TrackSmart AI Weekly Scorecard: NYRA & Regional Tracks

Cumulative performance through October 4, 2026, with the September 28 – October 4 week in focus

TrackSmart AI | Prepared for Saratogabets | Week Ending October 4, 2026

1. Overview & Methodology

This scorecard covers the cumulative reporting period of December 26, 2025 through October 4, 2026, with a close look at the week of September 28 through October 4. The tracking dataset spans Belmont at the Big A (BAQ), Finger Lakes (FL), Aqueduct (AQU), Laurel Park (LRL), Santa Anita Park (SA), Penn National (PEN), Saratoga Harness (SARH), Woodbine (WO), Saratoga (SAR), Del Mar (DMR), Hollywood Casino at Charles Town Races (CT), Kentucky Downs (KD) and Keeneland (KEE).

The headline measure is the Top 4 Strike Rate: the share of races in which the AI’s four selections include the winner. A race counts as a hit if the winner sits anywhere in the Top 4, and the Avg Payout column tracks what those winning horses paid. This week added 115 races across Finger Lakes, Belmont Park, Saratoga Harness and Keeneland, bringing the cumulative sample to 2,261 races.

2. This Week at a Glance

Volume Expansion

The dataset grew to 2,261 total races with the addition of 115 races from September 28 through October 4. Finger Lakes, Belmont Park, Saratoga Harness and Keeneland all contributed.

Strike Rate Resilience

The cumulative Top 4 Strike Rate eased slightly, from 78.0% to 77.3% (1,747 of 2,261). The week itself brought high-variance conditions and weather anomalies. Even so, the AI found the winner in its Top 4 in 74 of 115 races (64.3%), with especially strong multi-day hit rates at Keeneland and Finger Lakes.

Value Normalization

The average payout on AI hits rose to roughly $8.07, up from roughly $7.98. Deep value identification and scratch-protocol mapping produced a run of large overlays: Flyhawk El Derecho ($38.16), Sunday’s Brunch ($35.72), Let Em Know ($27.38), Diliello ($27.14), La Puma ($25.84), Dotsy ($25.02) and White Sport Coat ($22.30).

System Integrity & Scratch Protocol

The automated scratch-adjustment protocol was important again this week. It moves “Next-Man-Up” alternates into active slots, which protected win rates and lifted winners such as Funny Factor ($22.08), Damavand ($7.48) and Right in Time ($15.38) into actionable ticket structures.

3. Daily Performance Breakdown

Prior totals reflect cumulative data through September 27. Daily breakdowns from earlier reports are condensed into the Prev. Totals row. Late scratches shifted runners up in the rankings across all cards this week, producing a number of top-tier hits and lifting the average payout.

4. Winner Distribution by Rank (Cumulative)

Taken together, the Top Pick and 2nd Selection account for 1,116 of the 1,747 hits, while the 3rd and 4th slots contribute 631 more. Those back-of-the-ticket slots are also where this week’s biggest prices came from, so the full Top 4 is doing real work and not just padding the ticket.

5. Cumulative Statistics Summary

6. This Week’s Structural Findings

The Scratch Protocol Keeps Paying

Sliding alternates up into the Top 4 after late scratches proved lucrative once again. The rule is simple and structural, and it caught real payout winners. Damavand ($7.48) moved from 5th to 3rd at Keeneland, and Funny Factor ($22.08) and Vekomancer ($3.62) were adjusted up into the Top 4 at Belmont Park.

Turf Routing & Deep Value

The turf algorithms were a highlight. On October 3 at Belmont Park the AI correctly swept every turf event on the card. Across the week it found strong value in stakes and turf routes, identifying late-kick beneficiaries and capturing overlays like Let Em Know ($27.38) and Dotsy ($25.02).

Pace Mapping on the Half-Mile Oval

The model identified vulnerable, contested early fractions well. At Saratoga Harness it located late-kick advantages on a half-mile oval and pulled Sunday’s Brunch ($35.72) and Flyhawk El Derecho ($38.16) out of unsettled pace flows.

Off-Turf Chaos & the Speed Gate Blind Spot

Weather and surface changes hurt the AI’s effectiveness on September 30 at Belmont Park, where the day finished at 12.5%. Races moved to the All-Weather Tapeta broke the model’s predictive footing. Rigid filters such as the “Speed Gate” restriction wrongly faded horses taking large class drops. That exposes a blind spot, and it needs an explicit override condition for stakes and allowance droppers.

7. AI’s Final Analysis: Building the Path Forward

1. Keep the scratch protocol exactly as it is. It has now protected win rates and surfaced overlays across multiple weeks. Any changes should be additive, not a rewrite.
2. Add a class-dropper override to the Speed Gate. When a race moves off the turf, a stakes or allowance dropper should not be faded by the Speed Gate alone. This is the most clearly identified fix from the week.
3. Flag surface changes as a risk condition. The September 30 Tapeta switch was the week’s one real failure. Races moved to a different surface deserve a visible caution and a review of which filters still apply.
4. Trust the turf and route work. The October 3 turf sweep at Belmont Park suggests the late-kick logic is calibrated well for stakes and turf routes. Confidence there can stay high.
5. Keep reading pace shape on small ovals. The Saratoga Harness results show the contested-fraction analysis finding late-kick winners on a half-mile track. Preserve that logic as the harness card continues.
6. Respect the full Top 4 when building tickets. Third and fourth selections produced some of the week’s biggest prices, so structures that drop them give up the value this model is finding.

8. Data Notes & Caveats

• Reporting period runs December 26, 2025 through October 4, 2026; this week’s window is September 28 through October 4.
• Prior totals (2,146 races, 78.0%, ~$7.98) reflect data through September 27 and are condensed into a single Prev. Totals row.
• This week’s additions came from Finger Lakes, Belmont Park, Saratoga Harness and Keeneland only; the other tracks in the dataset had no new races.
• The 74 of 115 weekly figure (64.3%) is the Top 4 hit rate for the week, consistent with the 1st–4th pick columns in the daily table.
• Rank-distribution percentages are shares of all 1,747 cumulative hits, not of all races.
• Average payouts are for AI hits only and are shown as reported; the cumulative figures are approximate (~).
• Single-day results, such as the 12.5% at Belmont Park on September 30, come from small samples and should be read alongside the cumulative totals.