NFL heavy favourites went 268-0. The edge was still 1.0%.

We ran the favourite side of a market-bias hypothesis we had already tested from the underdog direction. The claim under test: books systematically underprice NFL heavy favourites, leaving a harvestable edge on the chalk.

The sample: 268 graded legs across 52 events. The record: 268-0.

That number will do more damage to a reader's intuition than any argument we could make, so let's put the rest of the table next to it immediately.

What we found

Market-implied win rate across the sample: 99.0%. Actual win rate: 100.0%.

So the hypothesis is directionally alive — the favourites did win more often than the price said they would. A one-point gap in the market's favour, realised.

And the return on that perfect record was 1.0%.

That is the whole lesson in one line. At 99.0% implied, you are laying enormous prices to collect a sliver. A 268-0 run is not evidence of a large edge; it is what a 99.0% market *should* produce, give or take. The record and the return are measuring completely different things, and only one of them pays.

The significance test settles it: p = 0.231. At n = 268 legs across 52 events, a one-point overperformance against a 99.0% baseline is comfortably inside what chance produces. We cannot reject the null. Verdict: NOT SUPPORTED.

We want to be precise about what failed. The favourites did not fail — they went 268-0. The *hypothesis* failed, because the hypothesis was that the mispricing is large enough to bet, and at 1.0% with p = 0.231 it is not. A signal that needs 268 legs to show a point of edge and still can't clear significance is a signal we cannot size.

There is also a structural problem sitting underneath the arithmetic. At 99.0% implied, the loss distribution is brutally asymmetric: one adverse outcome erases dozens of wins. A sample with zero losses tells you almost nothing about the tail you are actually exposed to, because the tail is the entire risk and it hasn't shown up yet. n = 52 events is not enough events to have met it.

What it means

The underdog side of this bias did not clear. The favourite side does not clear either. That closes the symmetric version of the question, and closing a question is worth something — it stops us re-running the same test in new clothing.

What we are testing next: the interesting region was never the 99.0% extreme. It is the middle, where outcomes are genuinely in doubt and a point of mispricing is a point on a price you can actually size. We are re-cutting the same event set into implied-probability bands and re-running the bias test inside each one, so a real edge at 65% isn't averaged into invisibility by the chalk. We are also splitting by whether the number moved after open, since a stale price and a well-traded price are different animals wearing the same digits.

What would change our mind: a materially larger event sample — not leg count, *event* count — showing the favourite overperformance holding at p < 0.05, or a band-level cut where the gap is wide enough to survive the vig without a 99.0% price attached.

21+ where legal. Research note, not advice.