Case studies

An agent lodges a return

A scenario following one person who authorizes an agent to lodge her annual tax return, and the subsequent interactions: from a question about eligibility to a disputed assessment. Each step shows how a pattern resolves in sequence, and how they depend on each other.

01

The journey

This scenario is a flow a service team can expect as agent use grows. Because these are emerging patterns, we can expect realized approaches to change rapidly as the technology matures.

There are seven steps, sequenced in the likely order a live service would surface them. Each step shows the associated pattern. The numbered rail marks where a step needs an earlier one already in place. The dashed lane follows the neighbor who files without an assistant, from the route choice down to the same assessment. Each surface is interactive, and shows the resulting effect for both the user and the receiving agency.

1

The question

Guidance must be checkable before she acts on it.

She asks whether her vehicle costs can be claimed. The answer arrives with the provision it read and a link to it, so she can check the claim before acting. An answer without its source reads the same whether the agent found the rule or guessed it.

Response surface · 8.7 · grounded answer
Can I claim my vehicle costs this year?
Based on the vehicle-expense rule and your logged business travel of 4,200 km, you can claim up to $3,570 using the cents-per-kilometer method. Whether it’s allowed is decided at assessment. I can prepare the claim with you.
1Source for “the vehicle-expense rule”
Authoritative sourceWork-related vehicle expenses · the vehicle-expense rule
CurrencyRates effective 1 July 2025 · read today, 09:41
How it was read: typed, read-only
get_rule(topic="vehicle_expenses", method="cents_per_km") → { rate_per_km: 0.85 }

The figures are retrieved from the record, never generated: a claim the system cannot cite is a claim it does not make. The same provenance rides the answer as machine-readable metadata for any downstream agent.

2

The other filer

Hold the non-agent route at parity.

In the same season a neighbor with no agent files the same return by phone. Every route leads to the same assessment; published times and outcomes are what hold the service to that.

Response surface · 7.3 · choose how to lodge
User choosing channel

Lodge your return

Every route leads to the same assessment. The times and outcomes below are what each route actually delivered over the last quarter.

You can change routes mid-filing without losing anything.

Choosing a route without an assistant never changes how your return is assessed. Publishing these numbers is how we are held to that.

System monitors equivalence
Return lodgment · channel equivalence, Q2
Decision time gap
Non-agent routes +1.6 days vs assistant route
within the 3-day tolerance
Approval gap
No significant difference by channel
within tolerance
Completion rate
Phone route down 6% this quarter
review opened 28 Jun

When the two channels diverge past the set tolerance, a review opens automatically. The numbers shown above come from this monitor.

3

The grant

Scope the permission; keep the default narrow.

To lodge, the agent needs authority. Each permission is a plain-language line she can switch off; a standard bundle is pre-selected for anyone who doesn’t want to read a list. She narrows it anyway.

Response surface · 2.1 · authorize your assistant
User authorizing agent
City Digital Services

Authorize your assistant

Choose exactly what your assistant may do to lodge this year’s return. You can narrow any of these now or change them later.

Standard bundle2 of 4 permissions granted
Prepare and lodge this year’s return
Prepares and submits this year’s return. Each lodgment still needs your confirmation.
Read the income data the service holds
Read-only, and only for this return. Cannot be changed or shared.
Touch prior yearsbroad
Access to returns from before this one. Off by default.
Act without asking each timebroad
Standing authority, with no confirmation per action. Off by default.
Or get in-person help at any service center.
4

The record of consent

Keep a consent receipt.

A consent receipt records what was granted, by whom, and until when. It belongs to her as much as to the service. Any later question of overreach is checked against it.

Response surface · 2.4 · delegation record
User reviewing record
City Digital ServicesRevoke this delegation

Delegation record — TaxBot

Operated by Ledgerwise · reference DEL-2026-1847

Active
14 Mar 2026
Delegation granted
Authorized to lodge your 2025–26 return and read the income data the service holds.
15 Mar 2026
Agent read your records12 reads grouped
Income record and mileage log.
18 Jun 2026
Agent lodged your 2025–26 return
Lodgment receipt TX-2026-4829.
Download this record
System reconciles record
City Digital Services · delegation audit
Reference DEL-2026-1847
Authorized
Lodge 2025–26 return · read income the service holds
Done
12 record reads · 1 lodgment TX-2026-4829
Every action fell within the grant

The same record, machine-verifiable. The agency checks what was authorized against what was done. It is the user’s evidence if they dispute it later.

Anchors to ③ the grant
5

The declaration

Ask how it was prepared, without making it a trap.

At lodgment the channel asks how the return was prepared, and the agent’s channel discloses itself as part of the submission. A question submitters are afraid to answer collects false or empty declarations at scale.

Response surface · 1.3 · inside the intake form
04 / 05How you prepared this

Did you use any tools to prepare your submission?

No further detail needed. You can continue.
6

The record

Prove the action; hold it beside the grant.

The lodgment settles into the official record, and at that moment she holds a receipt for the action itself. Held beside the consent receipt, the pair can prove the agent stayed inside its grant, or that it didn’t.

Response surface · 3.2 · action receipt
Your tax return has been lodged
Receipt TX-2026-4829 · 18 Jun 2026, 09:41
Action record
Filed byTaxBot v3.2, operated by Ledgerwise, on behalf of L. Morton
Under delegationDEL-2026-1847 · granted 14 Mar 2026
Data usedsole-trader income record, mileage log VH-2231, City Digital Services record
Outcome$3,570 vehicle claim lodged · assessment to follow
What happens next

City Digital Services will assess your return. You’ll receive your notice of assessment within 14 days. This receipt stays in Your activity. You don’t need to keep this page open.

Under authority of ③ the grant
7

The assessment, and the dispute

Reasons arrive with the assessment; the dispute anchors to the action.

Weeks later an assessment carries a debt she doesn’t recognize, with reasons that name the figures, the disallowed claim, and the rule. Because the reasons name the figures, she can see the claim rests on data mapped to the wrong label. The dispute opens from the assessment, pre-filled from the action record.

Response surface · 3.4 · reasons for decision
Preview a reason given or unavailable
Why this claim was disallowed

The $3,570 vehicle claim was disallowed because the mileage it rests on was recorded as private travel, not business, under the vehicle-expense rule. The debt is the tax on the reinstated amount.

  1. 1Claim $3,570 — 4,200 km at $0.85/km, cents-per-kilometer method
  2. 2Reclassified — the linked log marked this travel private, not business
  3. 3Rule applied — the vehicle-expense rule: private travel is not deductible
If you disagree, you can check the figures against your log or dispute the decision, pre-filled from the action record.
Response surface · 3.5 · dispute
User disputing

Dispute action TX-2026-4829

We’ve filled in the record for you. You only need to tell us what went wrong.

Pre-filled from your receipt
Filed2025–26 individual tax return
ByTaxBot v3.2 (Ledgerwise)
Date18 June 2026
Dataincome record, mileage log VH-2231
System shows case queue
No open disputes

New disputes arrive with the action record attached. The response clock starts when a dispute is received.

Submit the dispute on the left to watch it arrive here. Each dispute is tagged to an agent or model version and also counts toward the overall error rate. When many users report the same fault, it registers as one pattern rather than as unconnected phone calls.

Anchors to ⑥ the action record · provable against ④ the consent receipt

Both routes, one assessment

The neighbor’s return is assessed the way hers was. That holds only while someone owns the phone route and staffs it. Otherwise the agent channel gets cheaper to run, the phone queue lengthens, and the route without an assistant reaches a worse outcome on a slower clock. Which future arrives is a resourcing decision made long before either filer picks up the phone.

02

What this case informs

8.7 Grounding and source attribution in AI outputs Emerging

The opening question is the earliest point an error can enter the filing; a per-claim citation is what lets the filer check guidance before acting on it.

2.1 Fine-grained scope negotiation Frontier

The grant screen renders each permission as a plain-language line with a narrow default; the scenario's chain of records starts from a scoped, inspectable grant.

2.4 Consent receipts and records Frontier

The consent receipt is the scenario's baseline record: what was authorized, by whom, until when, held beside what was later done under it.

1.3 Self-attestation and disclosure Emerging

The preparation declaration only works as a safe, structured question; a form that reads as a trap collects false or empty declarations at scale.

3.2 Confirmation receipt showing what happened Emerging

The action receipt anchors everything downstream of the filing: proof of lodgment first, then the dispute.

3.4 Reasons for decision Frontier

Reasons that name the figures and the rule are what turn a surprise debt into a claim the filer can check.

3.5 Recourse and dispute resolution Frontier

The dispute opens from the identified action, pre-filled from its record, on a stated clock; the filer never re-narrates the history.

7.3 Non-agent fallback channels, kept first-class Emerging

The phone filer reaches the same outcome only if the non-agent route is held at parity as a design duty, and resourced accordingly.

7.9 Digital divide and exclusion risk Emerging

The scenario's two filers are the divide the pattern names; the gap between their experiences is decided before either channel launches.