100 agents. Every game. Daily reports with scenario distributions, sensitivity rankings, and market comparison.
Mesh Madness is 67 games in three weeks. Every game is a new matchup with no shared history — different styles, different conferences, different contexts. Bettors, analysts, and fans face the same problem: too many variables, not enough structured ways to think about how a game actually plays out.
Public models give you a win probability and a spread. They don't tell you why — which mechanisms produce a Houston blowout versus a competitive game, what has to go right for the 12-seed to stay live, or how fragile the favorite's edge really is.
We developed a Mesh Madness plugin for the Intellidimension Mesh Platform — a domain-specific configuration that focuses the mesh runtime on tournament basketball analysis.
The plugin defines an ontology covering the dimensions that drive college basketball outcomes: pace and transition control, three-point variance, paint offense versus rim protection, key player influence, foul environment, rebounding margins, late-clock execution, coaching adjustment edges, neutral-site composure, and rotation availability. Each dimension is modeled with discrete states and calibrated priors specific to the matchup.
The analysis template assembles a network of 100 specialized agents for each game — offensive and defensive scheme analysts, tempo and transition specialists, player-specific evaluators, matchup historians, coaching tendency analysts, injury and availability monitors, and environmental context agents — connected through a small-world network that balances deep positional expertise with cross-cutting perspectives.
The plugin produces three outputs for every game: a pregame analysis document synthesizing the mesh deliberation, a Many-Worlds scenario simulation with full margin distributions, and a market comparison showing where the mesh agrees and disagrees with betting lines.
100 specialized agents spanning the full analytical surface of a tournament game. Quantitative, tactical, operational, behavioral, and market viewpoints all compete for influence through the deliberation process.
Win probability, shot quality, possession value, single-game volatility, upset baselines, first-half game-state, free throw dependency, lineup synergy, shot-mix dependency, schedule normalization
Ball-screen containment, switching defense, zone offense, transition defense, press-break, after-timeout design, late-clock offense, endgame decision trees
Lead-guard decisions, rim finishing, catch-and-shoot roles, rebounding technique, foul-drawing craft, foul discipline, player tendencies, neutral-site shooting adaptation
Advance scouting, scout retention, practice fidelity, bench management, foul-trouble contingency, substitution timing, roster cohesion, player development arc
Biomechanics & fatigue, sports medicine, load management, hydration & cramp risk, competition fueling, rest differential
Pressure response, attentional routines, stress spillover, retention under game stress, bench role psychology, prospect visibility effects, program identity
Sportsbook risk management, market signal integrity, market-vs-model divergence, availability investigation, source verification, probability communication
Tournament compliance, team travel, arena operations, host-city atmosphere, crowd composition, weather disruption, site acclimation, postgame procedures
Bracketology, seed-history analysis, tournament history, game-book archive, conference-to-neutral translation, research librarian, procedural stoppage specialist
Broadcast narrative filtering, scouting visualization, game note interpretation, analytics-to-court translation, evidence literacy, pregame signal interpretation
Clusters of related expertise deliberate intensively among themselves before conclusions travel across weak-tie bridges to adjacent clusters — ensuring no insight stays trapped inside a single discipline.
The mesh ingests team profiles, season statistics, recent performance data, injury reports, and matchup-specific context. The 100-agent network deliberates through multiple generations — agents post arguments, neighbors pick them up or challenge them, and the Chair synthesizes emerging consensus and unresolved tensions. The output is a structured analytical document covering the key dimensions, critical matchup factors, and the specific mechanisms that drive each team's path to winning.
The mesh identifies the distinct game scripts — the named worlds that represent qualitatively different ways the game can unfold. A Georgia–Saint Louis matchup might decompose into six worlds: Georgia transition control, Saint Louis spacing and shot quality, Georgia close escape, Saint Louis steals the close game, and stress worlds where foul trouble or availability shifts the structural balance.
Each world carries a margin-calibrated tilt anchored to the betting spread, so "comfortable control" means something concrete — not a vague positive sentiment, but a specific expected margin grounded in the matchup context. The simulation runs 2,000 Monte Carlo draws per game, mapping dimension states to worlds and adjusting margins based on how many dimensions resolve favorably or unfavorably beyond each world's baseline conditions.
A perturbation layer then re-runs the simulation across 1,000 Dirichlet-sampled prior sets — 2 million total draws per game — producing the full predictive distribution that incorporates both game-level variance and prior uncertainty. This is the distribution a bettor actually cares about: not "what happens if our priors are exactly right" but "what happens across everything we might reasonably believe."
Every histogram is annotated with the current betting spread. The report surfaces where the mesh's center of mass sits relative to the market, which worlds create the disagreement, and which dimensions are most sensitive — so you know exactly what to watch in the first five minutes to update your view.
Each day's PDF covers every game on that day's slate. For each game:
Daily reports are published each morning covering that day's games, with post-mortems added after results come in. The full archive — pregame analyses, Many-Worlds distributions, sensitivity rankings, and post-mortems — builds over the tournament into a complete calibration record.