OutSmart: mined only by experts AI can't replace
The model panel sees the full case. The public only ever sees its hash.
4 of 5 models got it wrong
Sent to peer review
Test Ground Truth against Frontier AI.
Select a verified benchmark case below and run the 5-model evaluation live to inspect how frontier models succumb to high-stakes edge traps.
Acute LMCA Occlusion with Hyperacute Normal Troponin
58-year-old male with 45m substernal heaviness. Initial hs-cTnI is 8 ng/L (normal <14). ECG: diffuse ST depression leads I, II, aVF, V4-V6 with 1.2mm ST elevation in lead aVR. BP 98/62.
Where Machines Fail. Where Humans Excel.
Frontier LLMs excel at textbook recall, but collapse when cross-domain subtleties, temporal latency, or regulatory statutory nuances interact.
Hover over any axis vertex to inspect the specific failure vector.
The Declassification Lens
Simulate how OutSmart preserves total privacy while enabling rigorous benchmark evaluation. Drag the slider to adjust declassification transparency.
Patient record: [REDACTED_PATIENT_NAME] admitted with 45 min substernal pain. Initial troponin was hs-cTnI 8 ng/L. Lead aVR exhibits 1.2mm ST elevation with diffuse depressions.
Failure-Rate Weighted Epoch Pools.
Tweak the simulated AI failure rate per profession below to watch the weekly emission pools rebalance dynamically in real time.
SIMULATED WEEKLY $OUTS REWARD POOL PARTITION
From your hardest case to a verified win.
Submit a real case
A scenario from your work, the question, the right answer, a short explanation and a grading rubric. Text, images and attachments all work.
The AI panel answers
Your case runs against the latest models. Every answer is graded against your rubric by an AI judge, with human spot checks on a sample.
Peers check it
When most models fail, two or three experts in your field confirm the answer is right, the case is fair and nothing is ambiguous.
You earn and climb
Accepted cases earn points and rewards, and push your profession up the weekly leaderboard.
Chest pain, normal troponin
5 models tested
Valid, fair, unambiguous
Reviewers vote valid, invalid or ambiguous. Ambiguous cases go back to the author.
AI Slayer, Cardiology
Reward paid. Your case joins the private dataset and counts toward your profession's weekly rank.
Human Edge Index
Which professions does AI still struggle with most? Every week, professions are ranked by how often their verified cases beat the model panel.
Share what you beat, not what you wrote. Share cards show the result and your field, never the case itself.
Top 1% Human Edge, Cardiology
My case beat 4 of 5 AI models (Click to customize)
AI failure rate by profession
Live arena metrics$OUTS. Mined only by experts AI can't replace.
Weekly epochs
Each week's emission is split into profession pools.
Pools weighted by AI failure
If AI fails 80% of cardiology cases and 40% of accounting cases, the cardiology pool is bigger. The hardest fields to replace earn the most.
Split inside the pool
Rewards go to accepted cases, multiplied by models beaten and the expert's verification tier.
Edge decay
When a newer model cracks an old case, it drops off the leaderboard. Rewards already earned stay yours.
Revenue-backed demand
Dataset buyers pay in stablecoins. Part funds $OUTS buyback and burn, part goes to the experts whose cases were licensed. Sponsors buy $OUTS to fund sprint bounties.
Fixed total supply
500,000,000Fair launch on Robinhood Chain (ID 4663), no exempt wallets, public team vesting.
About 5 years, halving every 12 months
Gradual emission for peer review
Multisig, grants per profession
12-month cliff, 36-month linear vesting
LP locked for at least 12 months
The first 1,000 verified experts
For AI labs and vertical AI teams
License expert-level evaluation sets that have never been on the open web. Every case is peer verified, tested against current models and watermarked per buyer.
Dataset licenses by profession, by subscription or batch, with optional exclusivity.
Custom eval sprints fund a competition in exactly the domain you need.