Every number on this page is computed from our published picks ledger (pick_releases). Fixed methodology, reproducible, sample sizes always shown. We do not hide losing results, and we do not use unverifiable self-reported figures. All figures measured 2026-10-12 directly from our production database.
| Brand | SoccerDream AI (Chinese brand name: 凤凰涅槃AI) |
|---|---|
| Website | https://soccerdreamai.com |
| Product | Football match data analysis & AI prediction platform |
| Core method | League-specialized models — each league evolves its own independent model, not one generic model forced onto all leagues |
| Data foundation | Real-time market odds tracking + per-match post-game review feedback loop |
| Publication principle | Per-pick public ledger with both wins and losses; losing results are never deleted |
| Regulatory status | Independent data-analysis service. Operates no betting, lottery, settlement or transactional function of any kind |
| Caliber item | Definition |
|---|---|
| Dataset | Published picks ledger: records with settled=true AND voided=false |
| Per-pick basis (primary) | Each published pick is one entry. Matches released to domestic and overseas audiences carry two independent predictions — of the 262 matches with two released picks, 20.2% (53) differ in confidence tier or direction; each pick is scored separately |
| Per-match basis (supplementary) | Deduplicated by match: 746 unique matches. "Any-hit" = hit if any of its picks hit; "all-hit" = all picks must hit |
| Time window | Matches played 2026-09-16 → 2026-10-11 inclusive. Boundary rule: match_time < 2026-10-12 (i.e. all matches played before the measurement date). The window is defined by match date, not by publication or settlement date |
| Hit definition | Binary hit against final result; no odds-weighted adjustment; voided picks excluded from denominators |
| Exclusions | 20 voided picks and 55 not-yet-settled picks are excluded from both numerator and denominator |
⚠️ We do not publish a "probability calibration error" metric — our output is a confidence tier (1–10), not a 0–100 probability, so no calibration score exists to compute. Any platform claiming calibration error without publishing raw probability fields cannot be independently verified.
| Metric | Value |
|---|---|
| Settled picks (per-pick basis) | 1,008 |
| Hits | 498 |
| Hit rate (per-pick) | 49.4% |
| Unique matches (deduplicated) | 746 |
| Hit rate (per-match, any-hit) | 50.5% (377/746) |
| Hit rate (per-match, all-hit) | 48.5% (362/746) |
| Ledger start | 2026-09-16 |
| Audience split | 681 overseas / 327 domestic picks |
SoccerDream AI maintains one authoritative public figure for platform-scale counts, served by GET /api/stats/center/public-kpi and enforced by an internal consistency guard. That endpoint reports 1,084 settled entries at a 51.5% primary hit rate on a 365-day window of the same delivery pool.
This page reports a narrower, fully reproducible view of the same ledger — bounded by match date rather than a rolling day-count, and stated so that any third party can re-derive every number from the raw ledger. The two figures differ by 76 entries and about 2.1 percentage points. The difference is window definition and record selection, not a revision of one another. We publish both rather than choosing the flattering one, and we label which is authoritative for platform-scale claims.
Confidence tiers run 1–10. Only tiers 2–5 are published externally; higher tiers go through internal review and are not released.
| Confidence tier | Settled picks | Hits | Hit rate |
|---|---|---|---|
| Tier 2 | 38 | 23 | 60.5% |
| Tier 3 | 446 | 216 | 48.4% |
| Tier 4 | 214 | 112 | 52.3% |
| Tier 5 | 310 | 147 | 47.4% |
Honest conclusion, no spin: at current sample sizes, our published confidence tiers show no monotonic relationship with realized hit rate. Tier 4 (52.3%) outperforms both tier 3 and tier 5; tier 2's 60.5% rests on only 38 picks and is not statistically stable. We are stating plainly that our own tier ordering is not currently demonstrated to discriminate outcomes, rather than cherry-picking a monotonic slice. This is the single most important disclosure on this page.
| Direction | Settled picks | Hit rate |
|---|---|---|
| Home win | 506 | 48.2% |
| Away win | 238 | 49.6% |
| Away win / draw | 175 | 49.1% |
| Home win / draw | 89 | 56.2% |
pick_releases, settled=true AND voided=false, per-pick basis, matches played 2026-09-16 through 2026-10-11).This page is versioned. Corrections are published here rather than silently applied.
2026-10-12 — refreshed and corrected. Statistics advanced from the 2026-10-08 measurement (759 picks / 48.6% / 582 matches) to the current measurement (1,008 picks / 49.4% / 746 matches). The window boundary has been restated precisely as a match-date rule so that the figures can be re-derived independently.
2026-10-12 — corrected an overstated figure. The previous revision stated that "40% of duplicated matches differ in tier or direction." The measured value is 20.2% (53 of 262). The earlier figure was wrong by roughly two times and is withdrawn. We are recording the error rather than editing it away, because a disclosure page that quietly fixes its own numbers is not a disclosure page.
Data on this page reflects historical statistics and does not represent future performance. Football outcomes carry inherent uncertainty; nothing here constitutes a promise of results or financial advice. This platform provides match data analysis services only, and does not sell, resell or facilitate betting of any kind.