Sports Analytics Engine

Serious betting research means finding every historical game that matches a situation: streaks, season-to-date averages, travel and elevation context, and conditions on the opponent as well as the team. Filters that reference other games, "after three straight games below their season average", cannot be expressed in a spreadsheet, and are awkward even as SQL.

At the same time, the analyst's entire dataset and methodology lived in Excel: large workbooks with one 64-column sheet per team covering betting lines, box scores, rest days, and travel, plus summary formulas and betting-return calculations. The system had to preserve that established workflow, including its exact numbers, while making multi-condition situational queries fast and repeatable across an entire league's game history.

Rather than forcing a new data-entry workflow, the platform imports the existing workbooks as-is. A server-side pipeline parses the index sheet, conferences, teams, venues with coordinates and elevation, and every per-team sheet, then loads a normalized PostgreSQL schema through chunked, deduplicated upserts.

We modeled games from both perspectives: each fixture is stored once as a match and twice as team-perspective game rows, with a database view joining each game to both its own and its opponent's box score as embedded JSON, so the query engine reads complete game context in a single fetch.

The situational logic lives in a dedicated query engine. Filters that reference other games, streaks, season averages, correlated conditions, cannot be simple WHERE clauses, so the engine fetches each team's full ordered game history once and evaluates layered characteristic filters in application code, keeping the filter semantics explicit and testable.

An analyst combines date filters, scope filters, teams, conferences, game type, home/away/neutral, overtime, and characteristic conditions across roughly 65 queryable measures spanning betting fields, scoring, rest, ranking, travel distance, elevation change, and the full team and opponent box score. Each characteristic supports current-game thresholds, previous-game conditions with consecutive and in-a-row semantics, season-average conditions, and correlation, restricting prior-game calculations to games that satisfied other filters in the query. Team-side and opponent-side criteria are defined independently, and the engine returns team, opponent, and intersection result sets as switchable tabs.

Every result set feeds a returns panel that simulates a flat-unit betting strategy across the matched games in four modes, moneyline, spread, over, and under, reporting total risk, net return, plus/minus units, and ROI at a configurable stake, pricing spread and totals bets at standard -110 juice. The calculations replicate the formulas from the original Excel workbook, so results are directly comparable with the analyst's prior process.

Results render in a spreadsheet-style grid built for large outputs: row virtualization keeps thousands of rows responsive, grouped column headers mirror the workbook layout, and pinned summary and averages rows reproduce the Excel formula rows, including W-L-push records for spread and over/under outcomes. Around the core sit a League Stats view aggregated in a PostgreSQL view, per-team game logs with server-side pagination, browsable team and venue directories, and an Import screen that streams the entire workbook load over Server-Sent Events with live log lines and progress.