Round 12 · Issue 1
This week in 60 seconds
Our system runs four independent versions of the NRL try-scoring model for every fixture, then compares what each version says against the best price across seven Australian bookmakers. When the model thinks a player is more likely to score than the bookie price implies — that's an edge.
How we read the round
We don't rely on a single model. We run several reads of every fixture, each tuned differently, and put them side by side before we publish anything.
When all of them like a player, that's a strong signal something structural is in our favour. When they disagree, that's information too — usually a sign to size smaller and watch.
What you'll see below: the picks themselves, why we like them, how confident we are, and where each one could come apart.
What the model flagged this round
One high-conviction edge (Brooks), one medium-conviction edge from our newest model version at half stake (Laban), one borderline structural call at half stake (Duncan), and five small speculative bets concentrated across Cowboys vs Rabbitohs and Dragons vs Warriors. These are model-identified edges, not guaranteed winners.
The cheat sheet — 8 picks, 3.25 units nominal exposure
| # | Player | Fixture | Price | Model | Market | Edge | Stake |
|---|---|---|---|---|---|---|---|
| 1 | Luke Brooks | Manly v Titans | $4.10 | 32.5% | 24.4% | +8.1pp | 1.0u |
| 2 | Jacob Laban | Dragons v Warriors | $5.00 | 26.0% | 20.0% | +6.0pp | 0.5u |
| 3 | Tallis Duncan | Cowboys v Souths | $2.75 | 40.0% | 36.4% | +3.6pp | 0.5u |
| 4 | Wayde Egan | Dragons v Warriors | $7.00 | 21.9% | 14.3% | +7.6pp | 0.25u |
| 5 | Reed Mahoney | Cowboys v Souths | $6.80 | 22.9% | 14.7% | +8.2pp | 0.25u |
| 6 | Peter Mamouzelos | Cowboys v Souths | $7.50 | 28.4% | 13.3% | +15.1pp | 0.25u |
| 7 | Matthew Lodge | Cowboys v Souths | $11.00 | 11.8% | 9.1% | +2.7pp | 0.25u |
| 8 | Sean Keppie | Cowboys v Souths | $11.00 | 14.5% | 9.1% | +5.4pp | 0.25u |
Green = high-conviction (1.0u). Amber = medium / borderline (0.5u). Unshaded rows = small speculative bets (0.25u each). A unit ("u") is whatever you have decided one unit means for your bankroll — for some readers that's $10, for others $100. Pick a unit size you can lose without stress.
Fixture concentration warning
5 of 8 picks are on a single fixture (Cowboys vs Rabbitohs: Duncan, Mahoney, Mamouzelos, Lodge, Keppie). Those picks share covariance — a defensive grind, an early sin-bin, weather, or a late lineup change can move multiple picks together. The nominal total of 3.25 units treats them as independent. The risk-equivalent exposure on that single fixture is closer to 2.0 units. If you bet, consider a per-fixture stake cap rather than treating each pick as independent risk.
Per-match previews
Plain-English read on every R12 fixture — the story, the picks in that game, what to watch.
Round 12 · Two Hero Bets
Our high-conviction call
Of the 8 picks on the slip, one stands out as our highest-confidence call — a player where the model sees a clear edge over the market and the underlying data supports it.
Luke Brooks
The setup: Sea Eagles host Titans on Saturday, priced as $1.30 favourites — a comfortable home-win matchup. Brooks is Manly's playmaker; he scores at about 1 try every 2–3 games.
The edge: the bookies price him as roughly a 24% chance. We have him meaningfully higher than that. Most of our reads of the fixture agree; one is more cautious because the books themselves are unusually split on this market — a signal we'll keep watching across the next few rounds. At full minutes against a leaky Titans middle defence, this is our cleanest read of the weekend.
Jacob Laban
The new-model pick — at half stake. This is the pick the brand-new version of our model surfaced. Laban plays on the Warriors' attacking edge against a Dragons defence the headline numbers say is solid — but our new model architecture flags as leakier on the second row.
The edge: bookies imply ~20% at $5.00; model says ~26%.
Why half stake: only our newest read of the round surfaces him — our other reads stay quiet. That's an unproven signal, so we back it, but at half the size we'd use on a pick everyone agrees with. If calls like this start landing consistently we'll lean on that read more in future reports.
Round 12 · The Borderline Call
Our borderline call — half stake
Tallis Duncan is listed at centre in the official Rabbitohs R12 team sheet (NRL.com, named at #4). With Latrell Mitchell out of the named 17, Duncan is the centre rotation slot. Most of our reads rate him a high-tier pick; our most conservative read pulled him just below the line. We're backing the structural case at half stake — reflecting both the team-sheet position and the fact our reads aren't fully aligned.
Tallis Duncan
The structural case: Latrell out → Duncan named at centre #4 per the official team sheet. Cowboys concede tries to centres above league average. Rabbitohs score 1.4 tries per game from centres combined. The model's older versions had him at ~50%; the newest version's conservative filter pulled him to 40%.
Caveat worth noting: Duncan's recent grade output has been predominantly forward (back-row / lock). The centre listing on the team sheet is the structural input here — if the lineup shifts or if his role is utility rather than primary centre, the case weakens. Half stake reflects this lineup-confirmation uncertainty.
Where the edge lives: not in claiming Duncan-as-elite-centre (no track record there) but in the position–team–defence triangle. Whoever Souths name at centre against Cowboys on Sunday should score at roughly 35–50% — the market price implies 36% — small edge if the lineup holds.
What we'll learn
Duncan's result is one data point of many. We'll log him alongside the other R12 picks and re-evaluate the model's newest filter at sample size (n ≥ 20 borderline calls across rounds), not on a single outcome. If borderline picks like this systematically outperform the new filter's cutoff over R12–R15, we'll relax the threshold for R16.
Round 12 · Cross-Model Consensus
Five picks every model version agrees on
These five players appear on every single one of our model versions' pick lists. That kind of cross-version agreement is rare — every architectural change we made still kept them on the slip.
Warning — testable claim
All five are forwards/hookers on either Cowboys or Rabbitohs. Cowboys' forwards have scored 4 tries across 11 games this season; Rabbitohs' hookers have scored 0 tries in 10. The model thinks the matchup creates opportunity the season-long stats don't reflect. This is our most testable claim of the week.
Wayde Egan
Warriors hooker. Model 21.9% vs market 14.3% implied (+7.6pp). Scores when team is attacking, and the Warriors are heavy favourites here.
Reed Mahoney
Cowboys hooker. Model 22.9% vs market 14.7% implied (+8.2pp). Spinal threat near the line; model thinks the matchup creates more opportunities than recent season stats show.
Peter Mamouzelos
Biggest model-vs-market gap of the five picks — model 28.4%, market 13.3% implied (+15.1pp). But this is also the model's least-supported call: Rabbitohs hookers have scored 0 tries across 10 games this season, and the model's hooker calibration is unproven (we don't have settled-bet evidence that this position is correctly calibrated yet). The pick clears our edge threshold but should be read as a model claim under test, not a confident edge.
Matthew Lodge
Cowboys prop at the longest price of the round. Model 11.8% vs market 9.1% implied (+2.7pp). Edge is small but the $11.00 long-shot price means even a marginal probability advantage produces positive EV on the bet.
Sean Keppie
Rabbitohs prop. Model 14.5% vs market 9.1% implied (+5.4pp). Long-shot upside at $11.00 with a modest model edge. Last on the slip — completely speculative.
Round 12 · Discipline
What we're NOT betting (and why)
Being clear about what we skip is as important as what we back. Three categories of picks we considered and rejected:
1. The 7 picks our older model was hot on
Our older read of the round flagged players like Bradley Schneider, Isaiya Katoa, and Thomas Mikaele. Our newer, tighter reads all killed those picks. We're trusting the newer work — it's tuned to catch exactly the kind of false positives the older version was producing. If a couple of those eliminated names score this weekend, that doesn't mean we got it wrong — at typical scoring rates around 15–25% you'd expect one or two of seven to land by chance. We'll need a pattern across several rounds before we'd loosen anything back up.
2. The big-name tryscorers everyone's talking about
Alex Johnston (Rabbitohs winger) is the round's biggest tryscoring threat by reputation — scores in 9 of 10 games. The bookies price him at $1.55 (~65% implied chance). Our model says ~51%. That's a -14pp negative edge, not a neutral pass — the model sees him as materially over-priced relative to his matchup. If a lay market is available on a betting exchange (e.g. Betfair Lay Anytime Tryscorer), this is the round's strongest model-flagged lay candidate. We're not publishing it as a back bet because we don't currently bet lays via fixed-odds books, but readers with exchange accounts should be aware that the model's stance here is oppose, not ignore. Same shape for Brian To'o and Thomas Jenkins: elite tryscorers priced where the market has already accounted for the floor — we don't see edge on the back side, but model lean is on the unders.
3. The match-level head-to-head bets
We don't currently bet head-to-head outcomes. Our model is good at predicting match margins but less reliable at converting that to win probabilities. Until that improves (mid-season backlog), we stay in player markets where our edge is more measurable.
Round 12 · Responsibility
Honesty section
How confident should you be?
We're not making performance claims yet. Our public work has been live for one fully-settled round (Round 11). That's not enough to claim a track record — and we won't pretend it is. We'll publish hit rates and returns only when the sample is large enough to mean something.
What we did review from Round 11: every pick, settled honestly, with a short note on what worked, what didn't, and what each result told us about how to tighten the work this week. Available on request.
We track every prediction in a database that auto-settles against actual results each week. After R15–R16 we'll have ~100 settled picks across multiple model versions — that's when ROI numbers become statistically meaningful, and that's when we'll publish them. Until then, treat these picks as informational forecasts, not proven money-makers. Size your bets so a losing round doesn't blow up your bankroll.
What could go wrong
Injury risk — most Australian bookies stand bets on named-and-played players even if they go off injured early. A 6-minute HIA on your pick is the same as a loss. Smaller stakes protect you.
Bunker decisions — a try ruled out in the bunker is the same as no try for betting purposes. Mahoney or Mamouzelos can score and have it disallowed → still a loss.
Late lineup changes — if a picked player is dropped from the starting 17 before kickoff, the bet is usually voided. If they're named as 18th-man injury sub, also voided. Check NRL.com Saturday morning.