Models24
PLAYER PROJECTIONS
Season projection LIVE
Rest-of-season points per game and games played for every QB, RB, WR and TE.
Shows in: the draft board, the bye plan's later-week covers, OURS in the trade finder, tiers
Reads: three prior seasons' league points per game, games-weighted 0.5 / 0.3 / 0.2, this season's weeks so far, blended in with SEASON_TO_DATE_K games of prior, position availability base rates, age tables, the rookie draft-capital prior, team volume from prior seasons with a pass-rate-over-expected adjustment
| active method | B |
| DEFAULT_SEASON_WEIGHTS | 0.5, 0.3, 0.2 |
| SEASON_TO_DATE_K | 4.0 |
| AVAILABILITY_BASE | QB 0.88, RB 0.8, WR 0.85, TE 0.84 |
| VOLUME_SHRINK_K_DEFAULT | 10.0 |
| PROE_ADJ_WEIGHT | 0.5 |
2026-08-15docs/backtest/2026-08-15-bakeoff.md
Five candidates, four holdouts: the naive last-season model A won the season bake-off and was promoted; the designed opportunity-share models did not beat it.
2026-10-13docs/backtest/2026-10-13-inseason-ros.md
Model B with this season's points blended in (k = 4 games) beat the preseason-only fit on rest-of-season points and became the active season method.
2026-10-13docs/backtest/2026-10-13-fit-winner-d-tiers.md
Method D's tiers were being walked on re-simulated default draws; found read-only, fixed in main the same week.
Last ran: 2026-11-03 06:00
Weekly projection LIVE
This week's points and p10-p90 range for every QB, RB, WR and TE: usage blended from preseason, season to date and the last four games, simulated, then scaled by matchup, Vegas total and injury status.
Shows in: the sim's own-team expectation, the report card, the brief, the bye plan's this-week cover, the lineup only where the market has no number
Reads: usage: targets and carries, preseason / season to date / EWMA over the last four games, team volume from prior seasons plus the coaching prior for a new play-caller, opponent: points allowed over expectation at the position, blended 50/50 with pass or rush EPA allowed, the team's Vegas implied total, z-scored against three prior seasons, Sleeper status: out = 0, Questionable 0.95, Doubtful 0.75
The lineup the page sets runs on the market's half-PPR number for QB/RB/WR/TE and K1/D1 for K/DEF; this model is the market's challenger, graded against it every Tuesday. Since 2026-10-28 team volume blends this season's own attempts into the prior-seasons fit (TEAM_VOLUME_K games of prior), so a defense that gets off the field faster reaches its offense's attempts as the games show it. Nothing moves a missing teammate's work to anyone: the 2026-10-28 vacated-work study redistributed an out player's share five ways and every arm raised held-out MAE, so an out player's share stays in the sim's untracked bucket and his teammates catch up only through the last-four-games blend.
| active method | D |
| usage weights (early / mid / late) | preseason 0.70 / 0.35 / 0.10, season to date 0.20 / 0.35 / 0.45, last four 0.10 / 0.30 / 0.45 |
| EWMA_LAMBDA | 0.75 |
| EWMA_WINDOW | 4 |
| MATCHUP_BETA | WR 0.1, RB 0.08, TE 0.12, QB 0.1 |
| FUNNEL_BLEND_WEIGHT | 0.5 |
| IMPLIED_BETA | QB 0.05, RB 0.11, WR 0.08, TE 0.03 |
| STATUS_MULTIPLIER | Questionable 0.95, Doubtful 0.75 |
| MULT_CLIP | 0.75, 1.3 |
| COACHING_PRIOR_MAX_WEIGHT | 0.4 |
| TEAM_VOLUME_K | 16.0 |
| LINE_BETA | QB 0.0, RB 0.0, WR 0.0, TE 0.0 |
2026-08-15docs/backtest/2026-08-15-bakeoff.md
Method D (C with grid-searched hyperparameters) won the weekly bake-off, 3 of 4 holdouts, and was promoted.
2026-10-07docs/backtest/2026-10-07-implied-total.md
The implied-total multiplier cut pooled weekly MAE 5.279 to 5.248 and lifted pairwise start/sit half a point; IMPLIED_BETA shipped.
2026-10-10docs/backtest/2026-10-10-def-oe.md
Points allowed over each opponent's expectation replaced raw points allowed as the matchup input: pooled MAE 5.234 to 5.230, better in 4 of 4 holdouts at WR.
2026-10-13docs/backtest/2026-10-13-matchup-beta.md
Walk-forward refit of MATCHUP_BETA met the 0.010 MAE margin but not the 2 SE condition; nothing shipped, the betas stay (QB/WR look oversized, RB/TE right).
2026-10-13docs/backtest/2026-10-13-weekly-weights.md
Grid test of the hand-set constants with a 2024-2025 holdout: no QB/RB/WR/TE constant moved; the funnel weight's +0.003 pairwise gain was the seed, not the model.
2026-10-13docs/backtest/2026-10-13-practice-status.md
A practice-report multiplier (Questionable x DNP/LP/FP) shipped nothing: the DNP cell's gain was 1.6 SE on 17 held-out rows.
2026-10-13docs/backtest/2026-10-13-winprob-lineup.md
A win-probability lineup differed from the mean-optimal lineup in 1 of 64 game weeks; it does not ship.
2026-10-28docs/backtest/2026-10-28-team-volume.md
PASS: blending this season's attempts into team volume (k = 16) cut held-out MAE 0.021 at 2.9 SE with start/sit accuracy unchanged and coverage holding; the gain grows with games played and survives removing the Vegas total. TEAM_VOLUME_K shipped.
2026-10-28docs/backtest/2026-10-28-receiver-corps-diag.md
Closed: which receivers played, and how deep they run, explained none of a quarterback's held-out week-to-week variance beyond his baseline and the game's expected score (incremental R² -0.0003 against 0.01).
2026-10-28docs/backtest/2026-10-28-vacated-work.md
FAIL: moving an out player's targets and carries to his teammates (pro rata, by league position pairs, by the team's own split, by the player pair, each shrunk toward the level above) lost on the fit years to doing nothing, and every arm raised held-out MAE (+0.03 to +0.10) while lifting start/sit accuracy by at most 0.005. By position the backfield improved (RB -0.07 at 2 SE) and WR got worse (+0.15): the mechanism fired on nearly every team-week, mostly on season-long reserve-list players and depth scratches whose stale shares it treated as real. Nothing shipped.
Last ran: 2026-11-03 06:00
Kicker K1 and defense D1 LIVE
This week's points for every kicker and defense from closed-form rates, not a simulation.
Shows in: the lineup's K and DEF slots, DEF THIS WEEK, the bye plan's K and DEF adds
Reads: K1: attempt distribution by distance, make rates shrunk toward league rates, wind, D1: sacks and takeaways over three seasons shrunk hard, the opponent's pressure rate
| K1_SD_FRAC | 0.55 |
| FG_TD_SHRINK_K | 12.0 |
| ATTEMPT_DIST_SHRINK_K | 30.0 |
| WIND_THRESHOLD_MPH | 10.0 |
| WIND_COEF | 0.015 |
| D1 SACK_SHRINK_K | 50.0 |
| D1 TO_SHRINK_K | 250.0 |
| D1 PRESSURE_ADJ_CLIP | 0.85, 1.15 |
2026-08-15docs/backtest/2026-08-15-bakeoff.md
Trailing-average K and DST baselines; no bake-off candidate beat them.
2026-10-13docs/backtest/2026-10-13-weekly-weights.md
K1's band was set by p10-p90 coverage alone; its point projection ranks kickers barely better than a coin flip (pairwise 0.54).
Last ran: 2026-11-03 06:00
Availability model OVERRIDDEN
How many games a player misses, from his own missed-week history shrunk toward his position, with multi-week spells.
Shows in: the season projection's variance (variance-only mode); P(miss 3+) as information
Reads: missed-week rate over the prior three seasons, position priors fit in-window, a two-state Markov chain for spells
| enabled | True |
| mode | variance_only |
| decision | shipped_post_hoc_variance_only |
| HISTORY_WEIGHTS | 0.5, 0.3, 0.2 |
| K_FLOOR | 20.0 |
| R_MAX | 0.9 |
2026-09-10docs/backtest/2026-09-10-availability.md
G1 passed 4 of 4 (it knows who misses games); G2 failed: season MAE worsened 2.51% against a 2% cap, so the means did not ship. Richard approved the variance-only use post hoc on 2026-09-12.
Last ran: 2026-09-10 18:41
Line state and line multiplier NOT SHIPPED
How healthy and settled each offensive line is this week; whether that is worth anything to its quarterback and backs was tested and is not.
Shows in: the lineup page's line table; LINE_BETA sits in the weekly projection at zero
Reads: lineman snap counts: the season's five anchor linemen, their snap share and how many are out, pressure rate allowed and rush EPA per carry, eight-game rolling z
Design docs/plans/2026-10-28-team-context-design.md. Shipped 2026-10-28: a traded player carries Sleeper's team the same day, and the line-state columns. Both gated studies failed the same day: a lineman's on/off record forecasts nothing about his next line (T1), and the line's rolling pressure and rush state adds nothing to player points once the Vegas total is in (T2). A real-world trade like Nelson to Seattle reaches a Seahawk's projection only through the Vegas total and the market blend.
| LINE_BETA | QB 0.0, RB 0.0, WR 0.0, TE 0.0 |
| LINE_Z_COL | QB prs, RB rush, WR prs, TE prs |
2026-10-28docs/backtest/2026-10-28-line-change.md
T1 FAIL: over 2,802 line-change events 2022-2025 the on/off-adjusted forecast of the next four games' pressure rate and rush EPA was no better than the rolling mean (relative gain -0.4% and -1.4%, sign agreement 55%).
2026-10-28docs/backtest/2026-10-28-line-multiplier.md
T2 FAIL: a line multiplier fit walk-forward moved pooled QB+RB start/sit accuracy by +0.0002 against the +0.010 required and worsened MAE at every position; LINE_BETA stays at zero.
Last ran: week 8 of 2026
SEASON OUTLOOK
Rest-of-season value curve LIVE
What a player is worth for the rest of the season, in points per game, from his FantasyPros rest-of-season rank.
Shows in: the trade scan, THEIRS fallback in the finder, the claims drop rule, free-agent floors
Reads: FantasyPros rest-of-season positional ranks (latest scrape), the rank-to-points curve fit on 2022-2025 in the playoff backtest
| RANK_MAX | QB 40, RB 80, WR 100, TE 40, K 32, DEF 32 |
| FLAT_TOP | 12 |
| curve fit seasons (first, last) | 2020, 2025 |
2026-10-26docs/backtest/2026-10-26-blended-ros-value.md
Blending season-to-date points into the curve value (k 24-48) made the paired Brier worse; the consensus curve stays, the blend did not ship.
Last ran: 2026-11-03 06:00
Playoff odds sim OVERRIDDEN
Each team's chance to win this week, make the playoffs and win the title, from expected lineups plus a weekly and a season noise term.
Shows in: the Playoffs tab, the win column, the trade scan's gate
Reads: Yahoo standings and schedule, the market's weekly numbers for this week, rest-of-season ranks through the curve for later weeks, free-agent floors
Nothing in it is fit to results: it prices rank through the curve and adds noise fit to the league's own scores. On 2026-10-27 Richard chose the more accurate re-run calibration and turned the win column and the trade scan on by dated override; the 30-game SD(z) monitor only opens an issue.
| sigma (weekly noise) | 20.028734510917552 |
| tau (season noise) | 6.415029256937795 |
| kappa (projection shrink) | 0.91 |
| ship | win True, title False, playoff True |
| override | 2026-10-27: {'by': 'Richard', 'why': 'act on the best information now: the live monitor reads SD(z) 1.03 on 15 games under this calibration (1.34 under September); the 30-game monitor keeps running and rules at 30', 'date': '2026-10-27', 'what': 'ship.win true under the 2026-10-26 refit (backtest G1 slope 1.299 vs band 1.25)'} |
| variant | consensus |
2026-09-28docs/backtest/2026-09-28-playoff-sim.md
Pre-registered walk-forward on the league's own seasons: the playoff gate passed, a weekly slope gate failed and the one pre-declared remediation (kappa) ran.
2026-10-26docs/backtest/2026-10-26-blended-ros-value.md
Re-run with consensus inputs: sigma 20.03, tau 6.42, kappa 0.91 shipped; the win gate's slope was 1.25 at the band's edge in 2024.
Last ran: 2026-11-04 09:09
Standings and luck LIVE
Where each team stands, its all-play record, expected wins and how much of its record is schedule.
Shows in: the Standings tab
Reads: Yahoo standings and every week's scores
2026-10-05docs/research/league-history/README.md
Over 2021-2025 Richard's record was mostly schedule: 29-46 against about 34 expected wins; opponents scored 4.3 above their averages against him.
Last ran: 2026-11-04 09:09
ROSTER MOVES
Lineup LIVE
Who starts, by the number each slot is set on, with kickoff locks recorded.
Shows in: the Today and Lineup tabs, the brief's LINEUP line, the texted report card's grade
Reads: the FantasyPros weekly consensus (market_week), projection_week, player_status
Sources: the market's half-PPR consensus for QB/RB/WR/TE when a pre-kickoff row exists, our weekly model otherwise, K1/D1 for K and DEF, 0 for a player who is out. The win-probability lineup was tested and does not ship.
| WIN_PROB_MIN_GAIN | 0.015 |
| WIN_PROB_MAX_COST | 3.0 |
| DEFAULT_CV | QB 0.45, RB 0.85, WR 1.0, TE 1.15, K 0.35, DEF 0.4 |
2026-10-13docs/backtest/2026-10-13-winprob-lineup.md
The rule fired 0 times against the 5 the ship rule needed over 64 game weeks; does not ship.
2026-10-05docs/research/league-history/README.md
Richard's own start/sit calls beat 'follow Yahoo' by 0.4 a week over 2022-2025 while the league lost about 1.0.
Last ran: 2026-11-03 06:00
Waiver wire LIVE
Which free agents to add now, watch or skip, from usage before the box score: role jumps, a teammate out, quiet producers, matchup, consensus, red flags, a production floor and a roster fit.
Shows in: the Wire tab, the brief's CLAIMS
Reads: route, carry and snap shares and xfp per game (nflverse ff_opportunity), player_status and Yahoo out tags, FantasyPros rest-of-season rank and Yahoo % rostered, Richard's roster and need weeks
First live grading (2026-10-27, 1,507 rows, one season): ADD NOW hit the replacement line 6% of the time against WATCH 20% and SKIP 22%; by screen, consensus 39%, quiet producer 27%, red flag 21%, matchup 15%, role jump 3%, teammate out 2%. A strict yardstick (a missed game counts zero, WR26 is the line), but the tiers need re-testing with opportunity and the crowd together.
| ROLE_JUMP_DELTA | 0.25 |
| ROLE_FIRST_ABOVE | 0.6 |
| QUIET_GAP | 3.0 |
| FLOOR_XFP / FLOOR_SHARE | 6.0, 0.12 |
| OBVIOUS_PCT | 50.0 |
| BREAKOUT_POINTS | 15.0 |
| REPLACEMENT_RANK | QB 11, RB 26, WR 26, TE 11, K 11, DEF 11 |
| ROSTER_FIT_MARGIN | 1.0 |
| OV_K | 4.0 |
| OPPORTUNITY_AMENDMENT | 2026-10-28 |
2026-10-14docs/backtest/2026-10-14-wire-kickoff-windows.md
The contingent next-man rule passed both gates (10.12 points vs 10.0; +5.25 over other backups), the first by a hair; ships as names and windows.
2026-10-27docs/backtest/2026-10-27-opportunity-claims.md
Ranking claims by opportunity value beat the old rule by +2.9 points per pick per game (90% interval +1.5 to +4.5, both seasons); the opportunity-based roster fit failed G2 (19%) and G3 (11% vs 14%) and stays on Yahoo.
Last ran: 2026-11-04 09:09
Claims and drops LIVE
Which ADD NOW rows to claim, in what order, and which bench player each claim can afford to drop.
Shows in: the brief's CLAIMS block (texted), the Today tab's claims card
Reads: wire_rec, the rest-of-season curve and free-agent floors, yahoo_pending
Claims are the wire's ADD NOW rows ordered by opportunity value; a pending Yahoo claim is not re-offered. A drop must be worth less than the add on the sim's curve over the free-agent floor, never a starter (fixed 2026-10-27 after 'DROP Nabers' went out by text). The crowd flag here is the old rule; docs/plans/2026-10-28-league-targets-design.md replaces it with who else is reaching for the player.
| MAX_CLAIMS | 5 |
| DECISION_GAP | 2.0 |
| CONTESTED_ROS_RANK | 40 |
| CONTESTED_PCT | 50.0 |
2026-10-27docs/backtest/2026-10-27-opportunity-claims.md
G1 passed: opportunity-ordered claims beat the this-week-edge rule by +2.9 points per pick per game on the league's own free pool, 2022-2025.
Last ran: runs with every refresh and league sync; the brief is texted, not stored
Bye plan and quiet adds LIVE
Who covers each starter's bye week, and which free agent to add for it, ranked on this season's usage, trend, matchup and hurt teammates.
Shows in: the Bye plan tab, the brief's COVERAGE lines
Reads: xfp and usage shares over the last two games, the opponent's points allowed over expectation, player_status and Yahoo out tags
Judgment constants, never backtested: 10 points per full point of share trend, a hurt-teammate bonus capped at 2 that needs the teammate to have played in the last two scored weeks (Richard's 2026-10-13 call), matchup decay left unshrunk.
| TREND_WEIGHT | 10.0 |
| HURT_CAP | 2 |
| ROLE_SNAP_MIN | 0.25 |
| QUIET_SNAP_MIN | 0.7 |
| BYE_PLAN_MAX_SHOW | 3 |
Last ran: rebuilt with every team page
DEF this week LIVE
Whether to start the held defense or stream the best free one this week.
Shows in: the Today tab's DEF card, the brief's DEF line
Reads: FantasyPros weekly DST consensus before kickoff, D1, the opponent's implied total
| gate per season | 1.0 |
| gate pooled | 1.5 |
| result pooled pts/week | 2.58 |
| result by season | 2022 1.37, 2023 3.43, 2024 4.03, 2025 1.61 |
2026-09-29docs/backtest/2026-09-29-dst-streaming.md
PASS: streaming gained +2.58 points a week pooled and at least +1.0 in all four seasons against the +1.0 / +1.5 gates set before the run.
Last ran: 2026-09-29T22:00
Trade scan LIVE
Swaps that raise Richard's expected wins and leave the partner no worse on the market's own values, through the sim's win math.
Shows in: the Trades tab (TRADE IDEAS), the brief's TRADES line
Reads: the rest-of-season curve, the playoff sim's schedule and noise, Yahoo rosters
Gated on the sim's win flag, which is on by Richard's 2026-10-27 override. Values are FantasyPros rest-of-season ranks through the curve; a top-3 QB/RB/TE carries a market premium charge.
| MIN_MY_DEW | 0.1 |
| MIN_PARTNER_DEW | 0.0 |
| FAIRNESS | 0.9 |
| TOP3_PENALTY | 1.5 |
| KEEP | 8 |
| SHAPES | 1, 1, 2, 1, 1, 2, 2, 2 |
2026-09-28docs/backtest/2026-09-28-playoff-sim.md
The win math is the playoff sim's, validated there; the valuation itself is the market's and is not backtested.
Last ran: 2026-11-04 09:10
Trade finder LIVE
Trades Richard can sell: what the partner sees (Yahoo's projection), what we see (the season model), every team's open and weak slots, and how pressed each manager is.
Shows in: the Trades tab (SELL list and needs map), the pitch text
Reads: Yahoo rest-of-season projection per scheduled game (THEIRS), the season model's points per scheduled game (OURS), yahoo_transaction (manager activity), standings, points against, playoff odds, tripwires
GAP is a hypothesis, not a verdict: `--grade` tests OURS against THEIRS on what players then did. The churn term has a forward gate: the top three by pressed score must make 45% of the league's next-week moves over weeks 6-15 or its weight goes to 0.
| WINDOW_WEEKS | 3 |
| WEAK_GAP | 3.0 |
| MIN_RATIO | 1.0 |
| NEED_MIN_GAIN | 1.0 |
| RECORD_PRESSED | 3 |
| WEIGHTS | record 1.0, pa 1.0, odds 2.0, open_now 2.0, weak_now 1.0, open_soon 0.5, tripwire 2.0, churn 1.0 |
| CHURN_PRIOR_WEEKS | 2 |
2026-10-14docs/backtest/2026-10-14-desperation-map.md
Retrospective over weeks 3-5: churn alone picked 30% of next-week movers, exactly uniform; the forward gate over weeks 6-15 decides.
Last ran: 2026-11-04 09:10
EVALUATION
Weekly report card LIVE
Did the model beat the market on start/sit this week, and did the lineup start the right players?
Shows in: the Tuesday text, the Grades tab
Reads: frozen pre-kickoff projection_week rows, the last market_week snapshot before each kickoff, league-scored actuals
| PAIRWISE_TOP_N | QB 12, RB 24, WR 30, TE 12 |
Last ran: 2026-11-03 06:00
Lineup grades, moves ledger, luck LIVE
Every manager's lineup graded against his own best lineup from Yahoo's captured lineups, every add and drop's return, and schedule luck.
Shows in: the Grades, Moves and Standings tabs
Reads: yahoo_lineup (captured each sync), yahoo_transaction, league-scored actuals
2026-10-26docs/plans/2026-10-26-evaluation-blocks.md
Descriptive blocks; Yahoo-final gating and a per-player from-week rule; no model to validate.
Last ran: 2026-11-04 09:09
Wire and finder grading LIVE
What the wire's flagged players and the finder's pitched players then did, against the replacement line and against both valuations.
Shows in: `ff wire --grade`, `ff trades --finder --grade`, the design docs' result sections
Reads: wire_rec, trade_finder_idea, league-scored actuals, Yahoo % rostered two captured weeks later
| REPLACEMENT_RANK | QB 11, RB 26, WR 26, TE 11, K 11, DEF 11 |
| REPLACEMENT_WINDOW | 2 |
Last ran: 2026-10-27 17:23
DATA
Yahoo league capture LIVE
The league as Yahoo shows it: rosters, standings, scores, projections and % rostered, transactions, lineups, Richard's pending claims.
Shows in: every tab that names a leaguemate; Wednesday 9am and on demand
Reads: football.fantasysports.yahoo.com league 875177
Read-only browser capture in Richard's signed-in Chrome. Gaps: % rostered weekly since 2026-10-14 only, adds-last-week never fills, manager names are not read, trades under league vote are not seen until they process.
| payload | v3 |
| runbook | docs/runbooks/league-sync.md |
Last ran: 2026-11-04 09:09
Sleeper status feed LIVE
Who is out, questionable or doubtful, and since 2026-10-28 which team a player is on, the same day a trade happens.
Shows in: every projection and lineup; the news texts (RB1 out, trade tripwires)
Reads: Sleeper players API, manual_status_override
| out statuses | Injured Reserve, Non Football Injury, Practice Squad, Suspended, Out |
| scaled | Questionable 0.95, Doubtful 0.75 |
Last ran: 2026-11-03 06:00
nflverse, market and context ingest LIVE
Play by play, player weeks, expected points from usage, injuries, rosters, snap counts, FantasyPros consensus, odds and weather, refreshed every Tuesday and game day.
Shows in: every model above
Reads: nflverse releases, FantasyPros, ESPN odds, Open-Meteo
Depth charts have been skipped since August (the source changed shape). The practice report (DNP/LP/FP) is captured and shown on the page; nothing prices it.
| metrics seasons | 2018 to current |
Last ran: newest nflverse:market_week 2026-11-03 06:00, oldest nflverse:players 2026-08-15
IN DESIGN
Other managers' targets NOT SHIPPED
Who else will take each free agent this cycle, so a claim could be called contested or a free agent going fast, with a pivot.
Shows in: nothing yet: the model failed its gates; the manager names and waiver priorities it needed are live on the Trades and Wire tabs' data
Reads: trade_finder_run needs and activity, wire_rec, the league history's pickup log, yahoo_team.manager_nickname, yahoo_standings.waiver_priority
docs/plans/2026-10-28-league-targets-design.md. Built and replayed on the 2022-2025 pickup log the same day; killed by its own rule. What stays: Yahoo manager names and waiver priorities captured each sync, and the finder's activity rate now counts offensive adds with each manager's own history as prior.
| NEED_SCALE | 0.25 |
| CONTEST_LINE | 0.5 |
| VIS_POOL | 15 |
2026-10-28docs/backtest/2026-10-28-league-targets.md
G1, G2, G3 all FAIL, kill rule fired: last week's points alone predicted the league's adds better than the model (precision 0.30 vs 0.22), a manager's add rate alone matched the model on who adds (top-3 0.485), and the contest line was unreachable (p_lost peaked at 0.26). The usage term was the part that hurt: the league adds the box score.
Last ran: unknown
A missing teammate's work IN DESIGN
When a role player is out, who inherits his usage and how much, by the team's own past splits.
Shows in: the weekly projection
Reads: snap counts and usage shares, player_status
Sequenced after the team-context build (docs/plans/2026-10-28-team-context-design.md §2d); its own design and backtest, margins pre-registered above the +/-0.003 pairwise noise floor.
Last ran: unknown
DRAFT (DORMANT IN SEASON)
Draft board, draft sim and sleeper score DORMANT
Preseason: value over replacement by tier, a walk-forward draft-policy simulator, and a sleeper flag.
Shows in: the Draft kit
Reads: the season projection, ADP and consensus boards
2026-10-05docs/backtest/2026-10-05-draftsim.md
VOR_TIER won the policy simulator by all-play win rate, 0.508 against the 0.500 chalk control; ZERO_RB lost at 0.462.
2026-08-15docs/backtest/2026-08-15-sleepers.md
The sleeper score did not beat random same-price players (11 of 60 against 25% expected, ratio 0.73); it ships labelled EXPERIMENTAL and is not a signal to draft on.
Last ran: 2026-10-05 14:09