The verdict
In July we published forecasts for every 189 occupation pool we track, built on the June SkillSelect snapshot — and committed to grading them publicly when the real numbers arrived. July's SkillSelect data is in: we graded 67 189-pool forecasts, and our average error came in at 2.24% against 3.10% for the naive “no change” baseline — we won this month. 65 of the 67 landed inside the published 80% ranges; the two that missed are shown below, not buried.
The 189 pool
The 189 Points-Tested pool, graded against the July snapshot:
| Series | Predicted | 80% range | Actual | Error |
|---|---|---|---|---|
| 189 Points-Tested pool | 181,322 | 172,212–190,914 | 178,506 | +1.6% ✓ |
Big occupations
The eight largest 189 occupation pools. Early Childhood Teacher was our best call of the month — off by 0.3%.
| Series | Predicted | 80% range | Actual | Error |
|---|---|---|---|---|
| Early Childhood Teacher (241111) | 12,306 | 11,688–12,957 | 12,343 | -0.3% ✓ |
| Chef (351311) | 11,961 | 11,360–12,594 | 11,449 | +4.5% ✓ |
| Software Engineer (261313) | 9,939 | 9,440–10,465 | 9,756 | +1.9% ✓ |
| Civil Engineer (233211) | 9,392 | 8,920–9,889 | 9,154 | +2.6% ✓ |
| ICT Business Analyst (261111) | 8,234 | 7,820–8,670 | 8,037 | +2.5% ✓ |
| Mechanical Engineer (233512) | 8,190 | 7,778–8,623 | 7,880 | +3.9% ✓ |
| Accountant General (221111) | 7,941 | 7,542–8,361 | 7,682 | +3.4% ✓ |
| Developer Programmer (261312) | 6,866 | 6,521–7,229 | 6,771 | +1.4% ✓ |
Case studies
This is the largest 189 pool we track, and it was our best call of the month: the model landed within 37 EOIs of the actual figure, despite the July financial-year expiry turnover shaking up smaller pools around it.
The point forecast missed by 4.5% — the month’s worst error among the big-8 — but the published 80% range caught the actual comfortably. This is exactly what the range is for: point forecasts miss, and the range is where the honesty lives.
The pool shrank through the bottom of our range — by just 9 EOIs, but a miss is a miss. The only other miss of the 67 went the other way: Physiotherapist (252511) came in at 585, three EOIs above our ceiling of 582. Two misses in 67, one in each direction, both within a whisker of the boundary — that’s what 97% coverage looks like in practice.
The six-month blind test
Same blind protocol as our methodology post: we refit the engine on data up to January 2026 only, and asked for the next six months. February to May landed within 1% of the point forecast — the May call was off by less than 1,800 EOIs on a 185,000 pool. Then came the part no January model could know: the end-of-financial-year expiry wave pulled the pool to 178,506 by July, 8.2% below the point forecast — and the deliberately widening range absorbed it. All six months landed inside the band.
Three blind-test case studies, run under the same protocol — production fit, history cut at January 2026, six months predicted blind:
Engineering Technologist (233914) is the star case of our methodology post, re-run on six fresh months: refit on data through January 2026 and asked to predict February–July blind, the model landed within 1.5% of the actual at six months out, and all six months fell inside the published range.
At six months out, the point forecast landed on the actual to the exact EOI — Chemist (234211) predicted 618, and 618 arrived. This is a small pool (around 600 EOIs), so some of that precision is luck, but the full six-month path stayed inside the range too.
Secondary School Teacher (241411) is the honest one: the pool surged faster than the model predicted from February through May — four straight months above the range — before the July financial-year expiry wave pulled it back inside. The six-month endpoint looks respectable at +4.5%, but the path shows two errors cancelling out, not one good forecast; that’s exactly why we publish the whole trajectory instead of grading only the endpoint.
The queue at your score
The pool totals above grade whole occupations. But what most readers actually use is the queue at their score — how many 189 EOIs sit at or above 65, 70, … 95 points in their occupation. Here too the model beat the naive baseline — at every single score level (5.86% vs 6.87% across 545 matured threshold forecasts), with 97.8% of actuals inside the 80% ranges. The weak spot was the 80-point bracket (12.0% error), sitting on July's expiry-wave fault line.
| Score | Forecasts | Our error | Naive error | In-range |
|---|---|---|---|---|
| 65+ | 129 | 4.18% | 5.18% | 100.0% |
| 70+ | 118 | 4.65% | 5.87% | 100.0% |
| 75+ | 106 | 6.48% | 7.71% | 96.2% |
| 80+ | 73 | 11.98% | 12.87% | 90.4% |
| 85+ | 53 | 4.25% | 4.82% | 100.0% |
| 90+ | 42 | 5.24% | 5.94% | 97.6% |
| 95+ | 24 | 4.16% | 5.03% | 100.0% |
| All | 545 | 5.86% | 6.87% | 97.8% |
Both the 189 pool totals and the score-level queues beat the baseline this month; the grades that matter to a 189 applicant all came in green — and we’ll publish them every month, including when they don’t.
Where the errors came from
Nearly every pool came in slightly below the point forecast — July is the financial-year turnover month (a wave of two-year-old EOIs expires) and the model’s trend expected more drift than the wave allowed. The 2.24% average error and the two boundary-graze misses say the 189 fits are tight; the over-coverage (97.0% vs an 80% target) means the published ranges could narrow as more months mature — and we’d rather narrow them slowly than miss fast.
What matures next
In early September we grade the June vintage at two months out and the July vintage at one month out — every month, wins and losses alike. Method details: how we test our forecasts. Live numbers: EOI trends.