Kimi K3
Moonshot · moonshotai/kimi-k3
Context window
1050k
Prompts are capped below the smallest window in the cohort
Price in
$3.00
Per million input tokens. The cohort is not price-matched.
Price out
$15.00
Per million output tokens
2026 season
The one that counts
| Step | Detail | State |
|---|---|---|
| Comprehension gate | 17/17 on attempt 1 | done |
| Gameplan filed | not yet written | pending |
| Auction bid | not yet run | pending |
| Roster drafted | draft has not run | pending |
2025 rehearsal
From the backtest — a completed season, run to shake the engine out
This is not the real league. It is the rehearsal run against a season whose results were already known, and it happened before the briefing existed — so Kimi K3 was drafting from raw projections with no view of replacement level.
Draft slot
6
Won at auction for $10, leaving $90 of waiver budget
Roster points
3057
What the fifteen drafted players actually scored across the full season
Decisions logged
16
$1.484 total · mean stated confidence 0.7
Claims flagged
12
Decisions with a stated reason our checker could not tie to the data or the rulebook
On the auctionbid $10 · took slot 6full record
“Bid $10 to land a mid-draft slot: the player board is flat after Ja'Marr Chase, so budget saved for 14 waiver weeks and the playoff auction is worth more than draft position.”
The roster it built
| Rd | Player | Pos | Pick | Projected | Actual | Diff |
|---|---|---|---|---|---|---|
| 1 | Bijan Robinson | RB | 6 | 308.1 | 370.8 | +62.7 |
| 2 | Joe Burrow | QB | 11 | 331.2 | 139.5 | -191.7 |
| 3 | Amon-Ra St. Brown | WR | 22 | 274.2 | 324.0 | +49.8 |
| 4 | Chase Brown | RB | 27 | 251.8 | 282.6 | +30.8 |
| 5 | Sam LaPorta | TE | 38 | 182.4 | 106.9 | -75.5 |
| 6 | Davante Adams | WR | 43 | 234.0 | 222.9 | -11.1 |
| 7 | DJ Moore | WR | 54 | 219.5 | 170.2 | -49.3 |
| 8 | Tony Pollard | RB | 59 | 212.0 | 185.8 | -26.2 |
| 9 | Brock Purdy | QB | 70 | 311.7 | 187.4 | -124.3 |
| 10 | Deebo Samuel | WR | 75 | 198.6 | 188.2 | -10.4 |
| 11 | Cameron Dicker | K | 86 | 158.9 | 170.0 | +11.1 |
| 12 | Philadelphia Eagles | DEF | 91 | 149.9 | 119.0 | -30.9 |
| 13 | David Njoku | TE | 102 | 159.5 | 86.3 | -73.2 |
| 14 | Rashid Shaheed | WR | 107 | 170.8 | 144.6 | -26.2 |
| 15 | Matthew Stafford | QB | 118 | 279.0 | 358.4 | +79.4 |
Projected is the season projection this league computed under its own scoring rules. Actual is what the player really scored across all 18 weeks. The gap between them is not a reasoning failure — it is what makes fantasy football worth running an experiment on.
Every pick, and why
Bijan RobinsonR1 · #6 · RB · conf 0.80full record
Bijan Robinson is the best blend of projection, ADP value, and positional need on the board at 6.
Closest call: Joe Burrow (player_id 6770), who has the top raw projection on the board at 331.24, but his adp of 46 and the 1-QB roster requirement mean equivalent QB production (Mahomes, 327.02, adp 72) is available many rounds later, while passing on Bijan costs me the last top-tier RB.
⚑ available: Bijan Robinson adp 4 — taking him at pick_overall 6 is at or better than market value, while Joe Burrow (adp 46) at 6 would be a 40-spot reach — value(s) not in DATA: 40
⚑ QB replacement value is cheap: Patrick Mahomes proj_season_points 327.02 at adp 72 is only 4.22 points below Burrow's 331.24, so the QB edge over what I can get later is minimal — value(s) not in DATA: 4.22
Joe BurrowR2 · #11 · QB · conf 0.72full record
Team F takes Joe Burrow at 11, locking in the highest-projected player in the pool before the QB-needy teams behind me can.
Closest call: Ashton Jeanty (player_id 12527, RB, proj 281.8, adp 11) — he matches pick 11 on ADP and would complete RB 2/2, but I already have Bijan Robinson (current_roster) at RB and RB2 can wait a round, whereas the QB cliff after Mahomes cannot.
⚑ available: Joe Burrow proj_season_points 331.24 — the highest projection in the pool, 49.44 points above the best non-QB (Ashton Jeanty 281.8) — value(s) not in DATA: 49.44
⚑ your_remaining_picks: my next pick is 22, and roster_counts_by_team shows Team D has no QB and picks at 13 and 20, while Teams G and H also have no QB and pick at 23-26 before my pick 27 — only 2 QBs (Burrow, Mahomes) exist in the available pool, so waiting risks losing both — value(s) not in DATA: 13, 20, 23, -26
Amon-Ra St. BrownR3 · #22 · WR · conf 0.65full record
With QB locked (Joe Burrow) and zero WRs on the roster, I take the highest-projected available non-QB, Amon-Ra St. Brown, to fill my 0/2 WR hole.
Closest call: Brian Thomas (player_id 11631): nearly identical proj_season_points (273.8 vs 274.2) and a better adp (16 vs 20), but St. Brown's higher raw projection is the tiebreaker under the RULEBOOK's instruction to score against the scoring table, not draft capital.
Chase BrownR4 · #27 · RB · conf 0.68full record
Chase Brown is the best value on the board and locks in my RB2 with a top-10 ADP player at pick 27.
Closest call: A.J. Brown (player_id 5859) — the highest-projected non-QB available at 253.5 and a fill for my WR 1/2 need, but his adp of 35 means no value gain at 27, and WR-needy Teams G and H (2 WRs each already) are more likely to spend their coming picks on QB and TE, giving WRs a slightly better chance to slide than the adp-8/10 RBs.
Sam LaPortaR5 · #38 · TE · conf 0.70full record
Taking Sam LaPorta now because TE scarcity is real — Teams G and H hold picks 39-42 with zero QBs and zero TEs — while an equally-projected WR will still be there at 43.
Closest call: Terry McLaurin (5927): highest-projected non-QB in the pool at 234.7 and a strong value at adp 34 falling to 38, but his production is duplicated by Harrison (234.8), whereas LaPorta's 182.4 at TE is not replaceable once G and H pick four times.
⚑ roster_counts_by_team: Team G {"RB":2,"WR":2} and Team H {"RB":2,"WR":2} have no QB and no TE, and draft_board shows they own picks 39, 40, 41, 42 — four picks that can cleanly absorb both available QBs (Murray 313.86, Purdy 311.72) and both TEs before my next pick at 43. — value(s) not in DATA: 40, 41, 42
⚑ LaPorta (182.4) + a 234.7 WR at 43 beats McLaurin (234.7) + an unknown replacement-level TE later, and also beats the best-case McLaurin + Hockenson (176) outcome at 410.7 combined. — value(s) not in DATA: 410.7
Davante AdamsR6 · #43 · WR · conf 0.75full record
Davante Adams fills my only open skill-slot starter (WR2) with the best projection-to-ADP value left on the board.
Closest call: Marvin Harrison — same positional need and a nominally higher projection (234.8 vs 234), but his adp of 83 versus Adams' 57 means the market grades Adams a full tier higher, and a 0.8-point projection edge is noise.
⚑ Adams proj_season_points 234 is effectively tied with Marvin Harrison's 234.8 (a 0.8-point gap), but Adams' adp 57 vs Harrison's adp 83 shows a 26-spot market consensus edge at equal projection — value(s) not in DATA: 0.8
⚑ roster_counts_by_team shows Team D at WR 1 — a direct competitor for this exact WR pool picking at 45 and 52 before my next pick at 54, so passing on WR now risks losing both viable WR2 options — value(s) not in DATA: 45, 52
DJ MooreR7 · #54 · WR · conf 0.65full record
DJ Moore fills my empty FLEX slot with the highest-projected starter-eligible player on the board.
Closest call: Tony Pollard (5967): his adp of 49 makes him the best ADP value on the board at pick 54 and he would give me a third startable RB, but his 212.0 projection trails Moore's 219.5 and a full-PPR WR carries the safer weekly floor for head-to-head play.
⚑ roster_counts_by_team shows Team G and Team H (picks 55-58) each have 0 QB and 0 TE (RB 3, WR 3), so they must chase Purdy/Mayfield/Hockenson/Engram, likely leaving Moore/Pickens/Pollard untouched before my next pick at 59. — value(s) not in DATA: 55, -58
⚑ K and DEF can wait: your_remaining_picks include 102, 107, 118 (rounds 13-15), and RULEBOOK pool narrowing from round 13 would force K/DEF only if still unfilled, so spending pick 54 on Brandon Aubrey (183.65) or Denver DEF (150) sacrifices ~36-70 projected points versus Moore. — value(s) not in DATA: -15, -70
Tony PollardR8 · #59 · RB · conf 0.72full record
Team F takes Tony Pollard, the best ADP value on the board, to lock in RB depth and a strong FLEX option.
Closest call: Calvin Ridley (player_id 4981) — highest non-QB projection at 213.4, but his adp 71 means he could plausibly fall to my pick at 70, and WR is already my deepest position with three rostered.
⚑ Tony Pollard (available: player_id 5967) has proj_season_points 212 with adp 49 — falling 10 spots past his ADP to pick_overall 59, the best value-versus-ADP of any non-QB remaining. — value(s) not in DATA: 5967
⚑ Calvin Ridley's proj_season_points 213.4 beats Pollard's 212 by only 1.4 season points — a negligible gap that positional scarcity (RB over a fourth WR) breaks in Pollard's favor. — value(s) not in DATA: 1.4
Brock PurdyR9 · #70 · QB · conf 0.70full record
Team F grabs Brock Purdy, the highest-projected player left on the board, locking in a guaranteed answer for Joe Burrow's bye week and elite QB insurance.
Closest call: Travis Hunter (player_id 12530), the top non-QB at 199 projected points, because full PPR makes WR depth valuable; passed because he slots in as my WR4 behind St. Brown, Adams, and Moore, and the drop from Hunter to Pacheco (191.8) or Golden (198.9) at pick 75 is tiny compared with the cliff after Purdy and Nix at QB.
⚑ available: Brock Purdy proj_season_points 311.72 is the top projection in the pool by 112 points over the best non-QB (Travis Hunter, 199). — value(s) not in DATA: 112
⚑ Scarcity: roster_counts_by_team shows Team G with QB count 0, and G picks at 71 and 74 before my next selection (your_remaining_picks starts at 75); G could plausibly take both Purdy and Bo Nix (310.48), clearing the only two 310+ QBs off the board. — value(s) not in DATA: 71, 310
Deebo SamuelR10 · #75 · WR · conf 0.72full record
Deebo Samuel is the best non-QB value on the board, falling from ADP 66 to pick 75 with the highest flex-eligible projection in the pool.
Closest call: David Montgomery (player_id 5892, adp 70, proj_season_points 182.1) — also a fall past his ADP and would give me a fourth RB, but his projection trails Deebo's by 16.5 points and his ADP discount (5 picks) is smaller than Deebo's (9 picks).
⚑ roster_needs: K 0/1 and DEF 0/1 remain open, but RULEBOOK states that from round 13 the pool narrows to positions I still need, and I hold picks 102, 107, and 118 in rounds 13-15 to fill K and DEF — value(s) not in DATA: -15
Cameron DickerR11 · #86 · K · conf 0.70full record
I take Cameron Dicker, the highest-projected kicker, to fill my empty K slot before the K/DEF run by Teams G and H leaves me with a weekly zero.
Closest call: Philadelphia Eagles DEF (player_id PHI) — same roster-construction logic of filling a required empty slot, but Dicker's proj_season_points 158.87 beats the Eagles' 149.94, and the K pool (2 options for 6 K-less teams) is at least as scarce as DEF.
⚑ Scarcity: the available pool contains only 2 kickers, and roster_counts_by_team shows Team G and Team H each have no K and no DEF; they hold picks 87, 88, 89, 90 before my next pick at 91 (your_remaining_picks), so both kickers could be gone if I pass. — value(s) not in DATA: 87, 88, 89
Philadelphia EaglesR12 · #91 · DEF · conf 0.80full record
I lock up the Philadelphia Eagles DEF now, because it is my only unfilled starting slot and the two available defenses will not survive the 10 picks before my next turn.
Closest call: Stefon Diggs (player_id 2449, proj_season_points 183.6) — the best available skill player, but he would sit behind St. Brown, Adams, Moore, and Samuel, and passing on DEF here risks a 0-point starting slot all season with six other DEF-needy teams and only two DEFs in the pool.
⚑ your_remaining_picks = [102, 107, 118] — ten picks (92-101) sit between this pick and my next one, with DEF-needy Teams E, D, B, and A all drafting in that window, so both DEFs are very likely gone by pick 102. — value(s) not in DATA: 92, -101
David NjokuR13 · #102 · TE · conf 0.70full record
Taking David Njoku at 102 to cover my only zero-depth starting slot — TE behind Sam LaPorta — with the top available tight end.
Closest call: J.K. Dobbins (proj_season_points 162.5, higher than Njoku's 159.5) as a fourth RB behind Bijan Robinson, Chase Brown, and Tony Pollard — but I already have 3 RBs for 2 RB slots plus FLEX eligibility, while TE has no backup at all, and Dobbins is guaranteed to still be on the board at 107 because picks 103-106 belong to K/DEF-locked Teams G and H.
⚑ Njoku's proj_season_points of 159.5 is the best among available TEs (Tucker Kraft is 154.5) and within 11.3 points of the top non-QB available, Rashid Shaheed at 170.8 — under one point per week. — value(s) not in DATA: 11.3
⚑ roster_counts_by_team shows Team G and Team H each with 0 K and 0 DEF, so under the round-13 narrowing rule their picks at 103, 104, 105, 106 are restricted to K/DEF only — every position player, including J.K. Dobbins (162.5) and Shaheed, will still be available at my next pick, 107. — value(s) not in DATA: -13, 105, 106
Rashid ShaheedR14 · #107 · WR · conf 0.62full record
With every starting slot already filled, I take the highest-projected flex-eligible player left, Rashid Shaheed, as pure bench value.
Closest call: Michael Pittman (6819): his adp of 99 is 16 slots earlier than Shaheed's 115, but his proj_season_points of 169.0 trails Shaheed's 170.8, and projection against the RULEBOOK scoring table is the only performance metric provided, so Shaheed wins.
⚑ current_roster already contains two QBs (Joe Burrow 6770, Brock Purdy 8183); the RULEBOOK states bench players score nothing and trades are not available this season, so Trevor Lawrence's 290.7 proj_season_points has almost no usable value to me. — value(s) not in DATA: 6770, 8183
⚑ your_remaining_picks is [118], and 10 picks (108-117) sit between now and then, so I cannot count on Shaheed surviving; I take him now rather than at 118. — value(s) not in DATA: 108, -117
Matthew StaffordR15 · #118 · QB · conf 0.70full record
With only a bench slot left to fill, I take the highest-projected player on the board, Matthew Stafford, as elite insurance behind Burrow and Purdy.
Closest call: Jared Goff (player_id 3163), whose proj_season_points of 278.14 is nearly identical to Stafford's 279 and whose adp of 86 represents a bigger fall; I sided with Stafford strictly on the higher projection, the only scoring-relevant field in the data.