Cricket prediction performance, 12-18 October 2026

How the week went

Most of 12–18 October played out as a “chalk” week in the CSA T20 Challenge: the better-fancied sides generally did what the prices suggested, and when they won, they often won comfortably. The Lions were a good example of that pattern — a 5-run escape against Eastern Cape Linyathi (Border) at 1.16 was tighter than expected, but they followed it with a far more routine 8-wicket win over Northern Cape (1.15) with 54 balls in hand.

There were still a couple of results that cut against the market. Northern Cape beating Limpopo was flagged as an upset despite the prices being close (Limpopo 1.80 favourites, Northern Cape 1.91), and it arrived with a DLS twist: Northern Cape won by 14 runs (DLS method) after Limpopo were left chasing in a shortened reply. The other proper stumble for the favourite came when Eastern Storm (2.20) beat the Knights (1.57) by 18 runs — a clean result, not a last-over coin flip.

Beyond those, the week had a nice spread of match shapes: a couple of thrillers (Western Province getting home with one ball left against Rocks), a couple of nail-biters that still went with the favourite (Border losing by just 5 to the Lions; Limpopo falling 3 short against Border), and at least one outright blowout (Warriors winning by 9 wickets with 63 balls remaining after bowling South Western Districts out for 68). It was the kind of schedule where a forecaster’s biggest edge wasn’t necessarily “finding chaos”, but avoiding the few landmines.

Where the methods landed

On raw hit-rate, it was unusually compressed: ML, AI 1, AI 2 and Maths all finished 8 wins from 11 (73% accuracy). The Books line was better again at 9/11 (82%). That’s the headline — but the more interesting part is what happened to profit and ROI underneath identical accuracy.

AI 1 produced the best model return at +5.1% ROI (profit +56), with AI 2 next at +2.5% (profit +27). ML was positive but modest at +1.5% (profit +16). Maths is the cautionary tale: it matched the others on 73% accuracy yet barely moved the needle at +0.1% ROI (profit +1). That’s the cleanest illustration you’ll get in a single week that accuracy and profit are not the same thing: if your correct calls are concentrated on short prices, you can be “right” often and still not get paid for it.

Books topped both columns: 82% accuracy and +16.5% ROI (profit +181). Some of that is simply what happens when the market favourite wins a lot — but it’s also a reminder that, in a week without many long-shot winners, the path to profit is mostly about not dropping the wrong favourites rather than hunting heroic outsiders.

The matches that decided it

Northern Cape vs Limpopo (13 Oct) was one of the two true market upsets and it split the panel. Limpopo were the 1.80 favourite and four of the five methods backed them (ML, AI 1, Maths and Books). Only AI 2 went against the grain with Northern Cape — and got paid when Northern Cape won by 14 runs (DLS method). In a week where most results followed the market, that single contrarian hit mattered disproportionately for model-to-model separation.

Knights vs Eastern Storm (14 Oct) was the other upset that really bit. Knights were priced 1.57 with Eastern Storm at 2.20, and again most methods stayed with the favourite: ML, AI 2, Maths and Books all took Knights. AI 1 was the lone holdout on Eastern Storm and was rewarded when Storm won by 18 runs. If you’re looking for why AI 1’s ROI led the model pack despite the same 8/11 accuracy, this is the cleanest exhibit: it found value on the one upset the others mostly missed.

Western Province vs Rocks (14 Oct) was the big disagreement game that didn’t show up as an upset, but still shaped the week’s “feel”. Western Province were a slight favourite at 1.80 (Rocks 2.00). Books and ML sided with Western Province; AI 1, AI 2 and Maths all went with Rocks. Western Province won by 6 wickets with one ball remaining — close enough that it could have swung the other way, but on the ledger it gave Books and ML a key point of separation over the three methods that faded them.

What it means

This was a week where the schedule rewarded staying mostly aligned with the market — and the Books line was both accurate (9/11) and strongly profitable (+16.5% ROI). The models were consistent on accuracy (all 8/11), but the spread in ROI (+0.1% to +5.1%) underlines a basic truth: two methods can be “right” equally often and still have very different economic outcomes depending on which prices they tend to collect.

It’s also a reminder that a small number of games can dominate the story. Two upsets — Northern Cape over Limpopo and Eastern Storm over Knights — did much of the work in separating the methods this week, especially because they were not consensus picks. In a chalky run, performance often comes down to how you handle the handful of matches the market gets wrong, not how well you identify the obvious ones.