Mistake #4: Using an Old Roster in a New Forecast
A team name is not a complete description of the lineup playing today. If your model was built on matches with one group of players and a different group starts, check which assumptions still apply. That does not mean every substitute weakens a team. A familiar stand-in can fit a role, and the usual lineup can have other problems. The mistake is treating a change as irrelevant or attaching an invented fixed penalty to it.
This guide provides a verification process and hypothetical calculations. It has no dataset establishing a 12–22% loss of win probability, a 14.2% travel effect or a measured language delay. None of those numbers should be used as a universal adjustment. A confirmed substitution is evidence of who may play; it is not, by itself, a calibrated estimate of the effect on winning.
Separate fact and estimate: Record the source confirming the lineup, then explain how it changes your model. If the effect cannot be estimated reliably, mark the forecast uncertain or pass. Do not automatically back the opponent or add several unsupported penalties together.
1. Confirm the players for the relevant match
Start with the team's public announcement, the tournament's official communication and the applicable roster rules. A season registration, an approved substitute and the actual starting lineup are different facts. Check whether a statement concerns this event, this match or only a future roster move. Keep the original link and publication time, including its timezone. A screenshot without a source or date is not a complete confirmation.
Rules are event-specific. The ESL July 2026 update, for example, includes changes to lineup and registration provisions. It does not establish a universal deadline for every tournament. Recheck the current event's rules instead of assuming a lineup must be final 45 minutes before play. A rule permitting a substitute does not prove that the substitute will start.
Databases such as HLTV, Liquipedia or Dotabuff can help cross-check names and history. They should not be described as the tournament's official signed match sheet unless that exact document is provided. A player post may confirm the author's situation without confirming all five starters. If sources disagree, preserve the disagreement and wait for clarification; repeating the same unsourced report on several sites does not create independent evidence.
2. Ask what changes beyond one name
For CS2, useful questions include who calls, who uses the AWP, which positions change and whether the substitute has played those roles with this lineup. For Dota 2 or League of Legends, use the roles and game-specific responsibilities of that title. Do not copy a CS2 anchor penalty into a different game's model. These questions identify what to investigate; they are not a ranking of universal damage from each substitution type.
An individual rating compares a player within the matches and roles that generated it. It does not show that replacing a lower-rated player improves the entire team. Different opposition, maps, role allocation and sample sizes can make the comparison misleading. The player-statistics guide discusses these metrics; a rating alone cannot convert roster news into a team win probability.
Communication should be checked from actual information about the lineup, not inferred from nationality. A multilingual roster may already have a shared working language, while same-language players may still be unfamiliar with one another. This article has no experiment measuring a fixed number of milliseconds added by English calls. Do not assign such a cost to a team or assume that an international substitute forces a specific language change.
3. Use scenarios instead of invented deductions
Suppose a hypothetical model originally gave team A a 0.72 win probability at odds 1.50. Its EV per unit was +8%. After new information, consider three alternative estimates below. The values are chosen solely to demonstrate sensitivity. They are not validated deductions for an IGL, support player, travel schedule or language, and there is no claim that the effect must be negative.
On a narrow screen, scroll the table sideways to see every column.
| Hypothetical estimate | Probability | Odds | EV |
|---|---|---|---|
| Original model | 72% | 1.50 | +8% |
| Less favorable scenario | 63% | 1.50 | -5.5% |
| More favorable scenario | 76% | 1.50 | +14% |
A fall from 72% to 63% is nine percentage points, or a 12.5% relative reduction. Those are different units. Neither number becomes justified because the arithmetic is correct. Additive deductions for role, communication and travel can also count the same disruption several times. Without an estimated joint model, summing them creates false precision rather than evidence.
Hypothetical mutually exclusive scenarios:
P(ordinary preparation) = 0.60; P(A wins | ordinary) = 0.70
P(limited preparation) = 0.40; P(A wins | limited) = 0.55
P(A wins) = 0.60 × 0.70 + 0.40 × 0.55 = 0.64
EV at odds 1.50 = 0.64 × 1.50 - 1 = -4%
Break-even at 1.50 = 1 / 1.50 = 66.666667...%The mixture is valid if those scenarios exhaust the possibilities and the weights and conditional probabilities are defensible. Here they are explicitly assumed. If neither the preparation state nor its effect is known, the calculation shows a question to resolve, not a fact to publish as the team's true chance. The odds converter converts prices; it does not infer a roster effect from a player's name.
4. Treat travel and price movements as separate evidence
A public arrival time or schedule can identify a short interval before play. It cannot tell you how much a player slept, diagnose fatigue or determine a fixed reduction in reaction speed. Keep a confirmed itinerary separate from speculation about health or performance. Do not present a real team as having a visa or language crisis in a constructed example. The favorite-bias guide concerns evaluating evidence rather than the badge on the jersey.
A price move is not proof of insider knowledge. It can reflect roster news, another public fact, margin changes, demand or a model update. A difference of fifteen cents also has different probability meaning at different decimal prices. Compare the same market, timestamp and settlement rules. Neither a shorter price at one operator nor a later closing move certifies the original forecast.
This article has no measured reaction-time comparison between sportsbooks and provides no guaranteed 3–5 or 20–60 minute window. A displayed quote may change or a request may be rejected. There is no reason to increase a stake simply because a substitution was announced. Use line comparison to inspect equivalent offers, while keeping requested and accepted prices separate.
If news appears after a bet has already been accepted, the new estimate does not retrospectively change its price. Check the ticket status and applicable rules, retain the news timestamp and record the forecast revision separately. Wanting to cancel is not evidence that cancellation is available. An automatic opposing bet or a larger position is a new decision requiring its own price, probability and assessment of total exposure.
Keep cases where you did not bet when reviewing the model. Otherwise the ledger only shows decisions that passed your filter and hides selection mistakes. Compare the forecast with the actual starting lineup, information available before play and the result. Do not revise the original probability after seeing the score. If no original record exists, say that the initial forecast can no longer be checked.
5. A practical verification record
- Save the exact event, match, expected starters and public primary source, with publication and observation times.
- Distinguish registered, eligible and confirmed starting players. Record unresolved names rather than assuming the usual five.
- Explain which model inputs depend on the old lineup and which roles might change. Avoid automatic penalties or an automatic opposing bet.
- Keep the original forecast, revised scenarios and contemporaneous quotes. Recheck after a material update, not only at an arbitrary two-hour cutoff.
- Review all comparable forecasts after results, including those you passed. A single upset neither proves a substitution penalty nor validates a betting rule.
A pending roster field should remain pending. Save later confirmation as a new record so the original decision can be reconstructed.
