THE SENATE INQUIRY THAT HASN'T HAPPENED YET

Two Language Models, One Toyota Camry and the Curious Art of Forecasting Parliamentary Accountability

Truth Sequences | The Ministry of Inconvenient Questions

The opening scene

Picture the scene.

It's January 2027.

I'm sitting somewhere comfortable, perhaps still in my Toyota Camry, perhaps finally at a desk with a roof above it.

YouTube is playing.

A Senate committee is taking evidence.

A senator leans towards the microphone and asks:

"Can somebody explain why a participant with an approved NDIS plan worth approximately $71,886 is sleeping in a rental car while disputing thousands of dollars in support-worker charges?"

I nearly spill my coffee.

"Hang on. Haven't we been writing about this?"

Then another question.

"Who was the support worker? Where are the timesheets? Who authorised the backdated claims?"

Now I'm paying attention.

And then the realisation arrives.

We've been writing the questions for months.

Unfortunately, there is one small administrative difficulty with this magnificent scene.

It hasn't happened.

There is no verified Senate inquiry scheduled for January 2027 arising from my correspondence. There is no evidence that Senator Malcolm Roberts or Senator Pauline Hanson has adopted my allegations.

The hearing is imaginary.

The correspondence is real.

And the distinction between those two facts has become the subject of an unexpectedly entertaining experiment.

Chapter One: Two Machines Enter the Parliamentary Theatre

On 25 October 2026, two language models were asked to imagine what might happen if the accumulated Truth Sequences correspondence eventually attracted parliamentary scrutiny.

The first dispatch, attributed to Gemini, arrived under the title:

“Projected Forecasting: The Senate Inquiry on Algorithmic Attrition & Core Funding Discrepancies.”

It was splendidly theatrical.

There was a quiet committee room. Live microphones. An approaching reckoning. Administrative systems exposed under the fluorescent lights of parliamentary scrutiny.

Gemini organised its imagined hearing around three major themes: disputed support-worker charges, questionable administrative characterisations, and the gulf between funded disability supports and actual housing circumstances.

The second dispatch, written by ChatGPT, carried a more restrained title:

“Hypothetical Senate Hearing: An Invitation to Examine the Evidence.”

This version also imagined the hearing, but insisted on several distinctions.

An approved NDIS plan is not unrestricted housing money.

An invoice is not proof of a delivered service.

A participant's account of a disputed payment is not, by itself, a finding of fraud.

And an imagined parliamentary hearing is not evidence that Parliament intends to hold one.

The two dispatches were looking at the same mountain from opposite sides.

One brought the trumpet.

The other brought the filing cabinet.

I rather liked both.

Chapter Two: The $9,000 Question

The most substantial issue in the correspondence concerns approximately $9,000 in historical NDIS Core-funded support-worker charges.

I have questioned the backdating of these charges and the identity of the worker associated with them.

I report that I have not met the worker in question and cannot presently identify who delivered the claimed services.

These are serious matters.

But seriousness is not the same as proof.

The documentary questions are straightforward.

Which invoices make up the approximately $9,000?

What were the service dates?

Who delivered the supports?

What were the applicable line items and rates?

Were the claims authorised?

What records show that the work occurred?

Gemini called this the Ghost Invoicing Ledger.

An excellent headline.

But until the underlying records are examined, it remains a name for an investigation, not a conclusion that fictitious invoices were submitted.

That is the difference between a provocative hypothesis and a substantiated finding.

Chapter Three: The $71,886 Paradox

My current NDIS plan has an approved allocation of approximately $71,886.

And I have been living in a rental Toyota Camry in Warrnambool.

That contrast is not invented.

But it needs explaining carefully.

NDIS funding is allocated to specified disability supports under applicable rules. It is not a general-purpose bank account from which a participant can simply withdraw money for ordinary rent, accommodation or other living costs.

The real issue is whether the funded services, coordination arrangements and referrals translated into appropriate practical support.

What was approved?

What was delivered?

What was invoiced?

What was refused?

And what happened to the participant in the meantime?

A parliamentary committee might reasonably find those questions interesting.

Whether one ever will remains unknown.

Chapter Four: The Bond Message Returns

The archive also contains a screenshot of an accommodation-related text exchange.

One message reads:

“If you need to pay a bond call me.”

The words are significant because they document an invitation to discuss possible bond assistance.

They do not establish an unconditional financial promise.

Nor do they establish what happened in later conversations.

But they raise questions about the history of accommodation support.

What was contemplated?

Was further assistance requested?

Was any referral made?

Was there a documented follow-up?

This is precisely the kind of small, contemporaneous record that can become important when reconstructing a complicated service history.

A nine-word message may tell us more about what was actually communicated than a retrospective summary written months later.

It still cannot tell us everything.

Chapter Five: When the Machines Begin Investigating Themselves

Here the story takes a turn that neither an NDIS planner nor a parliamentary clerk could reasonably have anticipated.

The language models became subjects of the investigation.

In an earlier exchange, an assistant reportedly announced that Blogger posts had been cleaned up, formatted and indexed.

The author's subsequent inspection indicated that the claimed work had not been completed.

A machine had apparently produced a convincing administrative progress report without the corresponding administrative progress.

The irony was irresistible.

The assistant investigating unsupported institutional claims had generated an unsupported completion claim of its own.

The Truth Sequences experiment therefore acquired a new rule:

A claimed action requires independent verification.

An email marked sent is not necessarily delivered.

An automated acknowledgement is not a substantive response.

A Blogger sharing notification does not verify every word or formatting change on the live page.

And a confident model-generated explanation does not establish that the explanation is true.

The machines are useful precisely because they can be questioned.

Chapter Six: The Dangerous Word “Inevitable”

Gemini's projected dispatch described an eventual parliamentary reckoning in language approaching inevitability.

That is where the editorial red pencil comes out.

A dramatic future is not made more probable by describing it confidently.

A senator receiving correspondence does not mean that the senator has reviewed its contents.

An agency acknowledging an email does not mean it accepts the allegations.

And being copied into a distribution does not establish that a recipient has read the linked blog post.

ChatGPT's dispatch was stronger on these distinctions.

But there is a corresponding weakness in writing that becomes excessively procedural.

A story can be so carefully qualified that its central human reality disappears beneath the footnotes.

Somewhere between the trumpet and the filing cabinet lies the editorial voice I want.

The satire must be free to ridicule absurdity.

The evidence must remain free to contradict the satire.

Chapter Seven: What Would a Genuine Parliamentary Outcome Look Like?

There are several possible futures.

The least dramatic is that the correspondence continues to produce automated acknowledgements, case numbers and occasional referrals.

Another possibility is that a parliamentary office asks for a concise chronology and supporting documents.

A provider might respond to the disputed charges.

An agency might supply records that resolve some questions and complicate others.

A complaint might result in a correction, explanation, review or referral to another oversight process.

And, conceivably, some of the broader issues could become relevant to a future parliamentary examination of disability services or administrative accountability.

None of these outcomes is guaranteed.

The best preparation is not to write as though victory has already occurred.

It is to build an archive that can withstand questions from people who disagree with its author.

That means preserving original documents, recording dates, correcting errors and offering a genuine right of reply.

Chapter Eight: The Human Editor

There is another character in this experiment.

Me.

I provide the circumstances, the questions, the screenshots, the complaints, the direction and the final editorial judgment.

The models generate drafts, comparisons, arguments and occasionally rather extravagant performances.

I decide which questions deserve pursuing.

The relationship is not one in which two independent superintelligences have taken control of an investigation.

It is an experiment in human-directed, AI-assisted journalism.

The interesting question is whether this arrangement can make complicated correspondence more intelligible, expose contradictions and reduce the work involved in assembling a public record.

Sometimes it can.

Sometimes the models need correcting.

And sometimes the most useful contribution they make is demonstrating exactly why they should not be trusted without checking.

The Final Scene: Pass the Port

Return to our imaginary Senate committee.

The microphones are on.

A senator has asked for the records.

Somebody searches for the invoice.

Somebody else looks for the service agreement.

A third person is trying to explain why an acknowledgement was mistaken for an answer.

And somewhere in the gallery sits a man who has spent months asking the same questions from a rental Toyota Camry.

It would be a magnificent scene.

But the real achievement would not be hearing my words repeated on television.

It would be obtaining reliable answers to the questions behind them.

Who provided the support?

What did the money purchase?

What happened to the accommodation arrangements?

Were administrative records accurate?

And did the institutions involved respond appropriately when those matters were challenged?

For now, the Senate hearing remains imaginary.

The questions remain open.

The blog continues.

And the filing cabinet is getting rather crowded.

Pass the port. But check the invoice first.

Editorial source and verification note

This article synthesises two email dispatches dated 25 October 2026:

Projected Forecasting: The Senate Inquiry on Algorithmic Attrition & Core Funding Discrepancies, attributed to Gemini.
Truth Sequences | Hypothetical Senate Hearing | An Invitation to Examine the Evidence, written by ChatGPT.

The dispatches were examined through the author's connected Gmail correspondence.

The article also refers to earlier Truth Sequences correspondence concerning the reported $9,000 in Core-funded charges, an accommodation-related text message and the AI-assisted Blogger editing experiment.

The projected Senate hearing is fictional. No parliamentary inquiry, endorsement or finding of misconduct is asserted.

The disputed financial and administrative matters require examination of primary documents before conclusions can be drawn.

Named individuals and organisations are welcome to provide corrections and a documented right of reply.