One operator, real models, verifiable results.
ZenHodl is a sports modeling and fair-value platform built by Coy Evans. The live platform and API both point back to the same core idea: estimate win probability from the documented model, then compare that estimate to the market price — and publish the evidence needed to inspect the result.
The product is AI-assisted on the software side, but the trust layer is intentionally human-readable: public benchmark artifacts, a filtered results ledger, validation, and methodology are visible so you can inspect the work instead of taking marketing claims on faith.
Model win-probability estimates and available venue prices, with quote age and game state where reported. Coverage varies by sport and source.
REST and WebSocket access for games, edges, predictions, venues, and webhooks.
Pre-registered benchmark artifacts and hashes are published for the benchmark windows that use them; the filtered results ledger reports its measured execution-identifier and direct-transaction-hash coverage.
On Starter and above, signed HTTPS webhooks deliver matching edge notifications to your server. Delivery cadence does not guarantee that every source quote is fresh.
A trustable alternative to circular “fair value” tools.
ZenHodl started with a practical question: can a game-state model provide a useful reference alongside venue prices, and can the evaluation be inspected publicly?
The model stack uses game state, historical outcomes, rating context and pregame priors. The API exposes model estimates; the trading bot can apply additional calibration, market blends and risk filters.
That same model layer powers the live platform, the API, the Edge Finder, and the validation pages. The point is consistency: fewer disconnected stories, more inspectable output.
The API labels its exposed edge model_pre_blend, before the bot’s market-blend step. Check the named architecture and features in a dated evaluation; this field alone does not certify input independence.
The Edge Finder and API expose model estimates. The bot applies a separate trading chain; a displayed price gap is not proof that the bot traded it. See the disclosure in the API docs.
Benchmark artifacts, the filtered results ledger, methodology, and validation are public because trust should survive inspection.
Coy Evans
Founder, operator, and system owner
Coy owns the modeling direction, product decisions, validation standards, and trading workflow. The site is intentionally built to show the work rather than hide behind vague claims.
That means the public pages have to do more than sell. They need to explain what the system is, what it is not, and how someone can verify it before paying for anything.
AI-assisted development
Software built with coding copilots, disclosed on purpose
A meaningful part of the codebase has been built with AI coding assistants. That includes implementation help, refactors, debugging, documentation, and infrastructure work.
We call that out because it is better to be explicit than pretend every line was handwritten in isolation. The trust question is not who typed the loop. It is whether the product, results, and validation hold up in public.
If a claim matters, there should be a place to inspect it.
Inspect the public wallet activity here; the results page separately reports measured row-level execution-identifier and direct-transaction-hash coverage.
The filtered bot-attributed ledger includes wins, losses, and nonbinary outcomes, with its evidence coverage stated rather than implied.
Confidence intervals, calibration, p-values, and error analysis are published with the model story.
The public methodology explains what goes into the models and where the limits are.
Try the live platform first
If you want calibrated fair value, best price across venues, and API access without building the stack yourself, start with the platform.
See pricing →Check the record yourself
The linked NBA benchmark publishes the pre-registered artifacts and hashes for that defined evaluation window. Inspect those artifacts alongside the filtered public results ledger instead of trusting a screenshot.
Inspect the benchmark artifacts →