# We tested how 189 ranks EOIs on four rounds: what held and what didn't

Points and date of effect hold almost exactly: 99.96% of EOIs above their occupation's minimum were invited. The unit-group queue fits 48 of 52 cases where it can be tested, with three misses in November 2024. What the data shows, where it fails, and what it does not prove.

These are test results, not an official description: they describe what happened in four past rounds, not how Home Affairs runs a round.

Page: https://immitrend.com.au/blog/189-ranking-rules-tested-four-rounds

## The short answer

The two published rules hold almost exactly. Across four subclass 189 rounds, 99.96% of EOIs above their occupation's minimum score were invited, and one date-of-effect cut at the minimum put only 0.13% of EOIs on the wrong side. The unit-group queue is a pattern, not a law: where occupations in one unit group were published at different minimum scores, 48 of 52 cases fit one shared queue, 32 of 33 in the three rounds of 2025–26 and 16 of 19 in November 2024.

## What was tested, and with what data?

The Australian Government's public SkillSelect EOI data (https://api.dynamic.reports.employment.gov.au/anonap/extensions/hSKLS02_SkillSelect_EOI_Data/hSKLS02_SkillSelect_EOI_Data.html) keeps a status history for each EOI over a rolling 24 months. From it the queue was rebuilt on the day of four rounds: November 2024, August 2025, November 2025 and June 2026, and compared with the minimum scores Home Affairs published. The full method is in the paper: https://github.com/Evan-Kun/skillselect-189-sugar

## Did higher points always go first?

Almost always. Of 25,944 eligible EOIs that scored above their occupation's official minimum, 25,933 were invited (99.96%).

## Did date of effect break every tie?

Yes. At each occupation's minimum score, a single date-of-effect cut separates invited from uninvited EOIs with 36 of 27,197 on the wrong side (0.13%).

## Do occupations in a unit group share one queue?

Unit groups that fit one shared queue / unit groups with two or more invited occupations in that round (tolerance: five EOIs):

| Round | Same minimum | 5 points apart | 10+ points apart | Minimums differ | All |
| --- | ---: | ---: | ---: | ---: | ---: |
| November 2024 | 21 / 21 | 11 / 11 | 5 / 8 | 16 / 19 | 37 / 40 |
| August 2025 | 21 / 21 | 3 / 3 | 4 / 5 | 7 / 8 | 28 / 29 |
| November 2025 | 23 / 23 | 6 / 6 | 6 / 6 | 12 / 12 | 35 / 35 |
| June 2026 | 21 / 21 | 8 / 8 | 5 / 5 | 13 / 13 | 34 / 34 |
| All four | 86 / 86 | 28 / 28 | 20 / 24 | 48 / 52 | 134 / 138 |

The 86 cases with identical minimums cannot tell one queue from separate queues. The evidence is the 52 cases where minimums differed, of which 48 fit. Given each group's number of invitations, one group queue reproduces 96% of published minimums exactly; four ways of splitting that number between occupations reach 56–77%.

## Where did the pattern fail?

- November 2024: accountants, civil engineers and software programmers. Accountant (General) and Management Accountant at 95 against Taxation Accountant at 85; Civil Engineer at 85 against four siblings at 70; Software Engineer and Developer Programmer at 95 against two siblings at 85.
- August 2025: psychologists. The published minimum rests on an invitation the rebuilt data does not contain, so it cannot be settled.

November 2024 is also the least complete of the four reconstructions, and it falls in a program year without published ceilings.

## What about the 15 October 2026 round?

It is not part of the test above; so far only the EOI pool before and after the round has been compared, which is a weaker check. On that check most unit groups with more than one invited occupation are consistent with one queue, with two clear exceptions: psychologists, and electricians (Lift Mechanic was held at a higher score than its siblings).

## What does this not prove?

- It is not the Department's procedure; Home Affairs has not said over which set of EOIs the ranking is applied.
- It is four rounds, three of them in one program year. The unit-group pattern is a finding about 2025–26, and a future round may differ.
- The data has noise, bounded by a tolerance of five EOIs rather than removed.
- The tolerance and the exact definition of date of effect were fixed after looking at the same four rounds.

## What is SUGAR?

SUGAR (Sector and Unit-Group Aware Ranking) is the model built on this finding: given the size of a round, it ranks each unit group as one queue and places each occupation's cut-off. Using only information available before each of 5 rounds, its cut-offs were off by 3.45 points on average, and a simpler method that reuses the last round's cut-offs did better. Once the round's actual size and the ceiling table are known, it was exact for 63% of cut-offs, with an average error of 2.38 points, ahead of the simpler methods. Five rounds are too few to call either difference conclusive, and the model was designed on these same rounds. A prediction is a reference, not a promise.

The four rules explained, with a one-minute video: https://immitrend.com.au/blog/how-189-invitations-are-ranked. 189 prediction: https://immitrend.com.au/189-prediction

## FAQ

### Does SkillSelect really invite the highest points first for the 189?

In the four rounds tested (November 2024, August 2025, November 2025 and June 2026), 25,933 of 25,944 eligible EOIs that scored above their occupation's official minimum were invited, which is 99.96%. At the minimum score, one date-of-effect cut separates invited from uninvited EOIs with 36 of 27,197 on the wrong side (0.13%). Both results match the rule Home Affairs publishes: points first, then date of effect.

### Is it proven that occupations in the same unit group share one 189 queue?

No. It is a pattern in past rounds, not a published rule. Where occupations in one unit group were published at different minimum scores, 48 of 52 cases across four rounds fit one shared queue, including 32 of 33 in the three rounds of 2025-26. Home Affairs has not said over which set of EOIs the ranking is applied, so the finding describes outcomes rather than the Department's procedure, and a future round may differ.

### Which 189 rounds did not fit the unit-group pattern?

Four cases did not fit. Three are in November 2024: accountants, civil engineers and software programmers, where the largest occupation was held at a higher score than its siblings in the same unit group. The fourth is psychologists in August 2025, where the published minimum rests on an invitation that the rebuilt data does not contain. November 2024 is also the least complete of the four reconstructions.

### What is the SUGAR model for 189 invitations?

SUGAR (Sector and Unit-Group Aware Ranking) is a model that, given the size of a 189 invitation round, ranks each ANZSCO unit group as one queue and places each occupation's cut-off. In a backtest of five rounds using only information available beforehand, its cut-offs were off by 3.45 points on average, and a simpler method that reuses the last round's cut-offs did better. Once the round's actual size and the ceiling table are known, it was exact for 63% of cut-offs, with an average error of 2.38 points, ahead of the simpler methods. Five rounds are too few to call either difference conclusive.

Figures are from the paper and its released data (https://github.com/Evan-Kun/skillselect-189-sugar). General information only, not migration advice.