Will a Software-Assisted Document Authenticity Assessment Survive an Australian Court or Tribunal?

The question is not whether the software is open or closed. It is whether the software is the witness. Where an investigator forms and owns the conclusion, they are the person cross-examined, and the analysis is material they relied on. Statutory presumptions about machine output are narrow. Be ready to show what was run, on which file, under which version.

That distinction is the spine of this article, and most discussion of the topic misses it. Investigation teams evaluating document analysis tooling are usually told to worry about whether the vendor will publish its methods. The Australian and overseas authority points somewhere else entirely - at who is giving the evidence, and at whether they can account for what they relied on.

Where Does This Question Actually Land?

Very little Australian investigation output ends up in a courtroom. It ends up in front of the Fair Work Commission, an internal disciplinary decision-maker, a regulator, or a tribunal. Court is the tail risk, not the base case.

That matters, because several of those forums are expressly freed from the rules of evidence:

• The Fair Work Commission is not bound by the rules of evidence and procedure in relation to a matter before it (Fair Work Act 2009 (Cth) s 591).
• The Administrative Review Tribunal is not bound by the rules of evidence and may inform itself in such manner as it considers appropriate (Administrative Review Tribunal Act 2024 (Cth) s 52).
• An internal disciplinary process is not bound by the rules of evidence at all. It is bound by procedural fairness.

It is tempting to read that as a relaxation. It is the opposite. In many tribunals the practical contest shifts from strict admissibility to weight, fairness and the quality of your reasoning - and statutory constraints and procedural rulings still bite. A forum that is not bound by the rules of evidence is also not bound by any presumption that runs in your favour.

So the practical test for most Australian investigators is not “is this admissible”. It is “can I explain what I did, show what I relied on, and have it hold up when someone competent pushes back”. That is a harder test to pass with a black-box output and an easier one to pass with a documented human assessment.

The Fair Work Commission’s expectations of an investigation process are already about demonstrable reasoning. A document assessment is not a special case. It is one more part of the record that has to survive being examined.

What Does Australian Law Require of Machine-Produced Evidence?

Where a matter does reach a court under the uniform Evidence Acts, the position is more accommodating than practitioners expect - but the provisions are narrower than they are usually described.

Section 146: what the presumption does, and what it does not do

Section 146 of the Evidence Act 1995 (Cth), headed Evidence produced by processes, machines and other devices, applies where a party tenders a document or thing produced wholly or partly by a device or process, and asserts that in producing it the device or process produced a particular outcome.

Where the section applies, and where it is reasonably open to find that the device or process is one that, if properly used, ordinarily produces that outcome, it is presumed - unless evidence sufficient to raise doubt about the presumption is adduced - that the device or process produced that outcome on the occasion in question. The Australian Law Reform Commission’s standard illustration is the photocopier: a party need not call evidence that photocopiers ordinarily make complete copies.

Now read what that provision actually does. It assists in proving that a device or process produced its asserted ordinary outcome, for a document or thing that has been tendered. It is a proof-facilitating provision about machine output. It does not extend a presumption of reliability to analytical conclusions reached with software assistance, and it is not a statutory shortcut to “the analysis was sound”.

Three limits sit inside the section, and each is a place a challenge can begin:

Limit	What it means in practice
Applicability	The section is directed at a document or thing produced by a device or process, tendered by a party asserting the device produced a particular outcome. Whether it reaches a given piece of analysis is a question to be argued, not an assumption to be made
Ordinary outcome, properly used	The presumption depends on it being reasonably open to find that the device or process, if properly used, ordinarily produces the outcome asserted
Rebuttal	It is expressly rebuttable. Evidence sufficient to raise a doubt about the presumption displaces it

The ALRC reviewed ss 146 and 147 in Report 102 and declined to recommend a stricter threshold, finding no empirical case for one and noting that the adversarial system’s existing testing mechanisms provide the safeguard. That is a sensible position, and it is not the same thing as the statute vouching for anyone’s software.

The honest reading is the useful one. Even where a presumption of this kind is available, the party relying on an analysis should be ready to prove, positively, what was run, on which file, and under which version of the engine. That is not a fallback for when the presumption fails. It is the ordinary work of putting a software-assisted assessment in front of a decision-maker, and it is what a challenge will test first.

This is exactly why SentinelOps version-stamps every analysis and records it append-only - so an investigator can answer those questions from the case record rather than from memory, three years later.

Section 58: a narrow provision, but a useful one

Section 58 provides that if a question arises as to the relevance of a document or thing, the court may examine it and may draw any reasonable inference from it, including an inference as to its authenticity or identity. Subsection (2) confirms that this does not limit the matters from which inferences may properly be drawn.

Be precise about the scope. Section 58 operates where relevance is in issue. It permits the court to look at the artefact and reason about it. It does not by itself authenticate a document, and it does not resolve expert-evidence questions, which run through the opinion rule and its exceptions.

What it does confirm is that an Australian court can be shown the thing and reason about it directly. Carry that into practice as a habit of mind. An observation that an independent examiner can verify on the file itself is worth more than a conclusion only the software can vouch for, because the first can be put in front of a decision-maker and the second cannot.

That is a design constraint rather than a marketing one, and it is the reason SentinelOps surfaces observations for a human to assess instead of issuing a machine verdict for a human to accept.

Sections 48 and 51: the original is no longer the point

Part 2.2 of the uniform Evidence Acts (ss 47-51) governs documentary evidence. Section 48 sets out the ways the contents of a document may be proved, including by tendering a document produced by a device used to retrieve stored information. Section 51 abolishes the common law original document rule.

That is why a copy is ordinarily capable of being received, and why authenticity questions have shifted from “is this the original” to “what can be said about this file”. Only the second question is live in most modern matters.

Which Australian jurisdictions does this apply in?

This is where generalisations go wrong. The uniform evidence scheme has been adopted in most, but not all, Australian jurisdictions.

Jurisdiction	Evidence legislation	Uniform scheme?	Machine and computer evidence
Commonwealth	Evidence Act 1995 (Cth)	Yes	ss 146, 147; s 58
New South Wales	Evidence Act 1995 (NSW)	Yes	Mirrors the Commonwealth numbering
Victoria	Evidence Act 2008 (Vic)	Yes	Mirrors the Commonwealth numbering
Tasmania	Evidence Act 2001 (Tas)	Yes	Mirrors the Commonwealth numbering
Australian Capital Territory	Evidence Act 2011 (ACT)	Yes	Mirrors the Commonwealth numbering
Northern Territory	Evidence (National Uniform Legislation) Act 2011 (NT)	Yes	Mirrors the Commonwealth numbering
Queensland	Evidence Act 1977 (Qld)	No	s 95 - admissibility of statements in documents or things produced by processes or devices, with its own presumption and certificate mechanism
Western Australia	Evidence Act 1906 (WA)	No	s 79C - statements in documents, including information from devices designed for recording, measuring, counting or identifying
South Australia	Evidence Act 1929 (SA)	No	Part 6A, ss 59A-59B - verification requirements for computer output that South Australia’s own law reform institute has described as more exacting than the uniform scheme

Norfolk Island also adopted the uniform scheme, in 2004.

The practical point for a national investigation team: the reasoning in this article travels across the uniform jurisdictions and needs local advice in Queensland, Western Australia and South Australia. An investigator working a matter that may end up before the Adelaide or Perth courts should not assume section 146 thinking applies.

Has an Australian Court Admitted Proprietary Forensic Software?

Yes, and this is the most useful Australian authority available on the topic.

In Tuite v The Queen [2015] VSCA 148 (also reported at (2015) 49 VR 196), the Victorian Court of Appeal (Maxwell ACJ, Redlich and Weinberg JJA) considered DNA evidence produced using STRmix - proprietary, commercially licensed forensic software - challenged on the basis that the methodology had not been shown to be sufficiently reliable for use in criminal trials.

The Court did not exclude the evidence. Two strands of reasoning matter:

• It declined to read a reliability threshold into s 79. Section 79 of the Evidence Act, the exception to the opinion rule for opinions based on specialised knowledge, leaves no room for reading in a test of evidentiary reliability as a condition of admissibility. Specialised knowledge can qualify even where the field is novel.
• Reliability was treated as relevant to the s 137 assessment in that criminal proceeding. Section 137 requires a court in a criminal proceeding to refuse to admit evidence adduced by the prosecutor where its probative value is outweighed by the danger of unfair prejudice to the defendant. The Court also observed that the touchstone of reliability for scientific evidence is trustworthiness, and that trustworthiness depends on validation.

The Court additionally urged the establishment of a quality standards body, preferably national, equivalent to the United Kingdom’s Forensic Science Regulator and Forensic Science Advisory Council.

What Tuite settles, and what it does not

It does not make s 137 the general Australian answer to reliability, and it should not be cited as if it does. Section 137 is confined to prosecution evidence in criminal proceedings. It has no application in a civil proceeding, and none at all in the Fair Work Commission or an internal disciplinary process - which is where most of this audience’s work lands.

Forum	The provision that bites	What the contest is about
Criminal proceeding, uniform jurisdiction	s 137 - court must refuse prosecution evidence where probative value is outweighed by danger of unfair prejudice to the defendant	Admissibility, with reliability considered in the balance
Civil proceeding, uniform jurisdiction	s 135 - general discretion to exclude where probative value is substantially outweighed by the danger the evidence might be unfairly prejudicial, misleading or confusing, or cause undue waste of time. A higher bar for the party seeking exclusion	Admissibility, less readily displaced
Fair Work Commission, ART, internal process	Neither. Not bound by the rules of evidence	Weight, fairness, and whether the decision-maker accepts your reasoning

Tuite is also not a document authenticity case, and probabilistic genotyping is not this class of software. What Tuite supplies is an Australian appellate court confronting proprietary forensic software, declining to make source availability a gateway to admissibility, and locating the real inquiry in the evidence actually tendered by the witness actually tendering it. That reasoning travels. The specific statutory route does not.

On the Court’s call for a quality standards body: that remains a live question for the sector rather than a settled one, and it is the right question to put to any supplier. Ask what authority or assessment is being relied on, who performed it, and what it covered.

Why Do Some Courts Push Toward the Source Code, and Others Not?

The most instructive material sits in two American decisions. They are not binding in Australia, they were decided under United States constitutional and evidentiary rules with no Australian counterpart, and they should be read as illustration of a principle rather than as authority. Read that way they are clarifying, because they point in opposite directions and the reason is not what most people assume.

	State v Pickett (NJ App Div 2021)	People v Wakefield (NY Ct App 2022)
Citation	466 N.J. Super. 270 (App. Div. 2021)	2022 NY Slip Op 02771 (26 April 2022)
Software	TrueAllele probabilistic genotyping	TrueAllele probabilistic genotyping
Outcome	Defence entitled to source code and documentation under a protective order	Refusal of source code discovery upheld
Operative reasoning	The defence expert needed access to test independently whether the software operated as intended; the judge could not perform the gatekeeping reliability function without it	The source code was neither a witness nor testimonial under the Confrontation Clause; the witnesses to be confronted were the human analysts

Same software. Opposite results. The distinguishing factor is not openness, and it is not the quality of the software. It is what the software was being asked to be.

In Pickett, the reliability of the software’s own output was the contested issue, and the court was being asked to accept a machine-generated statistical conclusion as the substance of the evidence. Where the machine is effectively the source of the assertion, courts move toward the machine.

In Wakefield, the argument was framed as a confrontation argument - a right to cross-examine the witness against you. The court’s answer was that source code is not a declarant and cannot be cross-examined; the people who could be cross-examined were the human analysts who used the system and formed the conclusions. The concurrence left open whether access under a protective order might be appropriate in another case, which is worth noting rather than glossing over.

Read together, the principle is not secrecy versus openness. Where software is treated as the decision-maker, courts push toward disclosure of its inner workings. Where the software surfaces observations and a qualified human forms and owns the conclusion, the cross-examinable witness is the human, and the analysis is treated as material they relied on. That is the design principle SentinelOps is built around, and it is why the licensing question is the wrong place to look.

What Makes an Investigator’s Own Reasoning Hold Up?

If the investigator is the witness, the investigator’s reasoning is what gets tested - in a hearing, in a merits review, or across a boardroom table.

Makita (Australia) Pty Ltd v Sprowles [2001] NSWCA 305 remains the governing Australian statement on expert opinion. Heydon JA’s analysis requires that the intellectual basis, or essential integers, of the opinion be explained so the court can arrive at an independent assessment of the opinion and its value. The expert’s prime duty is to furnish the court with criteria enabling evaluation of the validity of the conclusions. Facts observed must be identified and admissibly proved; assumed facts must be identified and proved some other way; and the opinion must be shown to rest on the specialised knowledge relied on.

The same discipline appears in the Federal Court’s Expert Evidence Practice Note (GPN-EXPT, dated 25 October 2016), which annexes the Harmonised Expert Witness Code of Conduct adopted by the Council of Chief Justices of Australia and New Zealand on 21 October 2015 and since adopted or mirrored in several Australian jurisdictions. Its consistent theme is that the expert must set out the reasons for each opinion and identify the assumptions and material relied on.

Workplace investigators are not usually tendering expert opinion evidence, and this article does not suggest they are. But the Makita discipline is the clearest available statement of what “reasoning that survives scrutiny” looks like in Australia, and decision-makers apply a recognisably similar standard to an investigator’s report.

Apply it. An investigator who says “the system flagged it” has furnished no criteria at all, and the conclusion is not based on their own knowledge - it rests on a machine’s characterisation they cannot explain. An investigator who says “I was shown this observation, I examined the file and confirmed it myself, here is what I concluded and why, and here is the record of what I was shown at the time” is doing exactly what is asked.

The software’s job is not to be believed. It is to make sure the investigator had the chance to look.

What Should an Investigator Be Able to Produce if the Assessment Is Challenged?

Four things. None requires the supplier to publish its methods, and all four are things you can insist on when evaluating tooling.

What to produce	Why it matters	The question it answers
A reproducible record	The observation must be independently verifiable on the artefact itself, by a competent examiner given the same file	”Can anyone else see what you say you saw?”
A version-stamped analysis	Software changes. A conclusion recorded two years ago must be traceable to the exact version that produced it	”Which engine produced this, and can it be reproduced?”
An immutable record of what was observed at the time	The material the investigator relied on must be the material a challenger sees, not a later revision of it	”Is this what you actually had in front of you?”
The investigator’s own documented reasoning	This is the part that is actually being tendered	”What did you conclude, and on what basis?”

The middle two are routinely underrated, so it is worth being blunt about why they matter.

Version stamping is not bureaucratic hygiene. Software has defects, and the more careful suppliers in adjacent fields publish them: the STRmix miscode summary records fifteen coding faults disclosed since that project began, each scoped to specific software versions, with the affected version ranges stated. That disclosure is what made those defects manageable, because affected work could be identified precisely by version. An analysis that cannot say which version produced it cannot be checked against a later defect notice at all.

Immutability matters because the most damaging line of questioning here is rarely “was the software right?”. It is “is this the same record you relied on at the time, or has it been re-run since?”. A record that can be quietly revised invites that question. A record that cannot, closes it.

SentinelOps records each analysis with the file’s integrity hash, the rules version, the component versions used, the timestamp and the requesting user, so an assessment can answer the version question years later. Completed analyses are never modified. A re-analysis writes a new record: the prior version and the current version both remain, and the difference between them is visible. The record an investigator relied on and cited remains exactly as produced. Where a run is incomplete, the record says so rather than presenting as complete. This is part of why the SentinelOps immutable audit trail is built the way it is.

Does It Matter That the Analysis Uses AI?

It matters a great deal, and the distinction is finer than the public conversation allows.

There is a real and growing Australian concern about AI-generated conclusions entering evidentiary and decision-making processes. Law reform bodies, human rights commissions and judicial education bodies have all been active on it. The concern is not AI as a category. The concern is a machine-generated conclusion that no human can explain and no party can test.

In SentinelOps, the assessment is computed in code from documented rules. A language model is used to narrate observations into readable investigator language; it plays no part in producing or adjusting the evaluative outcome. That is the most important architectural fact for anyone assessing this question, and it is why an AI component in the workflow does not convert the analysis into a machine opinion. The narration layer is a presentation layer. It is not the assessor.

Compare that with the posture common in the wider document-fraud market, where products output a tamper score or a probability that a document is inauthentic. That output is a machine assertion. It is the thing that attracts a Pickett-style demand to look inside, and the thing an investigator cannot explain from their own knowledge.

SentinelOps produces no score and no verdict. It reports what it observes. An indicator is not a finding. An investigator reviews every indicator and records their own assessment before anything enters the case record. And a single observation cannot escalate on its own.

What About Comparing Two Versions of the Same Document?

Workplace investigation teams do a great deal of this by eye - an invoice against one the same person submitted last month, a medical certificate against the practice’s letterhead on file, a signed acknowledgement against the version the employer holds. It is slow, and it is precisely the task where fatigue produces misses.

Two items of evidence can be compared side by side in SentinelOps, textually and visually, because alteration is often only visible as the difference between two files. Examining each file on its own misses the common real-world case where an original and an altered copy arrive as two separate documents.

The evidentiary value of that is straightforward. A difference between two files is among the most demonstrable observations available. It can be shown to a decision-maker, and an independent examiner given the same two files can confirm it. That is reasoning meeting the Makita standard by the nature of the material rather than by assertion.

Where Is the Analysis Performed?

This gets asked earlier and by more people than any of the evidence questions. Investigation teams handling complainant statements, medical records and personal financial documents are frequently uneasy about tools where it is not clear where the data goes.

All SentinelOps analysis is performed in Australia. Evidence never leaves the country, and customer data is stored in Australia. The platform is built, developed and tested in Australia.

For Commonwealth and state government investigation teams this is often a threshold procurement requirement reached well before any evidentiary question. For private-sector workplace investigation teams it is usually a privacy and duty-of-care question about the people whose records are in the file. Either way it is a fair question, and it should be answered specifically rather than with a reassurance.

What Are the Honest Limits?

Any supplier answer to this question that sounds triumphant should be distrusted, including this one.

SentinelOps will never tell you a document is genuine. Absence of indicators is not evidence of authenticity. There is no value in the system that means “authentic”, and none may be added.

An indicator is not a finding. Where the evidence for an indicator is not conclusive, the system says less, not more. It degrades to silence, not to accusation.

On assurance claims generally - and this applies to every supplier in this category, including SentinelOps: treat claims of court approval, legal validation or accreditation with caution, and ask exactly what authority or assessment is being relied on. Who performed it. What it covered. Whether it covered the version you would be buying, or a version from three releases ago. A supplier who can answer those questions precisely is telling you something useful. A supplier who answers with a category label is not.

What SentinelOps offers is architectural: the conclusion is designed to be the investigator’s, so the thing being tested is a person’s reasoning, supported by a record that can be reproduced.

Practical Guidance: What to Do Before the Challenge Arrives

1. Decide, in writing, who owns the conclusion. If your process lets a software output become a case finding without a named human assessment, fix that first. Everything else depends on it.
2. Record the assessment, not just the output. The investigator’s own words, reasoning and date. This is the most commonly missing artefact in the matters that go wrong.
3. Confirm the observation on the artefact yourself. If you cannot see it on the file, do not rely on it. An observation you cannot demonstrate is a black-box assertion with your name on it.
4. Capture the file’s integrity hash at intake, server-side. Establish what you received and when. This is the foundation of every later argument about the document.
5. Preserve the version stamp with the case - in the case record, not in a system log somewhere, where it will still be findable in three years.
6. Never overwrite an analysis. If circumstances change, run a new one and keep both. Prior version, current version, difference view.
7. Compare, where you hold two versions of a document. Alteration is often only visible as the difference between two files, and those typically arrive separately.
8. Write your report so the software is a source, not an authority. “I examined the document and observed X” is your evidence. “The system said X” is not.
9. Know which forum you are heading for. Section 146 and section 137 thinking is court thinking. In the Fair Work Commission or an internal process, the contest is about weight and fairness, and the standard of explanation is if anything higher.
10. Get jurisdiction-specific advice for Queensland, Western Australia and South Australia. The uniform Acts reasoning does not simply transplant.

Where SentinelOps Draws the Disclosure Line, and Why

SentinelOps publishes what the capability does and what its output means. It does not publish the methods by which it does it, because publishing them would function as a manual for defeating them.

This article is the argument for why that position is compatible with Australian evidentiary practice rather than in tension with it. The burden of explanation falls on the witness, not the toolmaker. Australian decision-makers ask the person putting forward a view to expose their reasoning - and where the analysis surfaces observations that an independent examiner can confirm on the artefact, the investigator can expose their reasoning completely without the supplier exposing anything.

More on how the platform handles evidence intake, custody and the case record sits on the evidence management page, and the workplace investigations page covers how this fits a Fair Work Commission-facing process.

Frequently Asked Questions

Is a document authenticity assessment admissible in an Australian court?

There is no Australian appellate authority specifically on document authenticity software, so no product can honestly be described as “admissible”. What can be said is that under the uniform Evidence Acts, s 146 can assist in proving that a device or process produced its asserted ordinary outcome, s 58 allows a court where relevance is in issue to examine a document and draw reasonable inferences including as to authenticity, and Tuite v The Queen [2015] VSCA 148 confirms that reliability is not a threshold condition of admissibility under s 79. What is actually put forward is the investigator’s assessment, and it succeeds or fails on the investigator’s reasoning and their ability to account for what they relied on.

Does section 146 mean the court will presume our software analysis was correct?

No, and this is the most common overstatement in the area. Section 146 is a proof-facilitating provision: where it applies, and where it is reasonably open to find the device or process ordinarily produces the asserted outcome if properly used, it is presumed that it produced that outcome on the occasion in question - unless evidence sufficient to raise doubt is adduced. It does not presume that an analytical conclusion is sound. Be ready to prove what was run, on which file, and under which version.

Is section 137 the test for whether our analysis is reliable enough?

Only in a criminal proceeding in a uniform-evidence jurisdiction, and only for evidence adduced by the prosecutor. In Tuite v The Queen [2015] VSCA 148 the Victorian Court of Appeal treated reliability as relevant to the s 137 assessment in that criminal proceeding rather than reading a reliability threshold into s 79. In a civil proceeding the analogous provision is s 135, which sets a higher bar for the party seeking exclusion. In the Fair Work Commission, the Administrative Review Tribunal or an internal process, neither applies and the contest is about weight and fairness.

Does the software need to be open source to be relied on in Australia?

No Australian authority requires it. In Tuite v The Queen [2015] VSCA 148 the Victorian Court of Appeal considered proprietary, commercially licensed forensic software and did not exclude the evidence. The more useful question is whether the software or the human is the witness.

Who is accountable for the conclusion - the software or the investigator?

The investigator. In SentinelOps, an indicator is not a finding: an investigator reviews every indicator and records their own assessment before anything enters the case record. The conclusion is the investigator’s. The software’s role is to make sure they had the chance to look.

Can the other side demand the supplier’s source code?

No Australian court has ordered it for this class of software. The overseas experience suggests the demand arises where the machine’s own output is the contested conclusion. Where a human forms and owns the conclusion, the cross-examinable witness is that human - the reasoning in People v Wakefield, 2022 NY Slip Op 02771, which is illustrative only and not binding in Australia. Note that the Pickett court in New Jersey reached the opposite result on a reliability challenge and ordered disclosure under a protective order, and that publishing a method openly and disclosing it to a court under protective conditions are different things.

Does section 146 apply in Queensland, Western Australia or South Australia?

No. Those three jurisdictions have not adopted the uniform evidence scheme. Queensland’s Evidence Act 1977 s 95, Western Australia’s Evidence Act 1906 s 79C and South Australia’s Evidence Act 1929 Part 6A (ss 59A-59B) each address machine and computer-produced material differently, and South Australia’s requirements have been described by its own law reform institute as more exacting than the uniform scheme. Take local advice.

What happens in the Fair Work Commission, which is not bound by the rules of evidence?

Section 591 of the Fair Work Act 2009 (Cth) means the Commission is not bound by the rules of evidence and procedure. In practice the contest shifts from strict admissibility to weight, fairness and the quality of your reasoning - and statutory constraints and procedural rulings still apply. This raises rather than lowers the value of a documented human assessment, because the decision-maker is assessing whether to accept your reasoning rather than whether a rule lets the material in.

Does using AI in the analysis create a problem?

It depends entirely on what the AI is doing. A machine-generated conclusion no human can explain is the thing courts and law reform bodies are concerned about. In SentinelOps the assessment is produced by documented rules, not by a language model; the model narrates observations into investigator language and plays no part in the evaluative outcome. That keeps the opinion where it belongs, with the investigator.

What should we keep on file in case an assessment is challenged years later?

Four things: a reproducible observation the investigator confirmed on the file itself; the version stamp identifying exactly which analysis produced the record; an immutable copy of what was observed at the time; and the investigator’s own documented reasoning and conclusion. SentinelOps records the first three by default. The fourth is the investigator’s professional responsibility, and no software can discharge it for them.

Does SentinelOps provide an expert witness?

The witness is the investigator. SentinelOps is built so the person who examined the document, formed the view and signed the assessment is the person who can be asked about it, with a version-stamped, append-only record behind them showing what they relied on and when.

Can SentinelOps tell us a document is genuine?

No, and it never will. The system reports what it observes and produces no score and no verdict. Absence of indicators is not evidence of authenticity. Treating a quiet result as a clean bill of health is the most dangerous thing an investigator can do with any tool of this kind.

Where is the analysis performed?

All SentinelOps analysis is performed in Australia. Evidence never leaves the country, and customer data is stored in Australia. The platform is built, developed and tested in Australia.

---

This article is general information for Australian investigation practitioners about how evidentiary questions are framed. It is not legal advice and must not be relied on as such. Evidence law differs between Australian jurisdictions, and the application of any provision depends on the facts of the particular matter. Obtain advice from a qualified Australian legal practitioner in relation to any specific proceeding.

Written by Lewis Smith, SentinelOps.