Updated 2026-04-24
View scorecard diff history → · JSON CSV Cite

The Sports Prediction Transparency Index

Most "sports prediction" providers won't tell you their Expected Calibration Error, keep their methodology secret, or hide pricing behind "contact us" forms. We score 27 real sources on seven dimensions anyone can verify. This page measures transparency, not predictive edge, ROI, or who has the best model.

Last automated re-check: 2026-04-24 · Last human-verified row: 2026-04-29 · Updated monthly · Suggestions reviewed by hand before they ship.

What this index measures

Public evidence: calibration disclosure, historical archives, methodology transparency, API access, visible pricing, track-record longevity, and sport coverage breadth. Equal weights are intentional. If you disagree with the weighting, compare rows dimension-by-dimension instead of relying on the total.

What this index does NOT measure
  • Predictive edge or profitability
  • Market-making quality or trading liquidity
  • Customer support, UX polish, or brand size
  • Who would win in a head-to-head model contest
Self-reported ECE roll call

Of 20 sports forecasters tracked, only 1 publish a numeric Expected Calibration Error

ECE is the single most important honesty metric for a probability product. Most providers won't publish theirs. Buckets: Numeric — an actual ECE figure on a defined holdout. Derivable — raw forecasts + outcomes are public so ECE can be computed. Partial — publishes accuracy / Brier / reliability but not ECE. Silent — publishes nothing about model accuracy.

7 sources are excluded from the denominator above because they are venues, aggregators, or closing-line providers — they don't claim to be forecasters.

1
Numeric
1
Derivable
7
Partial
11
Silent
ZenHodl disclosure
#2 of 27

How we score ourselves

ZenHodl is shown separately because this is our rubric. We include ourselves for disclosure, but we do not want a self-ranked row mixed into the main competitor tables. Our current score is 29/35, ranked #2 of 27 based on public evidence we link to below.

Self-reported ECE: 4.39% — verifiable from the linked NCAAMB season report. We commit to publishing this number for every sport, every season, including when it gets worse.

Last verified: 2026-04-24. Notes: Per-sport ECE published (NCAAMB 4.39% on 5,345 games). Every trade on /results. API + docs. Tiered pricing. Methodology docs partial — training code not open sourced. Multi-sport (8+) but young (<2 yrs).

Dimension scores
ECE█████
Archive█████
Method███··
API█████
Price█████
Years██···
Sports████·
Companion: live CLV scorecard

Public disclosure rubric — meet the public measurement.

This index scores public evidence of disclosure. /clv publishes the empirical companion: per-sport closing-line value across the 686 measured ZenHodl trades currently in the public dataset. The headline finding — trades that beat the close win 88.6% [84.8%, 91.5%] of the time, trades that lose the close win 11.3% [8.4%, 15.1%] — is the kind of measurement the index's calibration-disclosure dimension is designed to detect.

The two pages compound: this index ranks 27 sources on what they disclose; /clv shows ZenHodl's own disclosure in practice. /clv/repair documents which of our sports are paused for being CLV-negative and the explicit criteria for reactivation.

Open CLV scorecard → Open repair status →
/clv.json /clv.csv /clv/failures.json

Tip: click + compare on any row to add it to a side-by-side comparison panel. Selections sync to the URL so you can share the comparison link.

Model / Analytics Providers

Model-driven sources that publish projections, ratings, or probabilistic forecasts.

Source Rank Total ECE Archive Method API Price Years Sports Reported ECE Verified
FiveThirtyEight (archived) #
Model / Analytics Providers
Why this score?
Archived since 2023. All predictions + data on GitHub. Best methodology docs ever. Dead product.
#1 30/35 ███·· █████ █████ ███·· █████ █████ ████· Brier published; ECE derivable from raw data ↗ 2026-04-24
Inpredictable #
Model / Analytics Providers
Why this score?
Long-running NFL/NBA win-probability research blog by Mike Beuoy. Free. Detailed methodology posts. Charts archived. No formal API.
#4 24/35 ██··· ████· ████· █···· █████ █████ ███·· Brier / log-loss in blog posts (not ECE) ↗ 2026-04-29
Bart Torvik #
Model / Analytics Providers
Why this score?
College basketball. Excellent reliability tables every March. Free. Unstable JSON endpoints.
#5 23/35 ████· ████· ████· █···· █████ ████· █···· reliability tables (no single ECE figure) ↗ 2026-04-24
MoneyPuck #
Model / Analytics Providers
Why this score?
NHL-only, XGBoost-based. Methodology documented. Limited JSON availability.
#6 22/35 ███·· ███·· ████· ██··· █████ ████· █···· log-loss reported, not ECE ↗ 2026-04-24
Massey Ratings #
Model / Analytics Providers
Why this score?
Free college ratings. HTML only, no API, some methodology detail.
#8 22/35 █···· ███·· ███·· ····· █████ █████ █████ accuracy by sport published (not ECE) ↗ 2026-04-24
KenPom #
Model / Analytics Providers
Why this score?
College basketball only. Methodology in FAQ. $20/yr clear. HTML only, no API.
#13 19/35 ███·· ██··· ███·· ····· █████ █████ █···· — 2026-04-24
numberFire (archived) #
Model / Analytics Providers
Why this score?
FanDuel-acquired projections / DFS site. Now mostly content. Methodology never fully public; ECE never published. Listed as historical reference.
#23 14/35 ····· ██··· █···· ····· ███·· ████· ████· — 2026-04-29

Odds / Aggregation Tools

Products built around market comparison, line shopping, or expected-value screens.

Source Rank Total ECE Archive Method API Price Years Sports Reported ECE Verified
OddsJam #
Odds / Aggregation Tools
Why this score?
Aggregator, not a model. EV finder uses market consensus. Published pricing, tiered API.
#17 16/35 ····· ····· █···· ████· ███·· ███·· █████ — 2026-04-24
Unabated #
Odds / Aggregation Tools
Why this score?
Sharp analytics, line shopping. No published accuracy. Tiered pricing.
#18 15/35 ····· ····· ██··· ██··· ███·· ███·· █████ — 2026-04-24

Picks / Content Sites

Consumer-facing picks, expert selections, and media-driven betting content.

Source Rank Total ECE Archive Method API Price Years Sports Reported ECE Verified
Action Network #
Picks / Content Sites
Why this score?
Affiliate-driven content. 'Entertainment purposes only.' Expert picks tracked publicly per-author. PRO Reports document model methodology partially. No ECE published.
#19 15/35 ····· █···· █···· █···· ███·· ████· █████ — 2026-04-24
BetQL #
Picks / Content Sites
Why this score?
'AI-powered' picks. Published accuracy claims not ECE. Consumer subscription.
#20 14/35 ····· █···· █···· █···· ████· ███·· ████· accuracy% claims (not ECE) 2026-04-24
SportsLine #
Picks / Content Sites
Why this score?
CBS property. Expert picks. No ECE. Published subscription pricing.
#21 14/35 ····· ····· █···· ····· ████· ████· █████ — 2026-04-24
Prediction Machine #
Picks / Content Sites
Why this score?
Monte-Carlo simulation picks site (Bloomberg-syndicated for years). Claims accuracy but no holdout, no ECE, no auditable archive.
#24 14/35 ····· █···· █···· ····· ███·· ████· █████ — 2026-04-29
Covers #
Picks / Content Sites
Why this score?
Consensus picks aggregator. No methodology, no ECE.
#25 13/35 ····· ····· ····· ····· ███·· █████ █████ — 2026-04-24
Outlier #
Picks / Content Sites
Why this score?
Player props focus. Hit-rate / streak summaries shown in-app but no auditable holdout. No methodology, no API.
#27 11/35 ····· █···· ····· ····· ████· ██··· ████· self-reported user 'hit rate' on app 2026-04-29

Prediction Markets / Venues

Trading venues and market platforms. These are included for transparency comparison, not because they are model vendors.

Source Rank Total ECE Archive Method API Price Years Sports Reported ECE Verified
Manifold Markets #
Prediction Markets / Venues
Why this score?
Play-money. Real forecasters, real leaderboards. Full API. Not real-money.
#3 27/35 ██··· █████ ████· █████ █████ ███·· ███·· leaderboard Brier scores (not platform-wide ECE) ↗ 2026-04-24
Polymarket #
Prediction Markets / Venues
Why this score?
Global prediction market. CLOB API public. On-chain historical data. Not US-legal for direct trading.
#7 22/35 ····· ███·· ██··· █████ █████ ███·· ████· n/a (venue, not a forecaster) 2026-04-24
PredictIt #
Prediction Markets / Venues
Why this score?
Academic. Historical data downloadable. $850 position cap limits serious use.
#10 21/35 █···· ████· ███·· ███·· ████· ████· ██··· n/a (venue, not a forecaster) 2026-04-24
Kalshi #
Prediction Markets / Venues
Why this score?
CFTC-regulated prediction market. Full API. Not a model — a venue.
#15 17/35 ····· ██··· ██··· █████ ███·· ██··· ███·· n/a (venue, not a forecaster) 2026-04-24

Enterprise / Data Vendors

API-first or institutional data providers that sell feeds, odds, or probability products.

Source Rank Total ECE Archive Method API Price Years Sports Reported ECE Verified
The Odds API #
Enterprise / Data Vendors
Why this score?
Aggregated book odds, not a model. Transparent pricing, real API.
#9 21/35 ····· █···· ██··· █████ █████ ███·· █████ n/a (odds aggregator, not a model) 2026-04-24
Sportradar #
Enterprise / Data Vendors
Why this score?
Enterprise-grade. Published internal ECE in whitepapers occasionally. Full API. No public pricing.
#11 20/35 ██··· ····· ███·· █████ ····· █████ █████ — 2026-04-24
Stats Perform #
Enterprise / Data Vendors
Why this score?
Institutional-grade probability feeds. Sparse public benchmarks. Custom contracts.
#12 20/35 ██··· ····· ███·· █████ ····· █████ █████ — 2026-04-24
SportsDataIO #
Enterprise / Data Vendors
Why this score?
API-first sports data. Some ML/projections but no ECE. Published pricing.
#14 19/35 ····· ····· ██··· █████ ████· ███·· █████ — 2026-04-24
Genius Sports #
Enterprise / Data Vendors
Why this score?
Official data partner for NFL / NCAA / others. Probability feeds + trading services. Public ECE never published. Custom contracts only.
#16 17/35 █···· ····· ██··· █████ ····· ████· █████ — 2026-04-29
Kambi #
Enterprise / Data Vendors
Why this score?
Sportsbook turnkey provider. Powers many regulated US books behind the scenes. Quant team internal — no public model docs or ECE.
#22 14/35 ····· ····· █···· ████· ····· ████· █████ — 2026-04-29
Pinnacle (closing lines) #
Enterprise / Data Vendors
Why this score?
Gold-standard closing lines (since 1998). Implicit benchmark via market. Not a forecaster — listed for reference; transparency dimensions don't apply cleanly. No direct API sales.
#26 11/35 ····· ····· █···· ····· ····· █████ █████ implicit (closing line is a market, not a model) 2026-04-24
1. ECE
Do they publish Expected Calibration Error on a known holdout?
5 = reliability table per sport · 0 = never mentioned
2. Archive
Can you see every past prediction, including misses?
5 = full auditable ledger · 0 = no archive
3. Method
Is the methodology documented and reproducible?
5 = open-source or detailed paper · 0 = "proprietary AI"
4. API
Programmatic access for developers?
5 = full REST + SDK + docs · 0 = HTML scraping
5. Price
Pricing visible before you contact sales?
5 = tiered pricing published · 0 = "contact us"
6. Years new
How long has this source been publicly tracked?
5 = 10+ yrs · 4 = 5-10 yrs · 3 = 3-5 yrs · 2 = 1-3 yrs · 0 = under 1 yr
Longevity is a counter-weight to "new shiny model that hasn't been tested through a full season."
7. Sports new
How many sports does this source actually cover?
5 = 8+ sports · 4 = 5-7 · 3 = 3-4 · 2 = 2 · 1 = 1 · 0 = none
Coverage breadth recognises that a single-sport specialist (e.g. KenPom) is doing something different than a multi-sport platform.
Equal weights, by design
All seven dimensions weight equally. We don't tune the weights to favour our own profile — if we did, sport-coverage and longevity would weight lower because we score weakly on them.

Citing this index

The data is licensed CC BY 4.0 — quote, link, and reproduce freely with attribution. Programmatic access via the JSON / CSV endpoints is rate-unlimited for non-abusive use.

ZenHodl Sports Prediction Transparency Index. Accessed . https://zenhodl.net/transparency-index
Dataset
.json
Dataset
.csv
Provenance
scorecard diff history →

Why this index exists

There's no shortage of sites selling sports predictions. There's a chronic shortage of sites willing to tell you how accurate they actually are. We built this scorecard because when we went looking for a benchmark to beat, nothing existed. So we made one.

ZenHodl appears on this page for disclosure, but we separate our own score from the main grouped tables to avoid the obvious conflict of interest. We publish Expected Calibration Error per sport (4.39% on 5,345 NCAAMB games), keep every trade on /results, and document our methodology in our blog. We don't score ourselves 5/5 on methodology because we haven't open-sourced the training code, and we score weakly on track-record longevity because we're under two years old. Both honest about where we're not yet.

This index is re-checked monthly. If you think a score is wrong, tell us — we'll re-verify and add a public note in the next refresh.

Methodology. Scores are manually curated based on publicly visible homepage / methodology / pricing pages as of each row's Verified date. We do not score internal commercial claims we can't independently verify. Archived projects, picks sites, venues, and data vendors are grouped separately to reduce apples-to-oranges comparisons.

Fairness. Equal weighting is intentional and pre-registered. We did not tune dimension weights after seeing scores. The two newest dimensions (track record longevity and sport coverage breadth) are areas where ZenHodl scores weakly — they were added because credible challengers raised them, not because they help our ranking.

Disclosure. Pinnacle and prediction-market venues appear in the table for reference but they aren't forecasters in the same sense — they're priced markets. Their dimension scores reflect the rubric mechanically, but a low total there means "this is a market venue, not a model" rather than "this product is bad."

Reproducibility. The scoring source data is in api/transparency_index_data.json in our public methodology references; monthly snapshots are kept in api/transparency_index_history/ so the trajectory of every score is auditable. Suggested score changes from the monthly Claude re-check are reviewed by hand before they ever go live.

Found something wrong?
Challenge a score →
Send the evidence and we'll re-check it. Approved corrections are logged on the diff history page.
Missing a source?
Suggest a competitor →
Pre-filled email with the fields we need. We add credible sources within one monthly cycle.