# Foresportia This document is the extended, stable context for understanding Foresportia. It explains the product, its public prediction concepts, publication rules, structured-data surfaces, verification approach, and limitations. It does not contain today's predictions, a complete site map, private implementation details, or an API data dump. Use the four public layers for different purposes: 1. `https://www.foresportia.com/llms.txt` is the compact discovery map. 2. `https://www.foresportia.com/llms-full.txt` is the durable explanatory corpus. 3. `https://www.foresportia.com/ai/predictions-today.json` is the replace-in-place machine-readable selection of predictions currently published. 4. A canonical match page is the detailed public source for one match and may embed `foresportia-match-data`. ## Overview Foresportia is a multilingual football analysis platform. It uses statistical models and AI-assisted methods to estimate probability distributions for football matches, adds contextual and reliability-oriented signals, and preserves public historical observations so published predictions can be compared with final outcomes. The site covers domestic and international football competitions in French, English, and Spanish. Coverage and available analysis vary by competition, match, data quality, and publication stage. The existence of a fixture does not imply that a prediction or every supported market will be published. Foresportia is not a bookmaker, a betting operator, or a source of guaranteed outcomes. Its percentages are analytical estimates. They are intended to support probabilistic interpretation of a match, not to promise a result or eliminate football's inherent uncertainty. The general public workflow is: ```text observe match and competition data → estimate probabilistic scenarios → publish only eligible public analysis → preserve the pre-match observation → compare it with the verified final result ``` ## What Foresportia predicts When public and available for a match, Foresportia analysis may contain the following families. ### 1X2 The 1X2 distribution estimates three mutually exclusive outcomes: - `home`: home win; - `draw`: draw; - `away`: away win. The main pick is the outcome with the highest published probability. It is a ranking within the distribution, not a claim that the selected outcome is certain. ### Both Teams To Score BTTS estimates whether both teams score. A published BTTS value belongs to that market only. It must not be substituted for a 1X2, exact-score, or totals probability. ### Over and Under 2.5 These values describe the estimated total-goal distribution around the 2.5-goal line. They provide a goal-volume reading and do not identify which team will win. ### Double Chance Double Chance groups two 1X2 outcomes: home-or-draw (`1X`), home-or-away (`12`), or draw-or-away (`X2`). Only the combinations explicitly present in the public source should be used. ### Draw No Bet Draw No Bet compares home and away outcomes under the rules of that market. Its values must be interpreted as their own public market, not as replacements for raw 1X2 probabilities. ### Likely scores and exact-score scenarios Likely scores are individually ranked scoreline scenarios drawn from a wider score distribution. Several may be displayed because uncertainty is spread across many possible scores. ### Confidence, stability, and public context Foresportia can supplement outcome and market probabilities with a confidence indicator, a stability reading, and contextual warnings. These help interpret a prediction but are not additional outcome probabilities. No market is guaranteed to be present for every match. Missing content means unavailable or unpublished content. ## Publication states and data availability Publication state and match status describe different dimensions: - publication state describes what Foresportia has publicly released for the page; - match status describes the lifecycle of the sporting event. Consumers should use both and must give precedence to the fact that a played match is historical. The match-page publication-state vocabulary contains six values. Four describe prediction availability or historical verification; `postponed` and `cancelled` are page states derived from the event status and are not levels of prediction availability. ### Prediction availability and verification states ### `calendar_only` The fixture is publicly known, but Foresportia has not published a prediction for it. The page may contain identity, scheduling, competition, and team information. It must not be treated as evidence that a prediction exists. ### `early_prediction` An explicitly early analysis is available. It is further from kick-off and should be described as preliminary. Its presence does not imply that all later markets or context are already available. ### `prediction_available` A public prediction is available for the fixture. Only the fields and markets actually present are part of the publication. ### `played_verified` The match has been played and a final result is available for verification. Any retained prediction is a pre-match historical observation and must never be presented as a current recommendation. ### Event-status page states ### `postponed` The match is postponed. Scheduling may change, and no prediction should be described as currently active merely because an earlier analysis exists. ### `cancelled` The match is cancelled. It has no active prediction lifecycle leading to a normal final result. The separate `match_status` field describes the sporting event and can identify states relevant to interpretation such as scheduled, live, played, postponed, or cancelled. In particular, `played` is a match status, while `played_verified` is a match-page publication state. ### Interpretation rules - A missing field or market means that it was not made available in that public representation. - Never reconstruct, complement, or infer an absent market. - Never copy a signal from one market to another. - A `played` match is historical even if another publication label is temporarily behind during a refresh. - Postponed and cancelled matches do not carry an active prediction. - A past prediction may remain visible for accountability, but its tense and purpose must remain historical. ## Live scores and pre-match predictions For eligible fixtures, selected product views can temporarily overlay a fresh observed score, the current in-play status, and elapsed match time. This display is available on the date-based results view, Top Predictions, and Premium Odds Insights when the fixture identity, match timing, provider coverage, and snapshot freshness allow it. The live score is an observation of the sporting event. Published picks, confidence, stability, likely-score scenarios, odds analyses, and the original prediction probabilities remain the pre-match analysis throughout the live match. Foresportia does not currently recalculate these pre-match fields. Separately, selected supported fixtures may expose a beta live 1X2 probability, updated about every five minutes. The currently published live model starts from the archived Foresportia pre-match prior and the pre-match variables it requires, then uses the current score, elapsed time, and normalized red-card counters. It estimates remaining goals probabilistically and derives home-win, draw, and away-win probabilities; the current score is not added to its goal intensity. A red card lowers the sanctioned team’s remaining scoring intensity and raises its opponent’s before the 1X2 probabilities are recomputed. Displayed shots, shots on target, corners, possession, and live xG are contextual information for readers. They are not inputs to the currently published live model. Red-card counters are the sole displayed live statistic used by the model. In pre-holdout development validation, the red-card correction moved log loss from 0.748 to 0.744 overall and from 0.479 to 0.431 in red-card situations; these sample results are not a guarantee for an individual fixture. A live probability or individual statistic may be absent or partial because of fixture eligibility, provider coverage, or freshness. First half, half-time, and second half form one continuous live state while the live snapshot remains fresh. A half-time clock may remain unchanged without making the match non-live. If the live observation expires or cannot be verified, the interface must stop asserting that it is live. A verified canonical final result takes precedence over an older live observation. The presence of a published prediction does not guarantee live-score coverage, and the absence of a live badge must not be interpreted as proof that a match has not started. Consumers should distinguish the pre-match prediction timestamp from the freshness of any separately displayed live observation. ## How to interpret probabilities A probability represents the model's estimated likelihood of an outcome. It is not itself an observed frequency. Across sufficiently large and comparable samples, well-calibrated probabilities should correspond approximately to observed outcome frequencies. An estimated probability of 0.60 therefore remains an uncertain model estimate for the individual match; calibration is assessed by comparing many such estimates with later observations. Football is a low-scoring, high-variance sport. A single goal, dismissal, lineup change, tactical adjustment, or unusual event can change the result. A well-formed probability can therefore be contradicted by one match without that single observation proving that the model is invalid. ### Probability is not confidence Outcome probability estimates how likely an outcome is. Public methodology presents confidence as a separate, reliability-oriented indicator that complements the probability. A high outcome probability can therefore coexist with lower confidence, but confidence is not another estimate of the outcome's likelihood. ### Probability is not stability Stability describes how readable or concentrated the published 1X2 structure is. A distribution with one clearly dominant 1X2 scenario is more concentrated than a distribution in which home, draw, and away remain close. Stability does not convert the leading outcome into a certainty. ### Probability is not a bookmaker price Foresportia probabilities are analytical model estimates. They should not be described as bookmaker odds, market prices, guaranteed returns, or betting advice. ### Scale conventions The public daily JSON uses decimal probabilities from 0 to 1. Public HTML pages and `foresportia-match-data` may display or encode percentage values from 0 to 100. Consumers must read the convention of the source instead of assuming one scale globally. ### Distribution discipline When the three 1X2 values are present, they form the published home/draw/away distribution. Other market families have their own meanings and aggregation rules. Do not calculate an unpublished BTTS, totals, Double Chance, Draw No Bet, or exact-score value from unrelated displayed fields unless Foresportia explicitly publishes that value. ## Confidence and stability Confidence and stability are separate supporting indicators. ### Confidence Public methodology describes confidence as a complementary, reliability-oriented indicator intended to help distinguish a high probability from one supported by more robust comparable evidence. It is presented separately from raw outcome probability. Confidence is not a second estimate of the match outcome and should not be added to, averaged with, or otherwise treated as part of the outcome probability. Consumers should not infer a more detailed construction than the public methodology and displayed data provide. A consumer should interpret confidence in this order: 1. read the outcome probability; 2. read confidence as supporting evidence about robustness; 3. check stability and public context; 4. use verified history for performance questions. ### Stability Stability describes the readability and concentration of the 1X2 signal. It indicates whether one scenario stands out or whether the match remains open. Public labels can express levels such as risk, correct, stable, or very stable, with localized wording where applicable. Stability applies to the published 1X2 structure. Do not claim that it certifies BTTS, totals, Double Chance, Draw No Bet, or a likely score. It is also not a guarantee: even a concentrated distribution leaves probability mass on other outcomes. ### Common interpretation errors - Treating confidence as the probability that the main pick will win. - Treating stability as a guarantee or a universal quality grade for every market. - Assuming that the same displayed probability has identical robustness in every competition. - Judging calibration from one match or one matchday instead of a comparable sample. - Presenting a contextual warning as a deterministic prediction. ## Match-level analysis For a currently published prediction, the preferred discovery path is: ```text /ai/predictions-today.json → canonical_match_url → canonical match page → foresportia-match-data ``` The canonical match URL must be taken from the published source. Do not synthesize a match URL from team names, dates, translated competition names, or assumed slug rules. A match page can provide a human-readable analysis and an embedded public JSON object identified as `foresportia-match-data`. Depending on page state and data availability, this public object may contain: - schema and source information; - canonical URL and page language; - fixture and competition identity; - home and away teams; - scheduled date and time; - match status and final score when available; - publication state and prediction freshness; - the main 1X2 pick and probabilities; - available public markets; - confidence and stability; - likely-score scenarios; - public team context and warnings. The embedded object is a representation of the page, not a promise that every field is populated. A null value, empty collection, or absent market must remain unavailable in downstream answers. The match page is the preferred detailed source because it carries the canonical identity, language context, publication lifecycle, and complete public presentation for that match. Foresportia does not provide a separate unauthenticated JSON endpoint for arbitrary individual matches. ## Exact-score scenarios An exact-score prediction is a distribution over many possible scorelines. The score ranked first is only the individual scenario with the highest estimated probability. It normally represents a minority of the complete probability mass. A displayed Top 5 is a ranked subset of the wider distribution: - the first score is the highest-ranked individual scoreline; - the remaining scores are nearby alternative scenarios; - their displayed probabilities describe their individual estimated weights; - the Top 5 does not necessarily sum to 100%; - outcomes outside the Top 5 remain possible. Exact score is more specific than 1X2, BTTS, or Over/Under. Greater specificity means uncertainty is spread across more individual outcomes. A likely score must therefore be described as a scenario, not as a promised final score. Use broader markets to place scorelines in context: - 1X2 groups scorelines by home win, draw, or away win; - BTTS groups scorelines by whether both teams score; - Over/Under groups scorelines by total goals; - exact score isolates one precise combination. Historical exact-score evaluation should refer to the score scenarios that were actually published before kick-off and the verified final score. It must not retroactively replace the original scenarios. ## Current predictions and freshness Current public predictions must not be copied into this document. They change throughout the day and belong in: `https://www.foresportia.com/ai/predictions-today.json` This fixed URL is replaced in place. It is a narrow public projection of the same selected predictions used by the current Top Predictions surface, not a queryable API and not a public archive. ### Top-level fields - `schema_version`: version of the public JSON contract. - `source`: identifies Foresportia as the publisher. - `generated_at`: UTC time at which this representation was generated. - `valid_until`: freshness hint for the current representation. - `selected_date`: local football date represented by the selection. - `timezone`: time convention used by timestamps in the JSON. - `probability_scale`: declares the 0-to-1 probability scale. - `status`: indicates whether a current selection is published or no prediction is available. - `disclaimer`: short interpretation warning. - `matches`: current public match representations. ### Match fields Each match can provide: - `kickoff_utc`; - `competition` with its code and public name; - `home_team` and `away_team`; - `canonical_match_url`; - `publication_state`; - `match_status`; - the public `1x2` distribution; - `main_pick`; - `confidence` and `stability`; - available `markets` only; - ranked `likely_scores`; - `public_warnings`. ### Freshness procedure 1. Read `generated_at` and `valid_until`. 2. Confirm `selected_date` is relevant to the user's question. 3. Inspect both `match_status` and `publication_state`. 4. Treat `played` as historical, even during a short publication-state refresh lag. 5. Follow `canonical_match_url` for match detail. 6. If a field or market is missing, do not infer it. `valid_until` is a freshness hint for the representation. It is not a guarantee that the sporting event remains scheduled, that a lineup has not changed, or that an earlier prediction is still applicable. The daily JSON intentionally provides no arbitrary-date parameter and no public archive. Historical questions should use the dedicated verified-performance pages or the separate API according to its access rules. ## Historical performance and verification Foresportia distinguishes current analysis from settled observations. ### Current prediction A prediction is current only while it belongs to the active publication window of an unplayed fixture. It describes estimated pre-match scenarios. ### Played match Once the match is played, its final result is an observation. Any retained prediction becomes historical evidence and must be described in the past tense. ### Verified history Verified history preserves the prediction that was published before kick-off and compares it with the observed result. The historical record should not be recomputed from the final score, and the original probability should not be replaced by hindsight. ### Public performance families Foresportia provides separate public views for: - settled 1X2 predictions and outcomes; - settled multi-market picks such as BTTS, totals, and Double Chance when they were published; - national-team prediction history; - exact-score scenarios and verified final scores where the corresponding history is available. Performance statistics change as new matches settle. This document intentionally contains no current success rate, match count, or competition ranking. Use the linked history pages for current observations. Calibration and performance must be judged over a sufficiently broad and comparable sample. A single correct or incorrect match is not a meaningful estimate of model quality. Differences in competition, season phase, data volume, and football variance matter. ## Prediction methodology Foresportia's public methodology is probabilistic and verification-oriented. The following principles are stable public descriptions, not a disclosure of proprietary implementation details. ### Multiple signals Public explanations describe team strengths, attacking and defensive estimates, form, competition characteristics, and available reliable match context. They also describe a coherent goal distribution from which public outcome and goal-market probabilities are derived. Inputs may differ in completeness across competitions and fixtures. No individual signal should be treated as a deterministic rule. The output is a coherent probability distribution built from multiple pieces of evidence. ### From evidence to probability Public methodology states that the engine estimates expected goals, builds a coherent score-probability grid, and aggregates that distribution into 1X2 and available goal-related markets. Public presentation then keeps probability separate from confidence, stability, and contextual warnings. ### Calibration Calibration asks whether announced probabilities behave like observed frequencies over many comparable matches. For example, predictions grouped around a given probability should, over a sufficiently large sample, succeed at roughly the corresponding frequency. A model can rank outcomes correctly while still being poorly calibrated by consistently overestimating or underestimating probabilities. Foresportia therefore treats calibration and observable historical performance as important parts of reliability. ### Validation Published predictions are preserved and later compared with final outcomes. Public history makes it possible to inspect results rather than rely only on claims. Relevant evaluation can include probability calibration and scoring approaches that penalize confident errors, but no single metric removes the need to inspect sample size and context. ### Coverage and selectivity Publishing more matches increases coverage but can include weaker or less well-supported contexts. Publishing fewer matches can increase selectivity but reduces the available sample. Foresportia does not claim that one universal selection rule is optimal for every competition and period. ### Monitoring and drift Football distributions change. Team strength, playing styles, lineups, calendars, season phase, promoted or relegated teams, and competition behavior can shift over time. This is commonly described as drift or seasonality. Monitoring therefore matters. Historical behavior should not be assumed to remain identical indefinitely, and differences between competitions should not be dismissed as noise. Current performance and calibration should be checked against recent verified observations. ### Human-readable interpretation Public pages translate the probability structure into readable concepts such as balance of power, stability, confidence, likely scores, and contextual warnings. These descriptions are interpretation aids. The underlying uncertainty remains present. ## Data quality and publication limits Foresportia uses selective publication. A known fixture can remain `calendar_only`, and an individual market can be absent even when a 1X2 prediction is published. Public reasons for caution include incomplete or unreliable information, insufficient comparable evidence, unusual competition context, schedule changes, postponed or cancelled matches, and situations where the probability structure remains too open. The correct downstream behavior is conservative: - preserve absence as absence; - avoid fabricating values; - avoid claiming uniform coverage; - distinguish preliminary and active analysis; - recheck freshness near kick-off; - distinguish a page's publication state from the sporting match status; - use historical pages only for settled questions. This document deliberately omits private thresholds, internal diagnostic fields, supplier details, unpublished eligibility logic, and implementation-specific scoring rules. Public consumers need the meaning of the released data, not the private mechanics used to decide publication. ## Languages and canonical URLs Foresportia publishes in French (`fr`), English (`en`), and Spanish (`es`). This document uses English as a compact pivot language and does not repeat the full corpus three times. Use these rules for localized navigation: - start from the resource matching the user's language when one is listed; - use canonical and `hreflang` links exposed by public pages; - do not translate or guess URL slugs; - do not assume equivalent pages have identical filenames across languages; - use `canonical_match_url` for the detailed match page, then follow that page's language alternatives; - preserve the language and terminology of quoted public labels. Localized names such as “pronostic”, “prediction”, and “pronóstico” can refer to equivalent product concepts while using different canonical paths. ## Structured public data The unauthenticated public structured-data surface is intentionally narrow. ### Daily discovery JSON `https://www.foresportia.com/ai/predictions-today.json` is the best machine-readable discovery source for what is currently selected and public. It exposes only the current selection and links to canonical pages. ### Embedded match data `foresportia-match-data` is embedded in canonical match pages. It provides detailed public context for that page without creating an arbitrary unauthenticated match-query service. ### Public history pages Historical pages provide human-readable, current verification views. Their statistics should be read from the pages rather than copied permanently into this corpus. ### Surfaces that are not public data APIs Internal storage paths, raw snapshots, supplier files, backend artifacts, and private diagnostics are not public resources. They must not be discovered, linked, or treated as supported interfaces. ## Foresportia API Foresportia also provides a separate API product for programmatic access to structured football data. Depending on the documented plan and entitlement, it can provide match data, predictions, and verified historical results. The product API differs from the public daily JSON: | Public daily JSON | Foresportia API | | --- | --- | | Fixed public URL | Documented API endpoints | | No authentication | Authentication required | | Current selected public predictions | Entitled structured match and prediction access | | No arbitrary-date query | Date and competition operations according to documentation | | No public archive | Verified historical depth according to plan | | Compact discovery representation | Product schemas, quotas, errors, and usage rules | API authentication, endpoints, quotas, plans, historical depth, identifiers, and schemas are documented separately. Do not duplicate that reference here, expose credentials, or assume that the public JSON contract and API schemas are interchangeable. The API documentation map is: `https://www.foresportia.com/api/docs/llms.txt` Public API overview pages are available in French, English, and Spanish. ## Limitations - Predictions are estimates, not guarantees. - Football outcomes contain irreducible variance. - Data can be incomplete, delayed, corrected, postponed, or cancelled. - Some late-breaking, incomplete, or unavailable match context may not be fully represented. - Competition behavior and model calibration can change over time. - Low-volume contexts support weaker conclusions than broad comparable samples. - A leading 1X2 outcome can lose; a top exact score can fail while the broader distribution remains reasonable. - Confidence and stability support interpretation but do not remove uncertainty. - Historical success in one competition, period, or market must not be transferred automatically to another. - A missing prediction or market is not permission to reconstruct it. - Current and historical resources answer different questions and must not be mixed. The responsible interpretation is to state uncertainty, cite the relevant public source, respect its freshness and lifecycle, and avoid stronger claims than the published data supports. ## Key resources - [Compact Foresportia map](https://www.foresportia.com/llms.txt): Short discovery index for current predictions, methodology, histories, languages, match discovery, and API access. - [Current public predictions JSON](https://www.foresportia.com/ai/predictions-today.json): Replace-in-place structured discovery source for predictions currently published. - [Prediction methodology — English](https://www.foresportia.com/en/ia-prediction-football.html): Detailed public explanation of pre-match and beta live probabilities, inputs, contextual statistics, calibration, and limitations. - [Prediction methodology — French](https://www.foresportia.com/ia-prediction-football.html): French methodology and interpretation guide for pre-match and beta live probabilities. - [Prediction methodology — Spanish](https://www.foresportia.com/es/ia-prediccion-futbol.html): Spanish methodology and interpretation guide for pre-match and beta live probabilities. - [Exact-score analysis — English](https://www.foresportia.com/en/exact-score-football-ai.html): Public explanation and current presentation of probabilistic score scenarios. - [Verified 1X2 history — English](https://www.foresportia.com/en/past_results.html): Current settled 1X2 observations and final outcomes. - [Multi-market history — English](https://www.foresportia.com/en/past_picks.html): Current settled observations for available published markets. - [About Foresportia — English](https://www.foresportia.com/en/about.html): Project identity, purpose, transparency approach, and limitations. - [Foresportia API — English](https://www.foresportia.com/en/api.html): API product overview and access options. - [Foresportia API documentation map](https://www.foresportia.com/api/docs/llms.txt): Technical entry point for API schemas, endpoints, authentication, and usage rules. For a complete FR/EN/ES navigation matrix, use `https://www.foresportia.com/llms.txt` rather than expanding this explanatory corpus into a sitemap.