One in ten current qualifications lists a deleted unit
the register publishes no replacement for

What exactly is being counted

A qualification on training.gov.au publishes a unit grid: the list of units of competency it is made of, most marked core or elective, each one carrying the register's own usage recommendation, so the grid itself already says when a unit is deleted. Every unit also has its own record on the same register, with a status and a list of the products that supersede it. What no page joins up is the two readings against each other: a grid still listing a unit, against a record with nowhere to go.

A pair is counted here only when all four of these are true at once:

• the qualification's status on the register is current;
• the grid we read is the release the register publishes first for it, and the release number and date travel with the finding;
• that grid marks the unit as not current; and
• the unit's own record publishes a status that is not current and an empty list of what supersedes it.

If any one of the four is missing there is no finding, and the unit is declared as unread rather than counted; a missing field is never read as an empty one. Across the census and its re-runs, 0 units went unread and 0 qualifications showed the opposite contradiction (a grid killing a unit whose own record still says current). The zeros are worth as much as the findings: they say the 214 are not the residue of a partial read.

What this is not. It is not a breach count, and it moves no clock. The item 14 enrolment gate is arithmetic on a replacement's inclusion date, and here the register publishes no replacement, so there is no such date to compute. It is also not a claim about the world: the register publishing nothing is not the same as nothing existing, which is exactly what the worked example below shows.

The count

The proportion worth holding on to is the one in the middle. Counted as grid entries (product-unit pairs, so a dead unit sitting in fifteen grids counts fifteen times), current qualifications no longer recommend 3,708 of them, and 3,494 have somewhere to go. 5.8% do not. The ordinary case is a supersession with a published chain, and an RTO can work through it. These 214 are the remainder that no amount of chain-following resolves, because there is no published chain to follow.

Skill sets are the quiet side of this: six findings in 1,622 products. The weight is almost entirely in qualifications, which is also where the enrolments are.

When the unit is compulsory

Most of the 214 pairs are elective entries. An elective has a workaround inside the qualification's own packaging rules: choose a different elective. Core is different. The register's grid marks the unit compulsory for every learner in that qualification, so there is nothing to swap it for.

There are seven such pairs, six distinct units across six current qualifications. All seven, with the release of the grid each one was read from:

The same unit can be core in one qualification and elective in another, and that is not a detail: it is the flag of the pair, not of the unit. TAEASS401 is listed by 15 current qualifications and marked core by two of them. AURMGA101 is core in AUR21020 and elective in AUR30920. Reporting a unit as “core” because one product marks it that way overstates the finding across every other product that lists it. We made exactly that mistake in an internal draft before this was published, and the corrected count is the one above.

The worked example: TAEASS401

TAEASS401 Plan assessment activities and processes is the single most-listed of the 143. Here is what the register publishes on its own record, read 1 September 2026:

status Deleted · currency period ends 2022-12-06 · usage recommendation Deleted from 2022-12-07 · superseded by: an empty list · its only mapping points backwards, to TAEASS401B, flagged not equivalent (“Updated to meet Standards for Training Packages. Addition of new element.”, dated 2016-04-05)

training.gov.au, record for TAEASS401

And the same register lists it in the published grid of 15 current qualifications and one current skill set. In two of those qualifications the grid marks it core: AVI50419 and MEA60322. In the other 13 it is an elective. All 15 are current: AVI50419, HLT50221, MEA60322, PMA40116, PMA50116, PMA60116, PPM40121, RGR40419, RGR50118, RII40120, RII40320, RII40520, RII40920, SIS40621, SIS50421.

Now the part that decides how this gets written. On 2022-12-07 the register moved TAEASS401 and two units alongside it, in the same training package release. Two of them — TAEASS402 Assess competence and TAEASS403 Participate in assessment validation — the register superseded, and published TAEASS412 and TAEASS413 as their replacements, current from 2022-12-08. The third, TAEASS401, it deleted with an empty replacement list. Same package, same day: two units with a published chain to follow, one without.

That gap is the whole practical problem. An RTO reading its own scope mechanically (follow what the register says supersedes each dead unit) gets nothing here, because the register says nothing. And we will not fill the silence for you: we are not going to name the unit that “really” replaces TAEASS401, because the register does not name one and we have not verified one. That is why we write it publishes no replacement and never there is no replacement: the first is what we read; the second we do not know, and neither statement licenses us to invent the substitute.

How long this has been the case

These are not last week's deletions catching a grid mid-update. Grouped by the year the register recorded each of the 143 units as deleted:

The oldest deletion here goes back 17 years: AUM4003A, recorded deleted on 2008-10-05, and still listed today by four current qualifications (MSA30208, MSA40108, MSA50108, MSA60108). The heaviest single year is the most recent one: 2026 alone accounts for 48 of the 143. One reading cannot say whether the backlog is growing or shrinking: that needs two, and the second one is a repeat of this census. What it does say is that the spread runs from 2008 to 2026: these are not all one training package's bad week.

Where it concentrates

It is not spread evenly. The 214 pairs sit in 114 qualifications, an average of 1.9 each, and the heaviest carries 12:

And from the other direction, the units that turn up in the most current grids:

The six skill set findings are short enough to publish whole: ICTICT804 in TLISS00191 Establish Digital Supply Chain Skill Set (core); PSPGEN083 in PUASS00093 Biosecurity Response Public Information Management (core); LGALAND503A in AHCSS00079 Carbon Farming Services Advisor (CFSA) Skill Set; TAEASS401 in PMASS00015 Workplace assessor, trainer, coach; SIFXMGT002 in RGRSS00065 Racing Stable Administration Skill Set; RIISAM315A in RIISS00053 Advanced off-the-road and earthmoving tyre fitting.

What an RTO actually does with this

The honest answer starts with what we will not tell you: which unit to put in the gap. The register publishes no replacement, and choosing a substitute is a packaging-rules decision that belongs to the RTO and, where it matters, to its regulator. Anyone selling you the swap is selling you their opinion.

What is actually available to act on:

• Know which ones you hold. The finding lives at the level of a product on your scope, so the first question is whether any of the 114 qualifications or six skill sets is on yours. That is a lookup, not a judgement.
• Separate elective from core before anything else. An elective entry is usually absorbed by the packaging rules; a core entry is not, and only seven pairs are in that state. Treating all 214 as equally urgent wastes the attention the seven deserve.
• Check the enrolments, not just the scope. The question that has a person attached to it is whether anyone is enrolled in, or scheduled for, that unit right now.
• Write down what you decided and why, with the register printout attached. The value of this finding is that it is documented on the regulator's own register: whatever an RTO does next, it can show the two statements it was reading.
• Take it to your regulator when it is core. There is no mechanical answer, and no page on the internet (this one included) can give you one.

For the surrounding rules: what the instrument says about deleted products, and about superseded ones, is set out in there is no teach-out period for a superseded qualification. Whether and how that reads onto a deleted unit sitting inside a current qualification's grid is a question for you and your regulator; the register does not answer it, and we do not either.

Method — so you can reproduce it

Three queries against the register's public API, no account, no key. Everything below ran from Node on 1 September 2026, and each pass carries its finishing stamp: the census of qualifications 2026-09-01T19:02:18Z, skill sets 2026-09-01T19:08:46Z, the second-pass re-read 2026-09-01T19:12:29Z, and the full re-run of the qualification census 2026-09-01T21:24:48Z.

• The population is read, never typed. GET /api/search/training?pageSize=100&offset=N&orderBy=Code%20asc&includeTotalCount=true&filter=Type/Id%20eq%202%20and%20Status/Id%20eq%200. Type 2 is qualification, type 8 is skill set, status 0 is current. The register declares totalCount 1,164 and 1,622, and that is the denominator used here. Note offset, not skip: with skip the register answers 200 and returns page one forever.
• The grid, per product. GET /api/training/{CODE} for the first published release id, then GET /api/training/{CODE}/releases/{id}/unitgrid, which gives each unit's code, isEssential / isEssentialLabel (core or elective) and usageRecommendation.
• The unit's own record, in batches of fifteen: GET /api/search/training?filter=Code eq 'A' or Code eq 'B'… returning status, currencyPeriod.endDate, usageRecommendation and supersededBy. A finding needs status.isCurrent === false and supersededBy present and empty; an absent field is unread, never empty.
• Same code as the product. The census calls the identical function that renders the flag on our product pages, so the published count and what a reader sees on a page cannot drift apart. Two commands, 4.6 minutes and 6.4 minutes: node scripts/censo-incoherencias.js and node scripts/censo-incoherencias.js --tipo 8.
• Read again, and the corrections are listed. A second pass at 2026-09-01T19:12:29Z re-read, by a separate route, the population as the register declares it, the unit records behind the findings and the grids behind every figure on this page. Then the whole qualification census was run again end to end, finishing 2026-09-01T21:24:48Z, and reproduced every count in the table above: population, 1,150 grids, 3,708 entries no longer recommended, 214 pairs, 143 units, seven core pairs, 0 disagreements the other way and 0 unread. Reading the draft adversarially against that data changed three things, and they are worth naming: the core count is per pair, so it is seven pairs over six units, and not the five the first pass reported; the MEM31219 grid is release 2, not release 1 as first written; and a draft of this article named a current unit as covering TAEASS401's ground, which the register does not say and we had not verified. It is gone, and nothing replaced it but the contrast above.

Known limits

• It is a snapshot. The register can publish a replacement, or update a grid, the day after we read it, which is why every figure here carries the day it was read. A repeat run is the only way to know whether a given pair still stands.
• First published release only. We read the release the register publishes first for each product, and the release number and date travel with each finding above. An earlier release of the same product may list different units.
• Not every qualification publishes a grid. 14 of the 1,164 do not, and they sit outside the denominator: they are neither clean nor findings, they are unread by definition.
• It says nothing about the world. An empty replacement list is a fact about the register's publishing, not proof that no successor unit exists somewhere in a current training package. We did not go looking for one, and we do not name one.
• Some pairs carry no core/elective flag at all. For seven of the 214 pairs (three units) the register publishes no isEssential, so they can be neither counted core nor described as elective. The seven core pairs are therefore a floor, not a ceiling.
• It is not a compliance finding. We read the register and count. We do not assess anyone against the Standards, and nothing here says an RTO is in breach.

Frequently asked

Do current qualifications on training.gov.au list units that have been deleted?
Yes. Read on 1 September 2026: of the 1,150 current qualifications whose unit grid the register publishes, 114 — 9.9% — list at least one unit whose own record on the same register says it is not current and publishes an empty replacement list. That is 214 product-unit pairs and 143 distinct units. Both statements are the register's own.

Does “no replacement published” mean no replacement exists?
No, and the difference matters. It means the register's supersededBy list for that unit is empty, so there is nothing to follow — not that nothing exists anywhere. We do not name a substitute, because the register does not and we have not verified one. What the register does show is the contrast: on 2022-12-07 the same training package release superseded TAEASS402 and TAEASS403 and published TAEASS412 and TAEASS413 as their replacements, and on the same date it deleted TAEASS401 with an empty replacement list. Same package, same day: two units with a published chain to follow, one without.

What happens when the deleted unit is core?
Core means the register's own grid marks it compulsory for every learner in that qualification, so there is no elective to swap. We measured seven such product-unit pairs across six current qualifications, involving six distinct units. What to do about it is a decision for the RTO and its regulator; the register does not answer it, and neither do we.

How do I check my own qualification for this?
Open the qualification on training.gov.au, go to its Units of competency tab, and note every unit the grid does not mark as current. Then open each of those units' own records and look at two fields: the status, and the list of what supersedes it. A unit that is not current with an empty replacement list is one of these. Register Radar runs the same reading on every product page on this site, and inside the paid transition analysis. The free scope check does something different: it reads your live scope and computes the item 14 enrolment cut-off for the superseded products on it.

The rest of the unit-level picture

This is the third of three measurements that all read below qualification level, and they stack. Equivalence is decided unit by unit, not qualification by qualification, so reading only your qualification list hides most of what changed. Most providers whose scope looks clean at qualification level have superseded units and nothing else. And the dated half of the rule (when a superseded product stops taking new enrolments) is measured across the whole sector in the item 14 enrolment cut-off, measured. If you want to do any of this by hand first, the step-by-step on the public register is the place to start.

Check your own scope

Every product page on this site runs this exact check and shows the result. Two examples: RII30220, the qualification carrying the most of them, and CHC52025 Diploma of Community Services, where exactly one entry in a grid of 114 units is in this state. The register's own record for each unit is one click away from there, so nothing here has to be taken on trust: the RII30220 unit list on training.gov.au is the same grid we read.

Register Radar's free scope check reads your live scope from the register for your RTO code and computes the item 14 enrolment cut-off for the superseded products on it. It does not read unit grids: that reading is on the product pages above, one click from the list the check gives you, and inside the paid transition analysis. No account and no sign-in; how it works, stop by stop, is published. What we keep is the code you typed, how the check arrived, and a one-way fingerprint of the IP address (an HMAC, which cannot be turned back into the address) rather than the address itself, cleared after 90 days (privacy).

Sources

• training.gov.au — the National Register. The population, every unit grid and every unit record, read 1 September 2026 from its public API with the three queries set out under Method.
• The records for TAEASS401 and for its two package siblings TAEASS402 and TAEASS403, plus the replacements the register publishes for them, TAEASS412 and TAEASS413 — status, currency period, usage recommendation, replacement list and mapping information, read 1 September 2026.
• The published unit grids of AVI50419, MEA60322, AUR21020, UET30921, PSP50522, MEM31219 — the six qualifications behind the seven core pairs, each read at the release named in the table above.
• Register Radar, scripts/censo-incoherencias.js — the census, run over the 1,164 current qualifications and 1,622 current skill sets on 1 September 2026. It calls the same classification function that renders the flag on our product pages.

Measured on 1 September 2026 against the live register, then re-read the same day by a second pass (2026-09-01T19:12:29Z) and by the whole qualification census run again end to end (2026-09-01T21:24:48Z), which reproduced every figure published here; reading the draft adversarially against that data corrected three (listed under Method). Across the census and its re-runs no unit went unread, and the field is reported rather than assumed (0 in qualifications, 0 in skill sets). Everything above describes what the National Register publishes: a status, a grid, and an empty replacement list. Nothing here says a replacement does not exist, nothing here assesses any provider against the Standards, and nothing here is a finding about the regulator. The counts are ours and can be reproduced, or disputed, against the same public pages.