Correction, 26 August 2026. This story compared the four editions of the workbook we had when it was written, April to July 2026, and said so. All nine editions published since the programme began are now available, back to November 2025, and two statements below do not survive the fuller series. First, commencements have been revised once: November 2025 was first published at 4,160 and republished at 8,420 in the very next edition, where it has stayed. The claim that commencements never change holds for every month except the programme’s first, which predates the editions this story compared. Second, the revisions do trickle. Across nine editions every settled month’s exit figure falls progressively rather than in one step, taking several editions to settle, so the single jump described below is part of a continuous pattern rather than a distinct event. The central finding is unchanged and is in fact stronger: every one of the eight settled months has been revised down, none up, none unchanged, by between 2.5 and 17.2 per cent. Referrals also fall every month, by under half a per cent. The original wording is corrected in place below and the figures it reported are unchanged.
Inclusive Employment Australia replaced Disability Employment Services on 1 November 2025, and the Department of Social Services has published a monthly data workbook ever since. We read the first eight in August. The ninth, covering 31 July 2026, was published on data.gov.au on 11 August.
The headline is straightforward. The caseload reached 254,840, up 1,470 on June. That is faster growth than either of the two months before it, which had run at 1,045 and 965, so a slowdown that had looked settled did not continue.
That is the sort of sentence a monthly release invites you to write, and we nearly wrote a more dramatic one. July also recorded 5,340 exits, the highest in the program’s life. Before publishing that, we checked it against the older editions we still had on disk. It turns out that every previous month’s record exit figure has since been revised away.
Four editions, five series, one that moves
Each edition of the workbook republishes the whole series back to November 2025, not just the new month. So the same figure can be read four times, and any change between readings is the department revising itself. We compared the editions for April, May, June and July.
| Month | Apr edition (12 May) | May edition (12 Jun) | Jun edition (3 Jul) | Jul edition (11 Aug) | Change |
|---|---|---|---|---|---|
| Nov 2025 | 1,845 | 1,845 | 1,845 | 1,845 | none |
| Dec 2025 | 2,490 | 2,490 | 2,490 | 2,490 | none |
| Jan 2026 | 4,290 | 4,290 | 4,290 | 4,290 | none |
| Feb 2026 | 4,415 | 4,415 | 4,365 | 4,365 | −50 (−1.1%) |
| Mar 2026 | 4,730 | 4,730 | 4,570 | 4,570 | −160 (−3.4%) |
| Apr 2026 | 4,775 | 4,775 | 4,375 | 4,345 | −430 (−9.0%) |
| May 2026 | n/a | 5,270 | 5,095 | 5,025 | −245 (−4.6%) |
| Jun 2026 | n/a | n/a | 5,160 | 5,030 | −130 (−2.5%) |
| Jul 2026 | n/a | n/a | n/a | 5,340 | first published |
Source: IEA monthly profile, flow data tab, four editions, each downloaded from data.gov.au and read directly on 18 August 2026. Change is from the figure’s first publication to its most recent. Published counts are rounded to the nearest 5.
Every revision runs the same way. Not one exit figure has ever been revised upwards.
The other four published series behave completely differently. The caseload total is identical in all four editions, for every month. So is the count of commencements. Referrals move by 5 to 25 in a month, which is the scale of the department’s own rounding. Only exits move by hundreds.
Across these four editions the revisions arrived in one step
A count that settles as late paperwork arrives would drift a little each month. This did not. Between the April and May editions, nothing changed at all: February, March and April were republished with exactly the figures they had first carried. Then the June edition, published on 3 July, revised every month back to February in a single step, April by 400 in one go. The July edition then trimmed a further 30 from April and 70 from May, and took 130 off June’s first published figure.
So there is a distinct event in the June edition, affecting the whole back series, and a smaller ongoing settling on top of it.
Nothing in the workbook mentions it
These workbooks carry a change log. It is reasonably diligent: it records when the outcomes tabs were added, when rounding to the nearest 5 was applied across the report, and even a change to how a bar chart handles negative values. We compared the information tab of the June edition against the May edition line by line. The two are identical. Not one line was added to record that the exit series had just been rewritten back to February.
The same is true of the July edition against June.
What we are not saying
We do not know why the figures changed, and we are not asserting a cause. Employment data
is administrative: it comes from claims and records lodged by providers in an IT system, and
late, corrected or withdrawn records are entirely ordinary. A department finding an error and
fixing it is a department doing its job. The workbook also states plainly that its exit count
excludes participants referred but exited prior to commencement
, so the
published figure is a defined subset of departures rather than everyone who left.
The criticism here is narrow and, we think, fair: the revision is not disclosed. A reader who quoted April’s exit figure in May was quoting a number that is now 9 per cent lower, and nothing in the current workbook would tell them that. The department clearly has the habit of writing change notes. This change did not get one.
The bias runs against the program, not for it
The obvious suspicion when a government revises its own numbers is that the revisions flatter. Here they do the opposite, and it is worth saying so.
Exits are revised down, so the newest month always overstates how many people left. Outcomes move the other way. The 26-week outcome count for May was first published as 1,295 and now stands at 1,350; June’s was first published as 1,305 and now stands at 1,370. Both up about 4 to 5 per cent. That direction has an innocent and stated explanation: the workbook says outcome counts are based on when payments for outcomes were created, so claims lodged after an edition closes land in a later one.
Put together, the freshest month in any edition shows more departures and fewer successes than the settled data will. The most recent month, the one most likely to be reported, is the one that makes the program look worst.
Which brings us back to July’s 5,340 exits. On this record, expect that number to fall. We will report where it lands.
The rest of the July release
The caseload of 254,840 is 8,980 higher than the first monthly figure in November, a rise of 3.7 per cent across the eight monthly steps since. July brought 8,290 referrals and 7,015 commencements. Exits have risen steadily from 1,845 in November, but that comparison flatters the early months rather than indicting the later ones: in November the program was days old, and a participant must commence before they can be counted as exiting, so a low first-month figure was arithmetically inevitable rather than a sign of stability.
One movement inside the caseload is worth watching. First Nations participants rose 5.1 per cent over that same span against 3.7 per cent for the caseload as a whole, while participants recorded as homeless rose just 1.3 per cent. Both are small shifts over a short window, and we draw no conclusion from them yet, but they are the two lines moving away from the average.
Our view
Publishing the full back series in every edition is good practice, and it is the only reason any of this is visible. Most published statistics do not let you check them against themselves.
Having done that work, the department should finish it. A revision note costs one line in a tab that already exists and already carries notes of far less consequence. Without it, the workbook quietly asks every reader to assume that a figure they wrote down from an earlier edition is still the figure, and for one series in five, it is not.
How we sourced this
We downloaded four editions of the Inclusive Employment Australia monthly profile directly from the data.gov.au dataset on 18 August 2026: the editions covering 30 April, 31 May, 30 June and 31 July 2026. Publication dates are the dataset’s own resource timestamps. Every figure above was read from the workbooks themselves, from the flow data, caseload data and outcomes data tabs, and each comparison was made by reading the same month from each edition. The information tab comparison was a line-by-line comparison of the text content of that tab between editions.
What we have not done. We have not established why any figure changed, and we have not put that question to the department before publishing; we have reported only what the workbooks show. We hold four editions, not all nine, so revisions to the November 2025 to March 2026 figures before the April edition are outside what we can see. The department rounds published counts to the nearest 5 and warns that recalculating from rounded data compounds that rounding, so differences of 5 or so carry no weight; the exit revisions we describe are between 50 and 430. Percentages are our own calculation from the published counts. Nothing here is a claim about any provider, employer or participant, and we name none.
Sources
- Department of Social Services, DSS Inclusive Employment Australia (IEA) monthly profile, four XLSX editions all downloaded and read 18 August 2026: 30 April 2026 (published 12 May), 31 May 2026 (published 12 June), 30 June 2026 (published 3 July) and 31 July 2026 (published 11 August): every exit, referral, commencement, caseload and outcome figure above, the exit counting method, the note that outcome counts follow payment creation, the rounding note, and the change log used for the information tab comparison.
- Department of Social Services, Inclusive Employment Australia (program page, read 18 August 2026): that the program replaced Disability Employment Services on 1 November 2025.
Work in employment services, or know why these figures moved? Tell us and we will check it against the published data and log the outcome here.