Season one starts October 20
NBA grading begins on opening night. Results post after every slate. Tools are ranked once 20 slates are graded, expected in mid-November.
Every DFS tool claims to be the best.We keep score.
DFS Benchmark grades projections, ownership, optimizers and AI against what actually happened — under rules published before the first slate.
IntroducingRulebook v1
Every DFS tool, graded the same way.
Introducing DFS Benchmark
What we grade, how we grade it, and why nothing gets ranked until 20 slates are in. Read the report →01Captured before lockEvery tool at the same minute, on default settings.
02Graded after the gamesOfficial DraftKings results, one shared player pool.
03Published with rangesNo tool leads a row until its gap beats the noise.
NowRulebook v1 draft is public. It becomes final before the first capture.
Oct 20NBA opening night: the first graded slate.
Mid-NovemberThe first ranked leaderboard, after 20 graded slates.
Every MondayThe weekly email: standings, slate winners and perfect lineups.
Who's tested
The leading DFS tools, a house entry, Claude, and two simple baselines every tool should beat.
Tools Division
- SaberSimPending
- StokasticPending
- RotoGrindersPending
- FantasyLabsPending
Pending means awaiting the company’s consent. A tool is graded only after its company agrees. Any other DFS tool can apply under the same rules.
House entry
- The DFS KitchenHouse
Owned by DFS Benchmark's founders. Same rules, captured first, labeled House everywhere.
AI Division
- ClaudeAI
The same fixed prompt every slate, web search on, salary file attached.
Baselines
- Recent-form
- Salary
Recent form: last 10 games (NBA). Salary: salary × the slate's points per $1K.
What we measure
Eight headline benchmarks rank the tools. Beat-the-Baseline leads: how much less error a tool makes than a simple recent-form average.
| # | Benchmark | What it measures | How it's scored | Better |
|---|---|---|---|---|
| 1 | Beat-the-Baseline | Projection skill over a naive forecast | 1 − source error ÷ baseline error, same player pool | Higher |
| 2 | Points Accuracy | Average points missed per player | Mean absolute error in DK points, by position group | Lower |
| 3 | Rank Accuracy | Whether players were ordered correctly | Rank correlation of projected vs actual, within each position | Higher |
| 4 | Value Finder | Whether it found the slate's best values | Of the 10 best actual values (DK points per $1K), how many were in its top 20 projected values | Higher |
| 5 | Ownership Accuracy | Whether it knew the crowd | Mean absolute error in ownership %, reference GPP | Lower |
| 6 | Chalk Finder | Whether it spotted the most-owned players | Overlap of its top 5 projected ownership with the actual top 5 | Higher |
| 7 | % of Perfect | Same-engine lineup quality | Same-engine lineup points ÷ best possible lineup's points | Higher |
| 8 | Cash-Line Rate | Whether the lineup would have cashed | Share of slates the same-engine lineup beat the cash line | Higher |
AI Division
Can Claude beat the DFS tools?
Claude gets the same fixed prompt at the same minute every slate, with web search on and the slate's salary file attached. Its lineup is graded like everyone else's, plus two checks only AI needs.
See the AI Division →Valid lineup rateLegal lineups: salary cap, positions and players who are on the slate.
Stale info rateLineups that include a player already ruled out at the capture minute.
Published in fullClaude's whole answer is posted after grading, beside its results, with the model version it ran on.
Badges and season awards
A tool earns a badge only with a clear lead: its range clears the next tool's after the minimum sample. Awards are decided when each sport's season ends. Neither can be bought.
- Most Accurate Projections
- Best Ownership
- Best Value Finder
- Best Value for Money
- Most Improved
DFS Benchmark
#1
Ownership Accuracy
NBA 2026–27
Tool name
Example
How DFS Benchmark works
Read the full rulebook →1Same rules for every toolDefault settings, the same capture minute, the same grading.
2Rules before dataThe rulebook is published before the first capture. Changes never apply to past slates.
3Locked at lockEvery captured file is timestamped and fingerprinted before games start.
4Ranges, not hypeA tool only leads a row when the gap is bigger than the noise.
5Consent firstCompanies are graded by invitation. Anyone who declines is listed as declined.
6No pay for placementParticipation is free, and money never touches results.
The DFS Benchmark Weekly
Standings changes, slate winners, the week's perfect lineups and the biggest form movers. Every Monday, starting after the first week of NBA grading.