The Gate We Apply to One Sport, Skip in Another

For most of the shop's first two years, we ran a gate that asked whether a player had logged meaningful minutes in each of their last five appearances. It was built for basketball, where a player who has been coming off the bench for eight minutes a night is a fundamentally different proposition than one who has been starting. The gate was simple, it was cheap to run, and it worked. So we applied it to hockey without thinking about it much.

Hockey has lines. A forward who plays on the fourth line gets ten minutes a night by design — not because he is struggling, not because he is in a coach's bad graces, but because that is structurally what fourth-line forwards do. We were killing candidates who had no business being killed, and we were doing it confidently, which is the worst way to be wrong. It took a calibration review to surface the damage: our hockey ratings had been systematically thin in a particular category for nearly a full season, and the cause was a basketball gate we had never bothered to interrogate.

That is the version of the story where we catch it. There are versions where we still haven't.

Where the inconsistency actually lives in this piece

The problem is not that we apply different gates to different sports. We should. The problem is that we apply the same gate to different sports without realizing we've done it, and then we apply different gates to the same sport without realizing we've done that either.

The minutes gate is the clearest example, but it is not the only one. We have a gate that checks for back-to-back appearances — two games inside thirty-six hours — and flags any player scheduled into that window as a corpus risk. We built it for basketball, where fatigue compounds over a compressed schedule in ways the raw stat line never captures. We apply it consistently to basketball. We do not apply it to soccer, where a player who featured for ninety minutes on Saturday and is listed to start on Tuesday is in an arguably more fatigued state than any basketball player we have ever flagged.

When Renata asked why the soccer corpus kept producing ratings that graded out softer than expected in the back half of double-match weeks, I didn't have a good answer. The gate existed. We had written it. We were just using it in one sport and ignoring it in another.

"The gate wasn't wrong. We just forgot it was ours to apply everywhere, not just where we first needed it."
— Renata

This is distinct from the problem we wrote about when a gate built for one context quietly outlives its usefulness — that piece is about gates that should have been retired. This one is about gates that are still earning their keep, just not everywhere they should be earning it.

How we tried to audit which gates were actually sport-agnostic

After the hockey calibration review, we ran a full cross-sport gate audit. The idea was straightforward: list every gate in the system, identify which sport it was originally built for, and then ask explicitly whether the underlying logic applied to each of the other four sports. If it did, it should be running there. If it didn't, we should be able to say why in a sentence.

We found twenty-three gates. Eleven had been built for basketball. Four had been built for baseball. The remaining eight were distributed across hockey, soccer, and tennis with no particular logic — they had been added as problems surfaced, which is how most of the gates got built, and which is also a process that tends to produce gates that were applied to four sports before anyone thought carefully about whether they should be.

Of the eleven basketball gates, we determined that seven had logic that transferred to at least one other sport. Of those seven, three were already running in those sports. Four were not. The minutes gate was one of the four. The back-to-back gate was another. A third checked for stat-type volatility in small samples — a player with fewer than eight appearances in the corpus window — and it was running in basketball and baseball but not in tennis, where eight appearances is an extremely generous sample and where the stat types are, if anything, more volatile.

The fourth one I am still not comfortable describing, because we haven't finished deciding whether the transfer is valid or whether we are just telling ourselves it is.

What the inconsistency actually cost us in the grading record

The honest answer is that we don't know the full cost, and that not knowing is itself the finding. When a gate is missing, it doesn't produce a visible error — it produces a slightly worse corpus, and a slightly worse corpus produces ratings that grade out a little soft, and ratings that grade out a little soft look like noise until you have enough of them to see the shape.

What we could measure: in the season before the audit, our hockey ratings in the minutes-affected category had a calibration gap of roughly eleven points. We were stating confidence at a level that our observed accuracy didn't support. That gap is not catastrophic, but it is not nothing, and it had been there the whole season without us noticing because we were looking at hit rate, not calibration. This is the failure mode we have written about before — trusting a gate because we were the ones who built it is not a defense against the gate being incomplete.

The soccer finding was harder to quantify because the double-match-week effect is intermittent. Some weeks it showed up clearly in the grades; other weeks it didn't. Marcus spent three days trying to separate the fatigue signal from the opponent-quality signal in the same window and concluded that we probably couldn't, not with the corpus depth we had at the time. We noted the gap, added the gate, and accepted that we would need another full season of grades before we could say whether it helped.

That is the part nobody tells you about gate work: you often fix a problem and then wait a year to find out if you fixed it correctly.

What the audit actually changed, and what it didn't

We added the back-to-back gate to soccer. We recalibrated the minutes gate for hockey with a sport-specific threshold — the question is no longer "did the player log meaningful minutes" but "did the player log minutes consistent with their structural role," which sounds like a small change and is actually a significant one. A fourth-line hockey forward clearing that gate looks completely different now than he did under the basketball version.

We did not add every gate everywhere. The volatility gate for small samples, which we applied to tennis, turned out to behave strangely there — it was killing candidates who had cleared the gate in previous seasons and who had every reason to clear it again, just because a compressed stretch of the calendar had pushed them below the threshold. That felt like the gate doing more harm than good, and we pulled it back. The principle that some gates are applied last precisely because they eliminate the most applies here in reverse: a gate that eliminates candidates without improving calibration is just noise with authority.

What the audit did not change is the underlying process for building gates. We still build them in response to problems. We still build them in the sport where the problem appeared first. The audit was a one-time correction, not a structural fix, and we know it. The structural fix would require us to think about every new gate in every sport at the moment we build it, and we are a small shop, and that discipline has so far eluded us.

Renata's read on it was that the inconsistency is probably permanent: "You will always be one sport ahead of your own rules." I think she is right, and I think the correct response is to schedule the audit rather than to pretend the inconsistency won't recur.

What I keep returning to is that the gates we applied consistently were not necessarily the right ones — they were just the ones we remembered. Consistency and correctness are not the same thing, and a gate that is wrong in the same way across five sports is arguably worse than one that is missing in only one. I don't know how to resolve that. It might be the kind of thing that doesn't resolve.

Note: PlayerGem is a fictional analytics shop and these accounts are invented. Nothing here is a pick, a recommendation, or betting or investment advice, and the players, teams and competitions described do not exist.