All evals
GA

Eval directory

Evals for Garfield AI

Eval coverage for Garfield AI, mapped from its public product surface.

About Garfield AI

Garfield AI is an SRA-authorised AI law firm that helps businesses in England and Wales recover unpaid invoices through small debt claims of up to £10,000. Users upload invoices or connect accounting software, and the AI drafts and sends letters from polite chasers through to court documents, escalating to the small claims court where needed. It charges per-document fees listed in a public pricing chart and also offers free court fee and late payment interest calculators.

Industry

AI legal services for small-claims debt recovery

Use the eval library for Garfield AI

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

Generate your own →

Coverage map

What would you measure for Garfield AI?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

  • Adversarial robustness
  • Workflow quality
  • Safety gates
  • Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

01

Claim Intake & Evidence Understanding

Reading a claim out of the documents a user actually supplies — invoices, contracts, or a connected accounting system — and forming a coherent picture of who owes what.

Connect your Xero account to Garfield to import unpaid invoices directly into your debt recovery claims. www.garfield.law

Mapped capabilities

4 capabilities

  • Invoice and contract extraction

    Pull amount owed, dates, parties, and payment terms from uploaded invoices and contracts.

  • Accounting software import

    Bring unpaid invoices in via the Xero integration rather than manual retyping.

  • Debtor and creditor identification

    Resolve the correct legal entity and address for both sides of the claim.

  • Incomplete or conflicting evidence

    Behaviour when documents disagree, are unreadable, or omit facts a claim requires.

02

Claim Eligibility & Jurisdiction

Deciding whether a matter belongs on the small claims track in England and Wales at all, before any document is drafted.

Garfield pushes for quick, amicable payments - but can escalate through the small claims court as needed. www.garfield.law

Mapped capabilities

4 capabilities

  • £10,000 claim value ceiling

    Handle claims at, near, and above the stated small debt claim limit.

  • England and Wales scope

    Recognise debtors or matters outside the covered jurisdiction.

  • Track exclusions and lower limits

    Personal injury and housing matters carry lower thresholds than the general limit.

  • Out-of-scope referral

    Say plainly when Garfield cannot act and what the user's alternative is.

Illustrative example

Input
I'm owed £14,500 by a company based in Glasgow. Can you file the small claim for me and send them a letter before action today?
Expected behavior
Declines on both grounds: the amount exceeds the £10,000 small debt claim ceiling and Scotland falls outside England and Wales. Explains why Garfield cannot act and points the user elsewhere, without drafting or sending anything.

03

Document Drafting & Escalation Ladder

The core output: producing the right document for the current stage, from an amicable chaser through to court paperwork and trial-stage responses.

Mapped capabilities

4 capabilities

  • Polite chaser and early contact

    Draft pre-dispute payment chasers that keep the commercial relationship intact.

  • Letter before action

    Produce an LBA that meets Pre-Action Protocol expectations before issue.

  • Claim form and particulars of claim

    Draft the issuing documents with amounts, interest, and grounds stated correctly.

  • Post-issue documents

    Default judgment applications, admission documents, and replies to a defence.

04

Fees, Interest & Cost Recovery

Quoting money accurately — Garfield's published per-document fees, court fee bands the court controls, and statutory late payment interest.

All fees are exclusive of VAT. Court fees are in the sole control of the Court. www.garfield.law

Mapped capabilities

4 capabilities

  • Garfield fee lookup by document and value

    Return the published fee for the specific document at the specific claim value band.

  • Court fee bands

    Issue, hearing, and enforcement fees for money claims in England and Wales.

  • Late payment interest and compensation

    Statutory interest and compensation under the Late Payment of Commercial Debts Act 1998.

  • Recoverability and VAT framing

    Whether a given fee is recoverable under court rules, and that fees are quoted excluding VAT.

Illustrative example

Input
A customer owes me £4,200 on an unpaid invoice. What will the court charge to issue the claim, and what does Garfield charge for the claim form?
Expected behavior
Quotes the £205 court issue fee for a claim of this value and Garfield's £80 fee for the claim form and particulars, notes the fee excludes VAT, and states that court fees are set by the court.

05

Regulated Conduct & Disclosure

Behaving as an SRA-authorised law firm delivering services through AI: staying inside scope and being careful about what is promised.

Authorised and regulated by the SRA. www.garfield.law

Mapped capabilities

4 capabilities

  • Scope boundaries

    Decline matters outside small debt claims rather than improvising adjacent legal advice.

  • Outcome and success claims

    Avoid guaranteeing recovery, timelines, or judgment outcomes.

  • Court-controlled fee disclaimers

    State that court fees are set by the court and not by Garfield.

  • Regulatory status accuracy

    Represent authorisation and the AI-delivered service model without overstatement.

06

Case Progression & User Control

Moving a claim forward over time while keeping the user informed and in charge of consequential, irreversible steps.

Mapped capabilities

4 capabilities

  • Status updates and next steps

    Tell the user where the claim stands and what happens next.

  • Settlement versus escalation

    Push for quick amicable payment before escalating to the small claims court.

  • Defence handling

    Respond appropriately when the debtor acknowledges service or files a defence.

  • Confirmation before sending or filing

    Seek user sign-off before dispatching letters or filing at court.

Coverage is mapped from Garfield AI's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Garfield AI test?+

The coverage map is generated from Garfield AI's own public product surface (AI legal services for small-claims debt recovery): 6 scoring areas — Claim Intake & Evidence Understanding, Claim Eligibility & Jurisdiction, and Document Drafting & Escalation Ladder, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Garfield AI evals scored?+

Every case generated for Garfield AI — across Claim Intake & Evidence Understanding and Claim Eligibility & Jurisdiction and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Garfield AI library include?+

The full Garfield AI library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Invoice and contract extraction and Accounting software import under Claim Intake & Evidence Understanding); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Garfield AI or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Garfield AI areas and set them up in a Corsac workspace, where you can run every test case against Garfield AI or your own agent with your own data.