2025/26 is a reconstructive replay (cutoff = first kickoff − 90 minutes, not a live two-hour freeze). 2026/27 pages are an illustrative sample of the live format. What that means

FPL Replay

Gameweek 3 · 2026/27

Rolling FT Heuristic vs MILP Active Bench Reconstruction

What-if for FPL 2026/27 Gameweek 3: Deploying the Free Transfer on Antonee Robinson (£4.6m) to extinguish zero-minute risk on departed Barco while promoting João Pedro into a 3-4-3 produces a higher expected floor than rolling 2 FTs for the post-international break window. Before kick-off the alternative was +3.7 projected points versus the control (64.1 vs 67.8). Match points are still pending.

Cutoff 31 Aug 2026, 08:00 UTCIllustrative samplesha256:7c92b7a2…f381Download snapshotIllustrative sample

Control vs alternative

PathFormationCaptainChipProjectedvs controlPointsPayoff
Template Heuristic (Roll Transfer) (control)
Preserves 1 FT to hold 2 Free Transfers entering GW4. Leaves Barco (0 projected min) occupying the 3rd bench spot and starts Rogers vs Leicester away.
3-5-2Haaland (2x)None64.1Pending
MILP Solver Execution (Robinson + 3-4-3) (alternative)
Transfers Barco -> Robinson (£4.6m, IPS away) and shifts formation to 3-4-3 promoting in-form João Pedro into the starting XI.
3-4-3Haaland (2x)None67.8+3.7PendingPending

Squad differences

Template Heuristic (Roll Transfer) (control)

Bank £0.2m · 3-5-2

Haaland (£15.2m), Palmer (£10.6m), Saka (£10.1m), Son (£10.0m), Rogers (£5.1m)

Preserves 1 FT to hold 2 Free Transfers entering GW4. Leaves Barco (0 projected min) occupying the 3rd bench spot and starts Rogers vs Leicester away.

MILP Solver Execution (Robinson + 3-4-3) (alternative)

Bank £0.2m · 3-4-3

Haaland (£15.2m), Isak (£8.5m), João Pedro (£5.6m), Palmer (£10.6m), Saka (£10.1m), Robinson (£4.6m)

Transfers Barco -> Robinson (£4.6m, IPS away) and shifts formation to 3-4-3 promoting in-form João Pedro into the starting XI.

What happened

The simulation highlights that the marginal shadow value of saving an extra Free Transfer (+1.92 xP) was insufficient to offset the immediate +4.20 xP security and fixture matchup provided by Robinson, alongside promoting João Pedro (4.60 xP) over Rogers (4.15 xP). The MILP arm captured a net model delta of +3.7 xP while reducing lineup variance from 14.8 to 13.2.

  • Holding deadwood players on the bench degrades effective auto-sub floor equity by 1.84 xP.
  • The 3-4-3 formation flex unlocked João Pedro's favorable Arsenal transition fixture equity while maintaining Rogers as first auto-sub.
  • Realised settlement is live/pending official Premier League match completion.

How to verify

Rolling FT Heuristic vs MILP Active Bench Reconstruction was written to a JSON snapshot at 31 Aug 2026, 08:00 UTC and hashed with SHA-256 (sha256:7c92b7a2d488e3fa0895c32b567b489ef0c239458e0a29367d5668b02ea6f381). Download the snapshot to inspect the frozen inputs. The freeze rule is described in Methods.

Download the JSON snapshot