer per day. RUQA's prompt reads that timeline and writes a draft daily standup — the engineer doesn't fill in a form.
Every approved standup distills into one or more outcome records: the PR that shipped, the spec that landed, the bug that got triaged. The outcome is the unit of work everything else hangs off.
An outcome triggers five independent signals — self-report, AI baseline estimate, git timestamps, volume regression, peer median. Disagreement above 30% creates a HARD flag for review.
Confirmed outcomes get scored against the 16 metaskills (prompt design, edge case detection, decision rationale, etc.). The score isn't a survey — it's derived from what the engineer actually shipped.
Capability data tells you which prompts your team is leaning on for which skills. Strong patterns surface as harness candidates; weak patterns get flagged for review. The harness library evolves automatically.
Before a harness is promoted to company-wide, it gets tested in Sandbox: same prompt across 4 LLMs and against alternative prompt structures. The winner ships. Loop closes back to Synthesis.
Three principles you can feel in every screen
Read what you already produce
Your team already creates commits, PRs, AI sessions, and decisions. RUQA refuses to ask you to fill another form. If a piece of information could be inferred, we infer it.
Every score is reproducible
Every number RUQA shows you — capability points, triangulation values, sandbox rubric scores — has a published prompt, rubric, and weighting. You can audit it. You can disagree with it. You can replay it.
Calibration is built into the product
For the first 90 days of a workspace, scores exist but don't affect compensation, exports, or aggregate analytics. The product enforces this — not a policy doc. By the time scores have weight, the team trusts the algorithm.
Things RUQA deliberately doesn't do
We don't ask 'what did you do today?'
We read what you did. The standup is auto-drafted from real signals.
We don't track tools.
We track meta-skills. Cursor / Claude / GPT churn doesn't affect your capability score.
We don't ship a single 'AI score'.
We ship five independent signals that triangulate. No one number to game.
We don't run Free-tier API bills.
Free is BYOK. Paid tiers are quota-based. Heavy users pay overage or attach their own key.
We don't keep the algorithm secret.
The synthesis prompt, the rubric, the triangulation math — all AGPL, all on GitHub, all auditable.
See the loop in motion
The demo workspace is pre-seeded with 90 days of synthesized history, 47 outcomes, 20 harnesses, and live triangulation flags. Click around — the connections become obvious in about three minutes.