Open questions about how DBSC's published HalSail results were modelled that we can't resolve from the data alone. DBSC has offered to help. Each answer lets us reproduce the affected results exactly and informs the general reconstruction.
Add to this list whenever the reconstruction hits a divergence that comes down to a HalSail/scoring decision rather than a bug on our side.
In HalSail, DBSC runs a tandem series per class/fleet, and each tandem has explicit, per-class control over which races it includes. The results manager chooses, class by class, exactly which heats land in each (class × tandem) cell.
That single mechanism is behind every divergence below:
- Q1 / Q5 — "flicking" single-competitor heats out of a class's tandem.
- Q2 — winding up the Sigma 33 one-design at 14 Aug (later heats just not included).
- Q3 — the abandoned 05 Jun heat included in Cruisers 3 ECHO's Series A but not its Overall.
- Q4 — one-design classes that share a start placing it in different series.
The per-case rationale / intent (deliberate vs oversight) is unclear and not material; the mechanism is the point. Implication for Sail Scoring: a sub-series here is one shared race set scored across all fleets, whereas a HalSail tandem's race membership is per-class. Reproducing these tables exactly therefore needs per-fleet (per-class) race membership within a sub-series — the gap that makes Q4 unrepresentable today and that the manual single-competitor exclusions (Q1) also land in.
Answer (Jun 2026). It is 100% manual enforcement of a DBSC Sailing Instructions clause by the results manager — a per-heat human decision, not an automatic rule. Every single-competitor heat is meant to be struck. The "kept" cases below that suggested a "primary class keeps its own start" rule were a results-manager miss: Cruisers 0 ECHO's 26 Jun one-boat heat should have been excluded too; it simply wasn't. So there is no tandem-vs-primary distinction — just the SI clause, applied by hand, with the occasional miss.
What we saw (the evidence that misled us):
| Fleet | What it is | Its single-boat day(s) |
|---|---|---|
| Cruisers 3 IRC | IRC tandem (the 2 IRC-rated boats of Cruisers 3) | excluded (9 of 15 days kept) |
| Cruisers 2 - Sigma 33 | one-design split of Cruisers 2 | excluded (12 Jun, 26 Jun, 31 Jul, 28 Aug) |
| Cruisers 5A ECHO | ECHO sub-fleet of the combined 4-5A VPRS pool | excluded (05 Jun — only boat 8237 of the 5A boats raced) |
| Cruisers 0 ECHO | the Cruisers 0 class's own series (own start) | kept (26 Jun) — ⚠ this is the error: should have been excluded |
| Cruisers 3 ECHO | the Cruisers 3 class's own series | kept |
We had reverse-engineered a "start-scoped, full-field-keeps" rule to fit the
"kept" row — but it was fitting a mistake. The proposed engine rule
(sailscoring/sailscoring#232) is therefore abandoned (closed not-planned; the
prototype lives on the local dbsc-single-competitor-rule-wrong branch).
Consequence for reconstruction. There is no rule to encode. To reproduce these tables exactly we'd have to replicate the manager's manual exclusions heat-by-heat (including their misses and inconsistencies) — best modelled as an explicit per-heat manual exclusion, not inferred scoring. Until that exists, the affected sub-fleets over-score the single-boat days (they count them as DNCs), and Cruisers 0 ECHO's 26 Jun is a published error we won't reproduce.
Separate from the single-competitor exclusion above (question 1), the Sigma 33 one-design table simply stops at 14 Aug, even though all three boats keep racing in Cruisers 2 ECHO through 28 Aug. In particular 21 Aug is excluded despite all three Sigma boats sailing it (≥2 — so it's not the question-1 exclusion). The one-design split looks like it was wound up mid-August while the parent class continued.
Mechanism (Jun 2026). Per-class tandem race membership (see the overview): the Sigma 33 tandem simply didn't include the post-14-Aug heats. Whether that wind-up was deliberate is unclear and not material.
On 05 Jun, the Cruisers 3 race was effectively abandoned — all ten boats
are RET / TLE / DNC, nobody finished:
46 RET · 246 RET · 1038 RET · 2102 RET · 5795 RET · 307 TLE · 11 DNC · 35 DNC · 1792 DNC · 2855 DNC
HalSail's own published Cruisers 3 ECHO tables treat that day inconsistently:
- Overall doesn't include 05 Jun at all (it runs 29 May → 12 Jun).
- Series A does include 05 Jun, scoring those boats their RET/TLE points.
So Series A contains a race the Overall excludes, for the same fleet. We drop a heat that nobody finished, which matches the Overall (parity-green) but not Series A (the RET/TLE boats diverge). The two published tables can't both be matched.
Mechanism (Jun 2026). Per-class tandem race membership (see the overview): the 05 Jun heat was included in the Series A tandem but not the Overall tandem. Slip or intent is unclear and not material; the two tandems just carry different race sets.
On the one-design days (Saturday, Tuesday), each class keeps its own Series A / Series B boundary, because the sparse classes (ILCA, IDRA, Melges, …) sail only a handful of races. The split can fall on the same physical start: on the 07 Jun 2025 14:03 start, Fireball scores it Series A while IDRA 14 and Melges 15 (who started alongside it) score it Series B.
In Sail Scoring we model a day as one series with the tandems as sub-series, where a sub-series is a single shared set of races scored across every fleet. That cannot represent two classes assigning the same start to different series. The day's Overall reproduces exactly (each fleet scores only the races it sailed); the per-class Series A/B cells diverge for the classes whose boundary differs from their start-mates'.
Three classes (Dragon, SB20, Sportsboats) additionally have one race that is in their Series A/B but not their Overall — the same Overall-vs-tandem inconsistency as question 3; our Overall-sourced build can't see those races at all.
Mechanism (Jun 2026). Confirmed: the one-design Series A/B are independent per-class race subsets (per-class tandem race membership — see the overview), so two classes on one start genuinely land it in different series. There is no single intended A/B calendar boundary.
Impact / status. Accepted as a modelling divergence (it is not a scoring bug): the combined day-file reproduces every Overall and most A/B cells, but cannot reproduce A/B for classes whose boundary disagrees with their start-mates' — because our sub-series share one race set across fleets, while HalSail's tandems are per-class. Closing this gap needs per-fleet race membership in a sub-series (see the overview's implication). See the Saturday/Tuesday one-design groups.
Answer (Jun 2026). Same resolution as question 1: the exclusion is manual SI enforcement, so the Thursday-vs-Saturday difference for the same fleet is just human inconsistency, not a rule. There is nothing to reverse-engineer.
The evidence, for the record — Cruisers 3 IRC (boats 998, 5795):
- Thursday: 15 days sailed, 9 kept — its single-boat days were struck, and all 9 kept heats have both boats finishing.
- Saturday: 5 kept, including two single-finisher heats that were not
struck: 21 Jun (
9981st,5795DNF) and 20 Sep (9981st,5795DNC, scored2/DNC).
The same fleet's single-boat heats were struck on Thursday but left on Saturday — a manual application that varied by day. This confirmed question 1's answer (a hand-applied SI clause with misses) rather than posing a separate puzzle.
Impact / status. No engine rule (sailscoring#232 closed not-planned). The
Saturday cruiser residuals — Cruisers 3 IRC, Sigma 33, and the small (±1) net
offsets that ripple through the progressive ECHO chains (Cruisers 1/5A/5B ECHO)
when a mid-chain heat is or isn't struck — all trace here, and would only resolve
by replicating the manager's manual exclusions heat-by-heat (see question 1).