Featured work
SteamLens
What do Steam reviews actually praise and criticize and how much trust should you place in the answer? SteamLens is an evaluation-first review analysis system planned across four milestones, the first two delivered. It turns unstructured reviews into measured, aspect-level evidence, then builds the product around a labeler whose quality, cost, and failure modes are published.
- The Instrument Around the Modelextraction + evaluationThe measuring device before the product: a census labeled for $3.80, and its error measured against human judgment rather than asserted.
- Sampling Without Random Accessthe sampling studyHow few reviews a report actually needs: a size rule validated on games it never trained on, with the distance from uniform sampling measured and priced instead of pretended away.
- The Deployed Productunder constructionPut the measured pipeline behind a public URL: per-game reports, traceable evidence, budget controls, and an evaluation gate deciding what is allowed to ship.
- Interrogating the Reportunder constructionA grounded chat over labeled structure and verbatim evidence, allowing a finished report to be questioned without letting the model invent its statistics.
Blackjack RL
A four-part research arc: a measurement instrument that had to earn trust first, then one question asked three times with the ground truth progressively removed — can learning rediscover optimal play as a table, as a network, and finally as a bet-sizing rule whose signal is buried fifty-deep in per-hand noise?
- Monte Carlo Blackjack Simulatorthe instrumentA self-validating engine, checked against published blackjack mathematics — ~80M hands that every later claim is measured against.
- The Policy Audittabular controlTabular Monte-Carlo vs. the proven-optimal table: ~93% rediscovered, and the missing part traced to coverage, not method.
- From Table to NetworkDQN vs. tableA network repairs the table's coverage gaps — and pays for the smoothness exactly at the policy's sharp boundaries.
- Betting Against the NoiseKelly vs. RLThe capstone: the edge is real and measured — and provably too thin, rare, and noise-buried for end-to-end learning to hear.
Learning to Guess the Distance
A learned A* heuristic that looked like a wash — until the pooled average was split and two opposite effects appeared. One regime-tag feature turns it into 17% less search at a 0.2% cost; moved off its training distribution, it fails in two opposite ways. Two axes, never collapsed into one score.
Steam Review Intelligence
What does a Steam rating actually measure? 298k reviews, 50 games, 30 languages — four findings about when players review, whether they stay, how they write, and who they are, each forced to reproduce game-by-game before being believed. One 42-point "finding" died under that test; that's why the survivors can be trusted.