AusIndustry does not prescribe a record format. It accepts any ordinary business record, created while the R&D was happening, that shows the activity met the core R&D activity test in s 355-25 ITAA 1997.
Registration is self-assessed and generally accepted on lodgement, so your evidence is usually tested only if the claim is later reviewed. Keep it for at least five years.
What AusIndustry actually asks for
At registration, AusIndustry asks for a description, not documents. You do not attach evidence to the application. The registration form (the online customer portal version released 15 August 2025) is a field-by-field questionnaire structured as Projects, then Core activities, then Supporting activities. Background sits on the R&D Tax Incentive program page.
Each core activity is split across separate fields with their own character minimums: the sources you investigated, why a competent professional could not have known the outcome, the hypothesis, the experiment, the evaluation, the conclusions, and the new knowledge generated.
That structure is the clue. Every field is a question your records have to be able to answer if the claim is examined. If you cannot point to a dated document behind a field, the field is an assertion.
Registration must be lodged within 10 months of the end of the income year, which is 30 April for a 30 June year end. That is a hard statutory deadline. It applies to the narrative, not the evidence; the evidence should already exist.
Evidence mapped to each limb of the eligibility test
AusIndustry assesses evidence limb by limb: proof the outcome was unknown before you started, a dated hypothesis formed before experiments began, observation records generated during the work, and evaluation records at the end of each cycle. The table below maps each requirement of s 355-25(1) to the records that prove it, and to when each record must have existed.
It splits the systematic-progression limb into its stages and adds the supporting-activity and expenditure rows. Those last two sit outside the four core criteria, but they fail claims just as often.
| Requirement | Evidence that proves it | When it must exist |
|---|---|---|
| Outcome could not be known in advance | Literature and patent search results, vendor and API documentation reviewed, expert emails, notes on prior internal attempts that failed | Before the activity starts |
| Purpose of generating new knowledge | Project plans, meeting minutes or recordings, expenditure approvals stating the R&D purpose | At planning or start |
| Hypothesis | A dated note or experimental plan stating the hypothesis and its scientific or technical basis | Before experiments begin |
| Experiment | Test protocols and experimental plans specifying variables varied, held constant and measured; CRO or research contracts | Before or during |
| Observation | Raw data logs, test reports, timestamped issue-tracker entries, run sheets, build and CI output | During the activity |
| Evaluation and conclusions | Analysis notes, test reports with results and their interpretation, records of negative results and the next hypothesis | During or after each cycle |
| Supporting activity nexus (s 355-30) | Plans showing the work was directly related to a named core activity, plus dominant-purpose reasoning where relevant | Ongoing |
| Expenditure nexus (ATO) | Timesheets, payroll data, invoices, cost allocation tables, apportionment method | Ongoing |
The first row is the one most often missing. A prior-art or literature search dated after the experiment finished does not show that you faced an unknown outcome when you started.
Records AusIndustry commonly accepts
There is no approved list, but reviews consistently accept the same categories of ordinary business documentation. What follows is the practical set for a software or health-tech claimant.
Planning and hypothesis records
Anything dated that states what you were trying to find out and why the answer was not already available: a technical design document, a research plan, an ethics or trial protocol, a Confluence page, a Slack thread, an email to a technical advisor.
The hypothesis needs to be scientific and testable. In GQHC [2024] AATA 409, vague or commercial hypotheses failed because they could not be validated or invalidated. "Reduce query latency below 200ms at 10,000 concurrent sessions by sharding on tenant ID" is a hypothesis. "Build a faster platform" is a business goal.
Active Sports Management Pty Ltd and Industry Innovation and Science Australia [2023] AATA 4078 put it plainly: the contemporaneous formation of a hypothesis is the essential starting point of the activities. A hypothesis recalled at registration time is not a starting point. How to document R&D activities for AusIndustry covers how to phrase a testable hypothesis.
Experiment and observation records
These are usually the strongest records a technical team already has, because they are generated automatically.
- Commit histories, branch names and pull request discussions tied to an experiment
- Jira or Linear tickets with dated status changes and technical commentary
- CI/CD logs, benchmark output, load test results, model training runs and metrics
- Lab notebooks, run sheets, raw data exports, instrument logs
- Trial data captures, screening logs, adverse event records
The requirement is that the record shows variables being varied and measured, not just work being delivered. A ticket that says "implement caching layer" proves delivery. A ticket that says "tested three eviction policies, LRU held 94% hit rate under burst load, LFU collapsed to 61%" proves observation. More on this in using GitHub commits as R&D evidence.
Evaluation and conclusion records
Observation without evaluation is a recurring failure. Records here include analysis notes, cycle-end review documents written when each experimental cycle concludes (these are contemporaneous, because they are created alongside the cycle they evaluate), results decks presented internally, and statistical analysis outputs.
Keep the negative results. An experiment that disproved the hypothesis and led to a revised one is textbook evidence of a systematic progression. Claims that only document successes look like development, not experimentation.
Expenditure and time records (the ATO side)
AusIndustry decides whether the activity is eligible. The ATO decides whether the expenditure is claimable, and it needs a nexus from every dollar to a specific registered activity.
That means timesheets or time logs allocating hours to an activity, payroll data, contractor invoices identifying the R&D work, and a documented apportionment method separating R&D from non-R&D and core from supporting. Amounts owed to associates must be paid in cash by year end, not merely accrued.
The ATO cites the Tier Toys and Ozone Manufacturing decisions on exactly this point: without contemporaneous records, the taxpayer could not show the expenditure was anything more than normal business expense. The activity may be eligible and the claim can still fail on this row alone. See the ATO's R&D tax incentive guidance for the expenditure side.
What does "contemporaneous" actually mean?
Contemporaneous means the record was created at the time the R&D was conducted, by the people doing it, as part of running the work. It does not mean formal, and it does not mean the record has to use statutory language.
A Slack message timestamped mid-experiment is contemporaneous. A polished technical memo written in April to support an April registration is not, no matter how accurate it is.
Absolute Vision Technologies [2022] AATA 2319 is the clearest authority. Documents prepared after the fact for an internal review were not treated as contemporaneous, there was marked discord between the records and the activities as registered, and amending a deficient registration with documents prepared well after the income year was not accepted.
The registration narrative itself is never evidence. It is a description that evidence must support.
For a full walkthrough of building contemporaneous records into your workflow, see contemporaneous evidence for a defensible claim.
Evidence that carries little weight
Some material feels substantial and does almost nothing under review.
- Marketing material, pitch decks and product roadmaps: these describe commercial intent, not experimentation
- Invoices alone, with no link to a named activity
- Consultant-drafted narratives with no underlying source records
- Statements that "the outcome was unknown to our team": the test in s 355-25(1)(a) is measured against a competent professional with access to knowledge available anywhere in the world, so new-to-us is never sufficient
- Undated documents, or documents whose only date is the file's last-modified timestamp after an export
- Screenshots reconstructed at claim time
Royal Wins Pty Ltd [2020] AATA 4320 found that where there is no contemporaneous documentation that a hypothesis was developed and tested, the activities are not eligible regardless of how much work was actually done. Volume of effort is not evidence.
Evidence for supporting activities
Supporting activities are governed by s 355-30 ITAA 1997 and need a different proof: that the work was directly related to a named core activity.
For excluded-category activities, or activities that produce goods or services, you also need to evidence dominant purpose, meaning records showing the work was undertaken predominantly to support the core activity rather than for its own commercial reason.
Practical evidence includes environment setup tickets referencing the experiment they enable, data labelling instructions written for a specific model experiment, and ethics submissions naming the trial they relate to. Group supporting evidence into broad categories per core activity rather than creating one supporting activity per ticket. The boundary is covered in core vs supporting R&D activities.
Software and health-tech claims
The evidence problem differs by where the uncertainty lives, and getting this wrong is the most common structural error in these claims.
Software: technical uncertainty against worldwide knowledge
For a software core activity, the file needs to show that a competent professional (see AusIndustry's eligibility guidance and Guide to Interpretation) could not have determined the outcome from publicly available knowledge. That is an evidence task, not a writing task.
Useful records: dated notes on the libraries, papers, vendor docs and Stack Overflow or GitHub issues you searched and why none resolved the question; architecture decision records listing rejected approaches with reasons; benchmark results at the scale that broke the known approach.
Moreton Resources Ltd v ISA [2019] FCAFC 120 confirms that applying existing technology in a new context can be a core activity where the outcome could not be known in advance and the purpose was to generate new knowledge. Coal of Queensland [2021] FCAFC 54 is the counterweight: applying known methods without a genuine unknown and a documented systematic progression still fails.
Clinical and behavioural endpoints
In health tech, the software is often routine while the genuine unknown is clinical: whether the intervention measurably improves an outcome. Both are legitimate, but they are assessed against different literature and different competent professionals, so keep clinical streams separate from build streams.
Evidence for a clinical stream looks different: a clinical literature review dated before the study, a pre-registered protocol stating the endpoint and hypothesis, ethics approval, statistical analysis plans, and de-identified outcome data.
One boundary matters. A structured study measuring a clinical or behavioural endpoint against a pre-stated hypothesis is not excluded market research, even if it runs on A/B testing infrastructure.
The discriminator is the pre-specified endpoint. The same experiment measured on conversion, churn or revenue is excluded market research under s 355-25(2)(a). Your protocol, written before the study, is what evidences which one you ran.
How long do you need to keep R&D records?
Keep R&D records for at least five years. The ATO generally measures this from when the relevant return is lodged, and longer where activities span years or a review is open. Where activities span multiple income years, keep the earlier years' records for as long as they support a later claim.
If AusIndustry or the ATO opens a review, retain everything until the matter is finalised, including exports from the underlying systems, not just the summary documents.
Practical risk: engineering tooling gets migrated. Ticket histories, CI logs and chat archives disappear when a company moves from Jira to Linear or changes Slack plans, and exports made at migration time are far more credible than reconstructions made three years later.
What happens if your evidence does not hold up?
An unsuccessful review means the activity is found ineligible, the associated expenditure is removed from the claim, and the offset is amended. Interest and penalties can follow on the ATO side.
Worked example. A company in a tax-loss position, with aggregated turnover under $20M and a 25% company tax rate, claims the refundable offset at 43.5% (the 25% rate plus 18.5 percentage points). It registers $900,000 of notional R&D deductions and receives $391,500 as a cash refund.
A review then finds one core activity, carrying $300,000 of that expenditure, had no contemporaneous hypothesis or literature search. The eligible base drops to $600,000, the offset to $261,000, and $130,500 becomes repayable, before any interest.
That is the arithmetic behind every record keeping decision. The through-line across Mount Owen [2013] AATA 573, Royal Wins, Absolute Vision and GQHC is identical: a hypothesis that was really a commercial goal, no documented worldwide knowledge gap, missing observation and evaluation, and records that were retrospective or did not match the registration. For what prompts a look in the first place, see what triggers ATO R&D reviews.
Building the record as you go
The strongest evidence file is not written for AusIndustry at all. It is the ordinary output of a team that plans experiments before running them, records what it observed, and writes down what it concluded, including when the answer was no.
Rand builds the registration from that engineering and clinical evidence directly, tying each core activity to dated signals and time logs so the narrative and the records match before anyone reads them. If you want a structured way to close the gaps, check your file against the table above, section by section.
This article explains the rules; it is not tax advice for your circumstances.