What the Spreadsheet Says vs What You Feel - Which Wins When They Disagree?

[SharpEddie47]

I've been doing this for twenty years and tracking every bet for fifteen of them. The spreadsheet doesn't lie. My record on bets where I followed the model output with no reservations: positive CLV, positive P&L over meaningful sample.

My record on bets where I had a specific reservation before placing but placed anyway: noticeably worse. Not catastrophically worse. Measurably worse.

What I haven't published, and what this thread is about: my override rate.

The times I identified a qualifying selection, the model said bet, and I didn't. Or the reverse: the model said no qualifying edge and I bet something anyway.

I went back through five seasons of records specifically flagging these two categories.

Override rate: approximately 12%. One in eight decisions involved some form of deviation from what the model was telling me to do.

The P&L on the overrides: slightly negative overall. The intuition that felt like information was producing marginally worse results than the model.

Here's the thing that keeps me honest. The 12% doesn't feel like 12%. It feels like maybe 3 or 4%. The overrides in my memory are the ones that worked out. The ones that didn't are harder to recall with the same clarity.

My honest question for this forum: what's everyone's actual override rate. Not what you think it is. What the record shows.

Because I suspect most people who describe themselves as systematic have a higher override rate than their self-description implies, and they don't know it because they're not tracking the overrides the same way they're tracking the selections.

 

[Klaus]

The override rate question is the correct question and I've tracked it specifically.

Bundesliga model, fourteen seasons of data. Pre-committed decision rule: if the model output meets the threshold, bet. If it doesn't, don't.

My documented override rate: 8.3%.

The 8.3% breaks down as follows.

Approximately 5% are "failure to bet" overrides. The model qualifies a selection but I find a reason not to place. Usually this is a specific piece of late information that the model hasn't incorporated yet. Sometimes this is justified. Often it is not.

The remaining 3.3% are "bet without qualification" overrides. The model doesn't produce a qualifying output but I place something anyway.

The P&L on the second category is specifically and consistently negative across all fourteen seasons.

The P&L on the first category is genuinely mixed. Sometimes I correctly identified information the model missed. Sometimes I talked myself out of a good bet using a plausible story.

The honest conclusion: I am less systematic than the framework I've built implies. The framework is more systematic than I am.