Was FFT right?

The call record

Every BUY and FADE below was written down before the 2026 season started, with the evidence attached and a timestamp on it. Once games are played each one gets graded against the market’s own price — and the misses stay on the page, because a record you can only lose from is the only kind worth publishing.

Nothing has been graded yet.

Grading begins after Week 1 of the 2026 season. There are no results to score these calls against, so there is no track record on this page — and there will not be an invented one. What is here instead is 29 calls, timestamped 2026-08-08 06:38 UTC, which is the part that cannot be faked afterwards.

Scoreboard

Calls registered

Graded

Hit rate

—

No call decided yet. Grading begins after Week 1.

Closer than market

—

Needs a decided call.

Cut	Registered	Hit rate
All calls	29	—
BUY	15	—
FADE	14	—
QB	7	—
RB	7	—
TE	5	—
WR	10	—

A dash in the hit-rate column means nothing has been decided in that cut. It does not mean zero.

How these calls were registered

2026-08-08-preseason-board29 calls

EVERY directional verdict the board produced at boardComputedAt — all 15 BUY and 14 FADE from a scored pool of 615 players, 160 of whom carried a market price. No filtering by conviction, position or confidence, so the record cannot be flattered by selection.

Registered at: 2026-08-08 06:38 UTC
Board computed at: 2026-08-08 06:33 UTC
Model version at registration: fft-projection-1.0.0-beta
Edge engine version at registration: fft-edges-1.0.0
Scored pool: 615 players
Players with a market price: 160
Chain link: none — registered before the chain existed

Which model made these calls. Registered under fft-projection-1.0.0-beta (shipped 2026-08-07, superseded 2026-08-08). The model now shipping — the one behind /api/edges and the figures on /methodology — is fft-projection-2.3.0-beta. That difference is deliberate and it is not going to be tidied away: a call stays attributed to the model that actually made it. Restamping these rows with the current version would let a newer model inherit an older model's record, which is exactly the manoeuvre this page exists to make impossible. Grading uses the price and the outcome, not the model, so a superseded model does not make a call ungradable. What that version was: First market-independent opportunity model. Per-game volume shrunk toward a single flat positional mean; rookies from players.csv draft capital only. The edge engine carries its own version, independent of the model's: fft-edges-1.0.0 assembled this board. /api/edges reports the assembly version serving it right now, which is the one to compare this against — the two identifiers move on separate clocks, so a difference between them is expected rather than an inconsistency.

Consensus of the Yahoo, Sleeper and RTSports ADP columns in the static dataset snapshot bundled with the app at registration time — a snapshot, not a live feed. The ADP recorded on each call is the price the market was charging when the call was made.

What attests to these timestamps — and what does not

Nothing here is hash-chained yet. The append-only chain exists in the code and every future registration and grading run will be sealed into it, but no entry currently carries a link. Saying that plainly is the point: a claim about tamper-evidence that nothing is actually using is not a guarantee, it is an intention.
1 entry predates the chain. It was deliberately not retro-sealed. Sealing an old record with a new tool is exactly the move a laundered record would make, so it rests on the git commit that introduced it and nothing else — and this sentence stays here until that is no longer true.
The chain does not prove the clock. Whoever can rewrite this file can also recompute the chain. What supplies the time is the git commit that added each batch — a reviewer can read the history and see when the calls entered it — plus the chain head above, which is a single string covering the whole record and can be published somewhere with a clock we do not control. That is the honest ceiling of the current setup, and it is weaker than an append-only store with a server-assigned timestamp.

How a call gets graded

Against the market, not against a target we set

A BUY says the market is too low on a player: he should finish earlier than his ADP. A FADE says the opposite. The call is correct only if he finishes at least 8 picks on the called side of the price the market was charging — the same band the board had to clear to make the call in the first place. Inside that band, neither side was shown right, and it scores as a push.

On the same ruler the call was written on

The comparable pool — every player carrying a market price on that board — is frozen at registration and stored with the calls. At grading, the pool is ranked by actual points and each finish is mapped back onto the pool's own ADP scale. A pool that could be re-cut afterwards is a pool that could be cut to flatter the record.

Pushes are not counted as losses, and not counted as wins

The hit rate is right divided by right-plus-wrong. Pushes leave the denominator. Until something is decided the hit rate is not a number at all.

Injuries are flagged, never dropped

A player who plays under 60 percent of possible games is still graded and still counted, with a flag saying the result was about availability rather than production. Removing the injuries is how a track record gets laundered.

Small samples are labelled as small

Below 30 decided calls the percentage is dominated by luck, and this page says so next to the number rather than letting an early streak read as skill.

The pre-registered calls

15 buy · 14 fade

BUY — the market is drafting him later than we rank him

BUY

Cam Ward

QB · TEN

Awaiting result

Model rank: 142.2
Market ADP: 165.7
Divergence: +23.5

td regressionScored 17 TDs on 59 red-zone opportunities — opportunity implies about 23.43. Finishing that far below expectation tends to regress upward, and the market is still pricing the 17.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 1.00 · 145 of 160 by market price · 2026-08-08-camward-qb-buy

BUY

Jalen Coker

WR · CAR

Awaiting result

Model rank: 144.2
Market ADP: 162.7
Divergence: +18.5

usage qualityHis route-concept mix is worth 0.5 EPA per target against a 0.05 league baseline. His most common concept is quick out at 25% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 1.00 · 142 of 160 by market price · 2026-08-08-jalencoker-wr-buy

BUY

Travis Kelce

TE · KC

Awaiting result

Model rank: 81.8
Market ADP: 101.3
Divergence: +19.5

usage qualityHis route-concept mix is worth 0.36 EPA per target against a 0.05 league baseline. His most common concept is quick out at 28% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 1.00 · 92 of 160 by market price · 2026-08-08-traviskelce-te-buy

BUY

Stefon Diggs

WR · WAS

Awaiting result

Model rank: 135.2
Market ADP: 160.3
Divergence: +25.2

usage qualityHis route-concept mix is worth 0.58 EPA per target against a 0.05 league baseline. His most common concept is quick out at 32% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
efficiencycatches 16 points more often than his 8.52-yard target depth predicts, and +0.54 yards after catch versus expectation. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.84 · 139 of 160 by market price · 2026-08-08-stefondiggs-wr-buy

BUY

Kyren Williams

RB · LAR

Awaiting result

Model rank: 10.1
Market ADP: 28.7
Divergence: +18.6

usage qualityHis route-concept mix is worth 0.29 EPA per target against a 0.05 league baseline. His most common concept is flat at 56% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.79 · 27 of 160 by market price · 2026-08-08-kyrenwilliams-rb-buy

BUY

Baker Mayfield

QB · TB

Awaiting result

Model rank: 96.2
Market ADP: 140.7
Divergence: +44.5

td regressionScored 27 TDs on 64 red-zone opportunities — opportunity implies about 31.31. Finishing that far below expectation tends to regress upward, and the market is still pricing the 27.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.72 · 120 of 160 by market price · 2026-08-08-bakermayfield-qb-buy

BUY

Courtland Sutton

WR · DEN

Awaiting result

Model rank: 69.2
Market ADP: 86
Divergence: +16.9

usage qualityHis route-concept mix is worth 0.32 EPA per target against a 0.05 league baseline. His most common concept is quick out at 33% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
opportunity7.29 targets per game across 17 games. Volume is the most stable input year over year, so this is the safest part of any projection.nflverse season stats

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.68 · 84 of 160 by market price · 2026-08-08-courtlandsutton-wr-buy

BUY

Oronde Gadsden II

TE · LAC

Awaiting result

Model rank: 100.8
Market ADP: 136.7
Divergence: +35.8

efficiencycatches 4.5 points more often than his 9-yard target depth predicts, and +1.45 yards after catch versus expectation. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)
usage qualityHis route-concept mix is worth 0.27 EPA per target against a 0.05 league baseline. His most common concept is quick out at 22% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.68 · 118 of 160 by market price · 2026-08-08-orondegadsden-te-buy

BUY

Sam Darnold

QB · SEA

Awaiting result

Model rank: 115.2
Market ADP: 152
Divergence: +36.8

td regressionScored 25 TDs on 72 red-zone opportunities — opportunity implies about 29.01. Finishing that far below expectation tends to regress upward, and the market is still pricing the 25.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.67 · 133 of 160 by market price · 2026-08-08-samdarnold-qb-buy

BUY

Khalil Shakir

WR · BUF

Awaiting result

Model rank: 71.2
Market ADP: 158
Divergence: +86.8

efficiency+1.86 yards after catch versus expectation. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.63 · 135 of 160 by market price · 2026-08-08-khalilshakir-wr-buy

BUY

Matthew Golden

WR · GB

Awaiting result

Model rank: 143.2
Market ADP: 151.7
Divergence: +8.5

usage qualityHis route-concept mix is worth 0.5 EPA per target against a 0.05 league baseline. His most common concept is screen at 20% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
td regressionScored 0 TDs on 4 red-zone opportunities — opportunity implies about 2.35. Finishing that far below expectation tends to regress upward, and the market is still pricing the 0.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.62 · 131 of 160 by market price · 2026-08-08-matthewgolden-wr-buy

BUY

Romeo Doubs

WR · NE

Awaiting result

Model rank: 98.2
Market ADP: 142
Divergence: +43.9

usage qualityHis route-concept mix is worth 0.49 EPA per target against a 0.05 league baseline. His most common concept is quick out at 27% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
efficiencycatches 4.4 points more often than his 12.81-yard target depth predicts. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.59 · 122 of 160 by market price · 2026-08-08-romeodoubs-wr-buy

BUY

Bo Nix

QB · DEN

Awaiting result

Model rank: 32.2
Market ADP: 106.7
Divergence: +74.5

td regressionScored 30 TDs on 97 red-zone opportunities — opportunity implies about 33.46. Finishing that far below expectation tends to regress upward, and the market is still pricing the 30.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.58 · 96 of 160 by market price · 2026-08-08-bonix-qb-buy

BUY

Dalton Schultz

TE · HOU

Awaiting result

Model rank: 112.8
Market ADP: 172.7
Divergence: +59.8

usage qualityHis route-concept mix is worth 0.38 EPA per target against a 0.05 league baseline. His most common concept is quick out at 42% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
efficiencycatches 6.3 points more often than his 6.26-yard target depth predicts. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.57 · 148 of 160 by market price · 2026-08-08-daltonschultz-te-buy

BUY

Juwan Johnson

TE · NO

Awaiting result

Model rank: 107.8
Market ADP: 151.3
Divergence: +43.5

usage qualityHis route-concept mix is worth 0.35 EPA per target against a 0.05 league baseline. His most common concept is quick out at 25% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
efficiencycatches 6.2 points more often than his 7.35-yard target depth predicts. That is earned production rather than volume, and it tends to persist.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.57 · 130 of 160 by market price · 2026-08-08-juwanjohnson-te-buy

FADE — the market is drafting him earlier than we rank him

FADE

Josh Allen

QB · BUF

Awaiting result

Model rank: 49.2
Market ADP: 21.7
Divergence: -27.5

td regressionScored 39 TDs on 92 red-zone opportunities — opportunity implies only about 32.56. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 1.00 · 22 of 160 by market price · 2026-08-08-joshallen-qb-fade

FADE

Derrick Henry

RB · BAL

Awaiting result

Model rank: 47.1
Market ADP: 21.3
Divergence: -25.8

td regressionScored 16 TDs on 64 red-zone opportunities — opportunity implies only about 10.13. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.98 · 21 of 160 by market price · 2026-08-08-derrickhenry-rb-fade

FADE

Brock Purdy

QB · SF

Awaiting result

Model rank: 129.2
Market ADP: 104
Divergence: -25.2

td regressionScored 23 TDs on 62 red-zone opportunities — opportunity implies only about 17.64. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.89 · 93 of 160 by market price · 2026-08-08-brockpurdy-qb-fade

FADE

Josh Jacobs

RB · GB

Awaiting result

Model rank: 39.1
Market ADP: 28.3
Divergence: -10.8

td regressionScored 14 TDs on 54 red-zone opportunities — opportunity implies only about 9.58. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.74 · 26 of 160 by market price · 2026-08-08-joshjacobs-rb-fade

FADE

Zach Charbonnet

RB · SEA

Awaiting result

Model rank: 157.1
Market ADP: 134.3
Divergence: -22.8

td regressionScored 12 TDs on 53 red-zone opportunities — opportunity implies only about 8.18. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.64 · 115 of 160 by market price · 2026-08-08-zachcharbonnet-rb-fade

FADE

Kenny Gainwell

RB · TB

Awaiting result

Model rank: 133.1
Market ADP: 107.3
Divergence: -25.8

td regressionScored 8 TDs on 36 red-zone opportunities — opportunity implies only about 5.3. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.45 · 97 of 160 by market price · 2026-08-08-kennygainwell-rb-fade

FADE

Cam Skattebo

RB · NYG

Awaiting result

Model rank: 86.1
Market ADP: 37
Divergence: -49.1

td regressionScored 7 TDs on 27 red-zone opportunities — opportunity implies only about 4.42. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.43 · 34 of 160 by market price · 2026-08-08-camskattebo-rb-fade

FADE

Dak Prescott

QB · DAL

Awaiting result

Model rank: 95.2
Market ADP: 83
Divergence: -12.2

td regressionScored 32 TDs on 111 red-zone opportunities — opportunity implies only about 29.88. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.35 · 79 of 160 by market price · 2026-08-08-dakprescott-qb-fade

FADE

Mark Andrews

TE · BAL

Awaiting result

Model rank: 148.8
Market ADP: 116
Divergence: -32.8

td regressionScored 6 TDs on 17 red-zone opportunities — opportunity implies only about 3.9. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model

Registered 2026-08-08 06:38 UTC · confidence low · signal strength 0.35 · 108 of 160 by market price · 2026-08-08-markandrews-te-fade

FADE

Davante Adams

WR · LAR

Awaiting result

Model rank: 109.2
Market ADP: 56.3
Divergence: -52.8

td regressionScored 14 TDs on 31 red-zone opportunities — opportunity implies only about 4.82. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model
usage qualityHis route-concept mix is worth 0.23 EPA per target against a 0.05 league baseline. His most common concept is quick out at 19% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play
efficiencycatches 7 points less often than his 13.22-yard target depth predicts, and -0.55 yards after catch versus expectation. Volume is carrying his production more than skill is.NFL Next Gen Stats (depth-adjusted)
opportunity8.14 targets per game across 14 games. Volume is the most stable input year over year, so this is the safest part of any projection.nflverse season stats

Registered 2026-08-08 06:38 UTC · confidence high · signal strength 0.32 · 57 of 160 by market price · 2026-08-08-davanteadams-wr-fade

FADE

Mike Evans

WR · SF

Awaiting result

Model rank: 155.2
Market ADP: 60
Divergence: -95.2

efficiencycatches 10.6 points less often than his 13.56-yard target depth predicts, and -0.51 yards after catch versus expectation. Volume is carrying his production more than skill is.NFL Next Gen Stats (depth-adjusted)
opportunity7.75 targets per game across 8 games. Volume is the most stable input year over year, so this is the safest part of any projection.nflverse season stats

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.29 · 60 of 160 by market price · 2026-08-08-mikeevans-wr-fade

FADE

RJ Harvey

RB · DEN

Awaiting result

Model rank: 90.1
Market ADP: 75.7
Divergence: -14.4

td regressionScored 12 TDs on 36 red-zone opportunities — opportunity implies only about 7.23. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model
usage qualityHis route-concept mix is worth 0.24 EPA per target against a 0.05 league baseline. His most common concept is flat at 49% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.26 · 69 of 160 by market price · 2026-08-08-rjharvey-rb-fade

FADE

Tee Higgins

WR · CIN

Awaiting result

Model rank: 58.1
Market ADP: 38
Divergence: -20.1

td regressionScored 11 TDs on 14 red-zone opportunities — opportunity implies only about 5.06. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model
usage qualityHis route-concept mix is worth 0.36 EPA per target against a 0.05 league baseline. His most common concept is quick out at 24% of targets. He is used in higher-value ways than his target count alone suggests.FFT route concepts derived from nflverse play-by-play

Registered 2026-08-08 06:38 UTC · confidence medium · signal strength 0.24 · 36 of 160 by market price · 2026-08-08-teehiggins-wr-fade

FADE

Rashee Rice

WR · KC

Awaiting result

Model rank: 56.1
Market ADP: 29.7
Divergence: -26.5

td regressionScored 6 TDs on 21 red-zone opportunities — opportunity implies only about 3.09. Touchdown rates that far above expectation rarely repeat, and the market is paying for them.Sportradar red-zone opportunity + FFT finishing-rate model
opportunity9.75 targets per game across 8 games. Volume is the most stable input year over year, so this is the safest part of any projection.nflverse season stats
efficiencycatches 6.1 points less often than his 4.46-yard target depth predicts, and +0.58 yards after catch versus expectation. Volume is carrying his production more than skill is.NFL Next Gen Stats (depth-adjusted)

Registered 2026-08-08 06:38 UTC · confidence high · signal strength 0.15 · 29 of 160 by market price · 2026-08-08-rasheerice-wr-fade

Read this before quoting anything above

No call has been graded. 29 calls are registered and grading begins after Week 1 of the 2026 season, when there are real results to score them against. There is no track record here yet, and an empty scoreboard is the honest state of one.
A call is graded against THE MARKET'S OWN PRICE, not against a target we set. It is correct only when the player finishes at least 8 picks on the called side of his ADP — the same band the edge board had to clear to make the call in the first place.
Beating naive statistical baselines in the projection backtest is NOT beating the market. This scoreboard is the first thing in the product that tests the market claim at all, and it can come back against us.
Availability is not modelled. A called player who misses most of the season is still graded and still counted, but flagged, because dropping the injuries is how a track record gets laundered.
1 entry in this record predate the hash chain and rest on their git commit alone. They were not retro-sealed, because sealing an old record with a new tool is exactly what a laundered record would do. 0 entries are chained and verify.
Batch 2026-08-08-preseason-board: 29 calls registered 2026-08-08T06:38:11.545Z from a board computed 2026-08-08T06:33:00.114Z. EVERY directional verdict the board produced at boardComputedAt — all 15 BUY and 14 FADE from a scored pool of 615 players, 160 of whom carried a market price. No filtering by conviction, position or confidence, so the record cannot be flattered by selection.

How the model works, and where it loses This record as JSON