The handicapper.
The handicap explains about a fifth of how long a ship takes over the course. The rest is pilot choice, traffic, berth readiness and tugs.
Expected times are a handicap, not a prediction of the clock. No stake, no prize.
The race itself is won on the actual course time (D-047); the handicap is the expected-time companion to that time, not a rival rule for the same contest. Race score is still computed, stored and shown beside every boat as vs expected, and it still decides nothing.
Across 3,500 races on 1,747 settled race days, the fastest boat over the course also posted the best handicap figure in 74% of them.
The model that is running
| Version | Fitted on | Sprints | R² | Holdout R² | Form prior |
|---|---|---|---|---|---|
| hc-1.0 | 19 September 2021 to 30 June 2025 | 11,671 | 0.196 | 0.187 | 0.053 |
The holdout is the tail of the window the fit never saw. A holdout R² that sits within a hundredth of the in-sample R² is the number worth trusting.
The spread a figure is measured in
| Class | Spread, log seconds | Share of the run | One figure on a 600 s run |
|---|---|---|---|
| Box ships | 0.155 | 15.5% | 93 s |
| Car carriers | 0.134 | 13.4% | 80 s |
| Tankers | 0.114 | 11.4% | 68 s |
| Cruise ships | 0.123 | 12.3% | 74 s |
| Other big ships | 0.170 | 17.0% | 102 s |
| Fallback | 0.152 | 15.2% | 91 s |
A figure is how far inside or outside her expected time a ship ran, in units of her own class’s spread. Positive is faster than expected. One decimal in public, and two figures within 0.03 of each other are a tie.
Is it calibrated
- Window
- 15 September 2026 to 12 October 2026
- Sprints
- 80
- Mean figure
- +0.172
- Spread of figures
- 0.842
- Inside the 10 to 90 band
- 86.3%
A calibrated handicap reads a mean figure near 0, a spread near 1 and about 80% inside its own 10th to 90th band. Higher coverage means the bands are too wide; lower means the model is more confident than it has earned.
| Class | Sprints | Mean figure | Spread |
|---|---|---|---|
| Box ships | 51 | +0.25 | 0.83 |
| Other big ships | 16 | +0.05 | 0.88 |
| Car carriers | 6 | -0.16 | 0.94 |
| Tankers | 5 | +0.54 | 0.45 |
| Cruise ships | 2 | -0.86 | 0.96 |
| Direction | Sprints | Mean figure | Spread |
|---|---|---|---|
| Outbound | 41 | +0.07 | 0.83 |
| Inbound | 39 | +0.28 | 0.85 |
| Conditions | Sprints | Mean figure | Spread |
|---|---|---|---|
| foul tide | 30 | +0.19 | 1.05 |
| fair tide | 29 | +0.17 | 0.66 |
| wet | 12 | +0.06 | 0.60 |
| slack | 8 | +0.33 | 1.06 |
| breezy | 1 | -0.07 | 0.00 |
Did the field work
| Race day | Entries | Ran | Overnight | Did not start | At berth | Scheduled | Offshore | On the field | Line error |
|---|---|---|---|---|---|---|---|---|---|
| 16 October 2026 | 0 | 0 | 0 | 0 | 0% | 0% | 0% | 0% | — |
| 15 October 2026 | 18 | 0 | 0 | 0 | 9% | 10% | 0% | 100% | 60.3 s |
| 14 October 2026 | 19 | 4 | 1 | 14 | 25% | 13% | 33% | 63% | 104.2 s |
| 13 October 2026 | 14 | 8 | 0 | 6 | 63% | 50% | 0% | 73% | 81.3 s |
| 12 October 2026 | 12 | 4 | 1 | 8 | 33% | 38% | 0% | 40% | 55.0 s |
| 11 October 2026gap in the data | 9 | 0 | 0 | 9 | 0% | 0% | 0% | 0% | — |
| 10 October 2026 | 11 | 4 | 2 | 5 | 33% | 40% | 0% | 100% | 97.6 s |
| 9 October 2026 | 19 | 3 | 5 | 14 | 43% | 0% | 0% | 50% | 37.9 s |
| 8 October 2026 | 14 | 2 | 3 | 10 | 14% | 14% | 0% | 57% | 81.6 s |
| 7 October 2026 | 18 | 7 | 4 | 9 | 57% | 27% | 0% | 75% | 119.9 s |
| 6 October 2026 | 21 | 3 | 3 | 15 | 27% | 0% | 0% | 75% | 79.8 s |
| 5 October 2026 | 17 | 7 | 1 | 9 | 57% | 30% | 0% | 73% | 56.5 s |
| 4 October 2026 | 19 | 2 | 3 | 14 | 0% | 17% | 0% | 83% | 77.3 s |
| 3 October 2026 | 12 | 3 | 3 | 8 | 0% | 38% | 0% | 57% | 26.6 s |
The three hit rates are the share of that kind of entry that ran. On the field is the share of the day’s runs made by a ship the field had named. Line error is the median gap between the opening line and the clock. A race day flagged for a gap in the data is published and left out of the scoring: 13 of 14 race days here are scored.
What goes in
One ordinary least squares on the log of the elapsed seconds: direction, length, class, draught, and the along-channel current signed so that positive always means it helped her. Every one of those is knowable the morning before she runs, which is the whole test of a handicap. Fitted separately for each season.
What stays out
Wind: measured, it added 0.0005 to the R². Traffic: passing an opposing big ship happens on about 18% of runs and costs around 6% of the elapsed time, but it is unknowable at the start, so it is recorded after the run as something that happened to her and never as an adjustment to what was expected of her.
Form
A ship’s own mean residual, shrunk toward zero: one run on record keeps a few percent of it, twenty keeps most of it. Form shifts the odds and never the expected time, because a figure has to measure the ship, and her own record cannot be baked into what she is expected to do.
Odds
The odds are the probability she posts the fastest time in her race (D-047), simulated over the day’s field from the same per-boat time distributions, printed as a fair fraction with nothing added. With no form on record the field is level and the honest answer to who wins today is any of them. Information only. No stake, no prize.
Where the numbers live
Every scored sprint is a row in the ledger, kept under the model version that scored it; a refit lands beside history instead of rewriting it. A run that lost an input to a fallback is flagged, and a page that shows its number says so. See the records and today’s race.