pick6

README

pick6

Terminal war room for fantasy drafts. Live-syncs Sleeper (and FPL Draft), tracks tiers, and tells you when the RB run means the value's gone.

A scripted room drafts, the cliff banner fires, the ranking re-sorts under it; a stops the clock, / finds a player, and tab drops to every number the engine holds. Recorded from the real thing. docs/demo.tape, vhs docs/demo.tape to regenerate.

what it does

• Survival odds on every player. The chance he's still there at your next pick, by simulating the picks between you and your turn 500 times against your opponents' actual rosters: their open slots, your room's measured appetites, and the measured share of real picks that go to players no board ranks. (-survival=adp falls back to the pure price-curve model.)
• Tiers that alarm. Amber when a tier is probably gone before you act, red when the man on screen is the last of it. Tiers come from your rankings file; where it runs out, they're derived from the value curve.
• Run detection that isn't noise. Four RBs in six picks only fires when it beats what the market itself expected for those picks.
• Your room, not the market. Feed it your league's past drafts and it reprices survival against how those people actually draft.
• A verdict, not a spreadsheet. On the clock it names a man and the two or three facts that justify him, in picks and odds rather than abstract points.
• Two sports, one engine. -sport fpl points the whole thing at a Fantasy Premier League draft (beta). The engine never asks which game it is scoring; it reads your lineup and derives the rest.

install

Download a release (no toolchain needed). Grab the tarball for your platform from the releases page, then:

tar -xzf pick6_*.tar.gz
xattr -d com.apple.quarantine pick6   # macOS only: clears the "unverified developer" block
mv pick6 ~/bin/                       # or anywhere on your PATH
pick6 version

checksums.txt on the release page carries the sha256 of every tarball.

Or with Go:

go install github.com/trisslazaj/pick6/cmd/pick6@latest

That installs to go env GOPATH/bin, which is often not on PATH. GOBIN=$HOME/bin go install ... puts it somewhere that is. Working on the code? go run ./cmd/pick6 <cmd> always runs current source.

quickstart

pick6 fetch                          # 1. pull data (players, adp, values, tiers)
pick6 live <draft_id> -user yourname # 2. sync your live sleeper draft

That's the whole thing. Your draft id is in the url: open the draft room in a browser and it reads sleeper.com/draft/nfl/<draft_id>. -user finds your seat from the draft order; -slot 4 sets it directly if the draft hasn't been seeded yet.

No Sleeper league? pick6 board runs the same war room with you typing the picks.

commands

pick6 fetch

Downloads the Sleeper player pool, half-PPR ADP, and player values; joins everything onto Sleeper ids; applies your rankings; caches it all. Run it once before drafting and again on draft morning; inside 12 hours it's served from disk.

pick6 fetch                            # defaults: 12-team half-ppr
pick6 fetch -format ppr                # other scoring
pick6 fetch -rankings my-tiers.csv     # your tiers and points win (see below)

pick6 live <draft_id>

Polls the draft every few seconds and updates the board in place. Every pick is cross-checked against the snake math; traded picks land on the right roster; a dropped poll self-heals on the next one. Reads your league's real lineup (two flex? superflex?) from Sleeper.

On the clock the board leads with a verdict: the man, his price, his odds, the tier he's the last of. Then the field, ranked by what the pick is worth.

pick6 board

For in-person drafts, or any platform without an API. / finds a player, x marks him gone, u takes it back. Everything downstream is identical, because the engine never asks where a pick came from.

pick6 board -slot 7 -teams 10 -lineup "qb rb rb wr wr te flex flex k def" -bench 6

pick6 * -sport fpl

Every command above takes -sport fpl and points at a Fantasy Premier League draft instead. See fantasy premier league below.

pick6 mock

Replays a scripted draft against the real UI with real players. Only the pick sequence is synthetic. -seed N replays the same draft every time; space steps, a autoplays.

Off the clock each ranked row is the position's depth: its best three by value, each with his odds of still being there when you pick. Under the ranking the pane fills with the board behind it, showing what the room is about to do (expect 10 wr · 7 rb · 4 qb, plus the men likeliest gone before you pick) and every position's tiers as a row of dots with the taken ones filled in.

pick6 tiers

pick6 tiers -pos wr -depth 15

The paper cheat sheet, but the tool crosses names off by itself: every man at the position, grouped by tier, straight from the cache. No live draft, no -slot, nothing but -pos and -depth to slice it.

the data tab

tab flips any board to the numbers, in views you switch with ←/→: value is every available player and every number the engine holds (value, tier, adp, spread, survival, format gap) with your rankings file's opinions riding the names; adp is the same table by price; tiers is the ladder, every man at every position grouped by tier, the taken struck through; and every rankings csv in ~/.config/pick6/rankings/ (or -rankings <dir>) is a view of its own, shown exactly as the file ranks them: its order, its tiers, its opinions, with nothing added but each man's odds of reaching your pick and a strike once he is gone. The file you gave fetch -rankings is always there. A name the board does not know gets a ?, not dropped. A big board — one overall order with the positions mixed — prints flat in the file's order with the file's own numbers, and p filters it without renumbering.

Late in the draft, kickers and defenses light up exactly when they should and not a round before:

the notes tab

tab again is your own notes, next to the board, so draft night is two screens (this and sleeper) and not three. It is a folder of markdown files in ~/.config/pick6/notes/, or -notes <dir>, rendered with player names in their position colours and struck through as they get taken, beside a map of the draft so far. global.md sits on top of everything; slot-N.md opens itself when you draw seat N; anything else you flip between with ←/→. e opens the current file in $EDITOR. Nothing on it feeds the engine. The notes are your side of the argument, and the board is the other.

An example folder lives in docs/notes-example.

for the curious

pick6 scout profiles each manager's tendencies from your league's cached drafts; pick6 calibrate backtests both survival models against real completed drafts and grades every model choice. The simulation ships because it won that backtest on every draft that could test it fairly. pick6 regret does the same job for decisions, which a survival backtest cannot grade: it replays your own completed drafts with your seat played by each policy in turn (what you really did, best-available, the formula, each of the two scorers) and prints the team each one walked out with. None of the three is needed to draft.

keys

key	does
/	search players; taken ones stay in the results with the pick that took them
x / u	mark taken / undo (board mode)
tab	cycle board → data table → notes
j k / p / ← →	scroll / position filter / switch view (data tab)
← → / e	switch file / open it in $EDITOR (notes tab)
space / a	step / autoplay (mock mode)
q	quit

fantasy premier league (beta)

FPL Draft is a snake draft over a 15-man squad, so it runs on the same engine.

pick6 fetch -sport fpl
pick6 live -sport fpl <league_id> -slot 3

Use your LEAGUE id, the number in draft.premierleague.com/…/league/<id>. Not the draft id your league details mention: that returns a real feed for someone else's league with the same number, silently. pick6 refuses a league that has already drafted and names it.

FPL publishes the draft order nowhere until round one happens, so pass -slot N once you see your seat. Add -user "your name" and the two get cross-checked as soon as you pick.

pick6 board -sport fpl and pick6 tiers -sport fpl work the same way.

what's different

	nfl	fpl
positions	qb rb wr te k def	gkp def mid fwd
squad	9 starters + 6 bench	15 men: 2 gkp, 5 def, 5 mid, 3 fwd
lineup	fixed, with a flex	11 of the 15, min 3 def / 2 mid / 1 fwd
price	average draft position	FPL's draft_rank
auth	none	none

You draft 15 and start 11, so 4 sit. Your second keeper is one of them: FPL starts exactly one, so he is a bench body and the board weights him like one. Everything past the guaranteed starters competes for the 4 flexible XI places, which is what a flex slot already means.

The squad caps are a legality rule. A sixth defender isn't a bad pick, it's one the app refuses, so the board reads def full and the simulated opponents won't take one either.

Both shapes come from FPL's own settings at runtime, so a rule change moves the board with it.

Notes and rankings live in ~/.config/pick6/notes/fpl/ and ~/.config/pick6/rankings/fpl/, so a football seat file never lands on a football board. Rankings CSVs work exactly as they do for the NFL board (see below); the only change is that names join exactly on name + position + team, since FPL names are one-word surnames and three men are called Wilson.

what beta means here

Tested: 105 real picks from a completed draft replay across all seven seats with zero desyncs, and the whole engine under a squad.

Not calibrated: the opponent model. NFL survival has a referee (pick6 calibrate) and NFL decisions have another (pick6 regret). FPL has neither yet. Measured against eight completed public drafts, the simulated opponents draft by rank order while a real room drafts by squad need, so they take too few keepers and defenders. A green safe tag on a defender is right about 60% of the time against 81 to 85% elsewhere.

At the table: trust the tiers, the value order, the roster pane and who's gone. Discount the survival odds on GKP and DEF, and take those a little earlier than the board says. Nothing measured from NFL drafts is used here; the room curve, escape rates and demand table are all switched off, and the frame says so.

bring your own rankings

A CSV turns the board into your board. Its tiers drive the cliff alarms, its opinions ride the names. Column order doesn't matter, unknown columns are ignored, FantasyPros exports load as-is:

name,position,team,tier,points,sentiment
Jahmyr Gibbs,RB,,1,,target
De'Von Achane,RB,,4,,avoid
Philadelphia Eagles,DEF,PHI,,,target

• tier: your tier lines, kept exactly; oversized tiers are subdivided, never merged.
• points: projected points; overrides the fetched values.
• sentiment: target / pass / avoid. Display-only: an avoid badge rides the player's name, and a verdict crowning a man you flagged says so. Rows with only a name, position and sentiment are valid; that's how kicker and defense calls work.

the math

Everything on screen is a probability with a paper trail: an opponent-aware Monte Carlo (rosters, room appetites, and a measured escape rate for picks that leave the ranked board), survival curves per player, an exactly-N correction, expected-best-later urgency, and probabilistic tier holds. Every constant was chosen by backtesting against real completed drafts. The whole derivation, the calibration methodology, and the ideas that measured worse and were removed (including the leak that briefly flattered the simulation) live in the engine paper (also attached to every release).

data sources

• ADP from Fantasy Football Calculator, whose free ADP API made this possible.
• Player values and ID crosswalk from FantasyCalc.
• Player pool and live draft feed from Sleeper.

license

MIT. See LICENSE.