TLDR: Technical uncertainty is the first limb of the core R&D activity test in s 355-25 of the Income Tax Assessment Act 1997: the outcome cannot be known or determined in advance from current knowledge, information or experience. The benchmark is a competent professional with worldwide accessible knowledge, not your team's skill set. Proving technical uncertainty in an R&D tax claim comes down to dated artefacts created before the experiments began: prior-art searches, rejected design options and written hypotheses.
What technical uncertainty means under Division 355
Section 355-25(1)(a) requires experimental activities "whose outcome cannot be known or determined in advance on the basis of current knowledge, information or experience, but can only be determined by applying a systematic progression of work" based on principles of established science and proceeding from hypothesis to experiment, observation and evaluation, leading to logical conclusions.
It is a statutory condition, not a description of how hard the work felt.
Two consequences of the s 355-25(1)(a) wording
First, uncertainty and method are welded together. If the outcome was unknown but you resolved it by trial and error rather than a systematic progression, you fail.
The Tribunal in Active Sports Management [2023] AATA 4078 (appeal dismissed, [2024] FCA 1346) found no systematic progression where there were no documents showing how hypotheses were formed, tested and recorded. The contemporaneous formation of a hypothesis is the starting point, and trial and error without isolable variables is not a repeatable experiment.
Second, limb (b) still applies. The activity must also be conducted for the purpose of generating new knowledge, including new or improved materials, products, devices, processes or services. Uncertainty alone does not carry an activity across the line.
Uncertainty is assessed per activity
Eligibility is self-assessed per activity, never per project or per product. A single release can contain one genuinely uncertain core activity and a great deal of routine build that is either supporting or not claimable at all. See core vs supporting R&D activities for how that split works.
The competent professional test
The reference point is a hypothetical competent professional working in the relevant field who has access to knowledge that is publicly available or reasonably accessible anywhere in the world. If that person could have determined the outcome in advance, there is no eligible uncertainty. Effort is not the measure.
The business.gov.au software development sector guide makes the same point: a core activity arises only where no existing knowledge, method or solution can resolve the hurdle, even for experienced professionals, and work that is commercially novel or technically complex is often still routine development.
In a review, expect to be asked what search would have answered the question and whether you ran it.
Why "new to us" fails
Novelty to the company is never sufficient. The following do not, on their own, create eligible uncertainty:
- Adopting a framework, cloud service or database the team has not used before.
- Migrating a working system to a new platform using documented patterns.
- Integrating a third-party API according to its published specification.
- Rebuilding a competitor's feature from its observable behaviour, which is also likely to fall within the exclusion in s 355-25(2)(g) ITAA 1997 for reproducing a commercial product or process.
- Configuring, tuning or customising off-the-shelf software within its documented parameters.
Two Full Court decisions mark the edges of the test. Coal of Queensland Pty Ltd v ISA [2021] FCAFC 54 (special leave refused) narrows it: the Court held that an activity's outcome is distinct from the data it generates, so not being able to predict the precise results of a trial or benchmark does not mean the outcome could not be known in advance.
Moreton Resources Ltd v ISA [2019] FCAFC 120 widens it: "experimental activities" is read broadly, and applying an existing technology in a genuinely new context or at a new site can be a core activity where the outcome could not be known in advance and the purpose was to generate new knowledge.
Moreton is not a free pass. Novelty of context can support eligibility, but the four-part test still has to be met and evidenced.
What counts as reasonably accessible knowledge
Reasonably accessible means what a competent professional could find with reasonable effort, not what your engineers happened to know on the day.
| Treated as accessible knowledge | Not resolvable from accessible knowledge |
|---|---|
| Vendor documentation, RFCs, published standards | Behaviour of a novel combination of components at a load no published benchmark covers |
| Peer-reviewed papers and preprints | Whether a published model architecture generalises to your data distribution |
| Public patent databases | Whether a documented technique holds under constraints the literature does not address |
| Open-source implementations and issue trackers | Whether a digital intervention shifts a validated clinical endpoint in your cohort |
| Conference talks, engineering blogs, textbooks | Trade-offs where sources conflict and no source resolves the conflict |
The right column is where eligible uncertainty can live. Each entry names a specific gap between what is published and what you needed to know, and whether the gap is real in your case is what the prior-art search has to show.
What does a documented prior-art search look like?
A defensible prior-art search is a dated record showing what you looked for, where you looked, what you found, and why it did not answer the question. It should exist before or at the start of the work, because it is the main evidence that the outcome could not be known in advance.
A workable format is a short document per core activity containing:
- The question, stated as a technical or clinical proposition, not a product goal.
- Sources searched, named specifically: databases, journals, patent registers, vendor documentation, standards bodies, open-source repositories, search terms and dates.
- What was found, with citations, including the closest prior work.
- Why the gap remains, stated as the delta between the closest published knowledge and what your activity needed to establish.
- Who conducted it, with the qualifications that make them a competent professional in the field.
The question has to be a proposition, not a goal: in Absolute Vision Technologies [2022] AATA 2319, hypotheses framed as project objectives were one of the reasons the claim failed.
Keep clinical literature reviews separate from technical prior-art searches. The benchmark for each is a different competent professional, so a clinical review is judged against the clinical literature and a technical search against the engineering literature.
Where eligible uncertainty lives in practice
Division 355 does not label uncertainty by type, but in practice eligible uncertainty shows up in two places, and a third category of work only looks like it.
Technical uncertainty: can it be built, will it hold at scale
This is uncertainty about whether a technical approach will work, and it usually presents as a question about behaviour under conditions no source covers.
Examples framed correctly:
- Whether a chosen consistency model can hold sub-100ms write latency across three regions at a stated concurrency, when published benchmarks stop an order of magnitude below it.
- Whether a retrieval architecture can maintain factual precision above a stated threshold on a domain corpus with characteristics the published evaluations do not include.
- Whether a compression scheme can hold reconstruction error below a stated bound on a stated inference task, at a memory budget no published implementation targets.
Each names a measurable threshold. "We had to build a scalable platform" names none, and reads as a commercial goal.
Scientific and clinical uncertainty: does the intervention work
Uncertainty does not have to live in the engineering. For health-tech companies it frequently lives in the outcome.
Whether an app measurably improves an IBS symptom score, or whether a monitoring algorithm detects deterioration earlier than standard care: these can be genuine knowledge gaps even when the software delivering them is routine to build.
The discriminator is the pre-specified endpoint and hypothesis, not the tooling. A structured study measuring a clinical endpoint against a pre-stated hypothesis is not market research, even when it runs on A/B infrastructure. The same experiment measured on conversion, churn or revenue is excluded market research under s 355-25(2)(a).
Clearing market research does not clear s 355-25(2)(d). An experiment built to answer how people behave, form habits or respond to messages is research in social sciences whatever endpoint it carries. More on framing in describing scientific uncertainty in a clinical claim.
One trap specific to this stream: clinical work often runs through a CRO or an offshore parent, and if the hypothesis, the results and the financial risk sit with them rather than you, the R&D may not be conducted on your behalf. Taxpayer Alerts TA 2023/4 and TA 2023/5 target exactly those arrangements. See clinical trial R&DTI eligibility.
Where uncertainty is absent
Some work simply does not qualify, and drafting around it is riskier than excluding it. Routine configuration, bug fixes, UI restyling and documented integrations resolve on known knowledge, and so do version upgrades. The software development sector guide linked above covers these patterns.
Software developed for the dominant purpose of internal administration fails for a different reason: s 355-25(2)(h) excludes it from being a core activity outright.
If the honest answer to "could a competent professional have predicted this outcome?" is yes, leave the activity out. Register it as supporting work only where it is directly related to a named core activity and, where the work is excluded-category or produces goods or services, undertaken for the dominant purpose of supporting that core activity.
Evidence that proves the uncertainty existed before the work
Timestamps do the work here. The evidence must be created at the time the R&D is conducted, and it must sit before the experiments it justifies. Royal Wins Pty Ltd [2020] AATA 4320 held that without contemporaneous documentation that a hypothesis was developed and tested, activities are not eligible, and the volume of work done does not rescue them.
Contemporaneous artefacts by source system
Most software and health-tech companies already generate the evidence. The failure is that nobody labels it or maps it to an activity.
| Test element | Artefact | Where it lives | When it must exist |
|---|---|---|---|
| Outcome unknown in advance | Prior-art and literature search, expert email stating current knowledge cannot determine the outcome | Notion, Confluence, email | Before the activity starts |
| New-knowledge purpose | Project plan, meeting minutes, expenditure approval stating the R&D purpose | Project docs, calendar, finance | At planning or start |
| Hypothesis | Spike ticket, experiment plan, architecture decision record with rejected options | Jira, Linear, ADR repo | Before experiments |
| Experiment | Test protocol naming variables varied, held constant and measured; CRO or research contract | GitHub branches, protocol docs | Before or during |
| Observation | Raw benchmark output, run logs, failed build records, trial data | CI artefacts, dashboards, EDC | During |
| Evaluation and conclusions | Analysis notes, retrospectives, results write-ups including negative results | Pull request discussion, reports | During or after each cycle |
| Expenditure nexus | Timesheets, payroll, cost allocation tied to a named activity | Payroll, time tracking | Ongoing |
Rand builds the claim from these signals directly, mapping pull requests, tickets and documents to a core activity. R&D time is estimated from each person's active days with placed records and becomes a confirmed time log when a director signs the year.
The uncertainty narrative then points at records that already carry a timestamp, rather than a story written in April. More detail in using GitHub commits as R&D evidence.
The timing problem: why post-hoc narratives fail
Reconstructing uncertainty at registration time is the classic loss. Absolute Vision Technologies [2022] AATA 2319 found that documents prepared after the fact for an internal review were not contemporaneous, and that there was marked discord between the records and the activities as registered. Only the activities as originally registered can be considered; later documents that describe different work are unlikely to be accepted.
GQHC [2024] AATA 409 adds the ATO's side: following obiter in Commissioner of Taxation v Auctus Resources [2021] FCAFC 39, the Tribunal held that where IISA has made no finding the Commissioner can decide whether activities are R&D activities, and that claim failed on thin records of data, protocols and analysis. Observation and evaluation must be evidenced, not asserted.
Retain records for at least five years after the claim is made, consistent with ATO guidance on keeping records and calculating notional deductions. See also R&D record keeping requirements.
Writing the uncertainty fields on the AusIndustry registration form
AusIndustry does not want one narrative. The form splits each core activity across separate fields (sources investigated, why a competent professional could not know the outcome, hypothesis, experiment, evaluation, conclusions, new knowledge), six of which carry a character minimum.
This section covers only the fields that carry the uncertainty argument; the full field-by-field walkthrough sits in our guide to what the AusIndustry registration form asks.
The current portal version was released 15 August 2025 and is structured as projects, then core activities, then supporting activities. Draft each field to its own requirement. Pasting the same paragraph into all of them is the fastest way to look templated.
Sources investigated
Name the sources and the search, not the conclusion. List specific databases, journals, patent registers, standards, vendor documentation and repositories, with search terms and the date the search ran, plus the closest prior work found and its limitation.
Weak: "We reviewed existing literature and found no solution."
Stronger: "Searched IEEE Xplore and arXiv on 14 March 2025 for "multi-region write consistency tail latency" and "per-tenant queueing", reviewed the vendor's published limits and two open-source implementations. The closest published result reports X under conditions Y, which does not address Z, the condition our activity required."
Why a competent professional could not know the outcome
State the gap between the accessible knowledge and the outcome you needed, in the language of the section. Identify the field, what that professional would already know, the specific variable published knowledge does not resolve, and why the outcome could only be determined by experiment.
Weak: "This had never been done before."
Stronger: "A competent distributed-systems engineer with access to the published benchmarks could determine system behaviour up to 5,000 concurrent sessions. No published source addresses the interaction of per-tenant queueing with cross-region consistency at 50,000 sessions, so the outcome could only be determined by experiment."
The same structure works for a clinical stream. Name the guidelines and trials a competent clinician would rely on, then state the endpoint, population or delivery mechanism those sources do not cover.
Hypothesis, evaluation and conclusions
The hypothesis field must state a proposition that could be invalidated by the experiment, with the measurable threshold written in. As Royal Wins put it, a hypothesis is a proposition tested by experiment, not an objective.
Weak: "We hypothesised the platform would scale."
Stronger: "We hypothesised that a queue-per-tenant design would hold p99 write latency under 100ms at 50,000 concurrent sessions."
The evaluation and conclusion fields then report what actually happened against that threshold, including results that invalidated the hypothesis and what you decided to test next.
What a failed experiment is worth
Failure does not reduce eligibility: entitlement turns on whether the activity met s 355-25, not on whether it succeeded commercially. An invalidated hypothesis is often the cleanest evidence that the outcome could not be known in advance.
Suppose the queue-per-tenant experiment above held the latency threshold to roughly 12,000 concurrent sessions and then broke on tail latency.
If the hypothesis was written down before the runs, the benchmark output and CI records exist from during the work, and the analysis names the point of failure and the next design to test, the activity has evidence for every limb of s 355-25(1)(a), from hypothesis through to conclusion.
The new knowledge is that the design does not hold above roughly 12,000 sessions, and why. That is a result a competent professional could not have read off any published source, which is exactly what the section asks for.
The failure is not what sinks the activity. The missing dated threshold before the runs began is. An activity that cannot evidence its uncertainty should come out of the registration, along with every hour and dollar attached to it.
A practical checklist before you register
Run each proposed core activity through these questions and keep the answers with the file. How each one resolves depends on your circumstances; this is a test of evidence, not a ruling on eligibility.
- Can you name the specific outcome that could not be determined in advance, in measurable terms?
- Can you name the sources searched and the closest published knowledge, with dates?
- Does a dated artefact state the hypothesis before the experiments ran?
- Do observation records exist from during the work, not after it?
- Do the conclusions record what was learned, including negative results?
- Does the activity avoid the s 355-25(2) exclusions, particularly market research, social sciences, internal administration software and reverse engineering?
- Do time records tie hours to this named activity, not to the project generally?
ATO and AusIndustry reviews probe the same points, and further context sits in the business.gov.au guidance on assessing if your R&D activities are eligible.
Anything you cannot answer is a review finding waiting to happen, and an activity that cannot answer question 3 should not be registered at all.
