TTLossbeater
← Back to Documentation

How Our Signal System Works

Deep dive into the technical details of our multi-signal consensus and edge detection system.

What TT Lossbeater Is (and Isn't)

TT Lossbeater is a sports analytics platform for professional table tennis. It covers the six most active pro circuits — Setka Cup, TT Elite Series, Czech Liga Pro, TT Cup, Setka Cup Women and Challenger Series TT — ingesting results and odds daily, maintaining per-league Glicko-2 player ratings, recent-form windows and head-to-head histories for thousands of players, and publishing a signal-based read on every scheduled match.

It is not a coaching or training tool, not a bookmaker, and not a tipster channel promising guaranteed wins. You cannot place a bet here. Every number on the site exists to help you understand a match, and every read we publish is recorded — wins and losses alike — for informational and educational purposes only.

Architecture Overview

Our system runs independent signals on every match, built on 470K+ historical table tennis matches. Each signal votes independently — no single model to fool. League Portrait tracks which signals are performing right now per league, and every match gets a Confirm or Contradict read with its outcome recorded publicly.

Signal Consensus

Independent signals vote on every match — agreement amplifies, disagreement dampens

League Portrait

Intra-day tracking of which signals are working NOW per league

Edge Detector

A Confirm or Contradict read per match, shown next to the live market price

Data Collection & Signals

Independent Signals

Player Statistics Signals

  • Form L5: win rate differential over last 5 matches (threshold ≥15pp)
  • Form L10: win rate differential over last 10 matches (threshold ≥10pp)
  • Streak: active winning streak of 3+ matches
  • Clean historical data — no reprocessing or future data leakage

Match Context Signals

  • Glicko-2 rating differential between players
  • Market direction (bookmaker favorite signal)
  • Odds Role (structural favorite/underdog patterns)
  • Last Activity recency signal

Pattern Signals

  • H2H: 60%+ win rate across 3+ head-to-head matches
  • First Set: historical 1st set win patterns
  • Confirm / Contradict read classification per match
  • Signal vs market comparison

Analysis Markets

Match Winner (1x2)
Signal consensus identifies the most likely match winner and produces a Confirm read (agreeing with the market favorite) or a Contradict read (backing the underdog), shown next to the live price.
First Set Prediction
Predicts who will win the first set. Uses H2H first-set history with low-spread shrinkage for close matchups. First Set is one of the dedicated signals.
Over/Under Total Points
Predicts whether total points will go over or under the bookmaker's line. Uses a statistical mixture model combining H2H scoring history, player tendency, scoreline distribution, and market signals to calculate calibrated probabilities.

Pick Quality Tiers

Not every match gets a confident read. The system classifies each pick by how its signals line up with the market:

Confirm Read

5+ signals agree with the market favorite — our highest-confidence Confirm read. See its real record on the Performance page.

Confirm Read (standard)

4+ signals agree on the favorite. A standard Confirm read; check its public record before following.

Value Bet

Contradict read: our signal consensus disagrees with the bookmaker's favorite and backs the underdog. High-variance by nature — the public record shows exactly how it does.

Data Integrity

Historical Data Foundation

470K+ matches of clean historical table tennis data. No reprocessing of historical results — signals are computed strictly from data available at match time.

Per-League Validation

Signal behavior is tracked independently per league. A pattern must hold up on live, point-in-time results — not just backtests — before it informs a league's Confirm or Contradict reads. The real per-tier record is public on the Performance page.

League Portrait Updates

League Portrait refreshes intra-day as matches complete. If a signal stops working in a specific league, it stops contributing to that league's signal consensus.

Track Record

~60%

Signal read vs the market (tracked)

~68%

Confirm read track record

~70%

Contradict read track record

These statistics reflect the real, point-in-time record of our Confirm and Contradict reads on live matches — not backtested data. Results vary by league, time period, and match conditions, and the full record (including losses) is public on the Performance page. Reads are produced daily across 6 leagues.

The Six Signals

Instead of one opaque model, six independent signals each cast a vote on the likely winner:

  • Rating — per-league Glicko-2 player ratings, updated after every finished match.
  • Form L5 — win rate across the player's last five matches.
  • Form L10 — the same read over a longer, steadier ten-match window.
  • Streak — the current winning or losing run, a momentum check.
  • H2H — the head-to-head record between these two specific players.
  • Market — the bookmaker's implied favorite, always the starting prior.

From Votes to a Read

A base probability blends the rating gap with weighted recent form (the last five matches count most, then the last ten, then twenty). Signal agreement then scales confidence: when five or six signals point the same way the read is amplified; when they split, it is dampened back toward a coin flip. The market is always the prior — signals move the needle relative to the odds, never in a vacuum.

On top of that, each league has its own calibration that acts as a ceiling: it can demote a read that looks overconfident for that league's conditions, but it can never promote one. Fast, volatile leagues are treated with more suspicion than stable ones.

Why Every Result Is Public

Every read is snapshotted at prediction time and graded only after the match settles — no hindsight edits, no silently deleted losses. The full ledger, including losing days and losing weeks, lives on the Track Record page. If the numbers have a bad month, you will see the bad month.