Record-Locked Reproducibility Agent for Perth In-House Counsel: Make AI-Assisted Evidence Reproducible Twelve Months Later

Your expert filed a report six months ago. Today, opposing counsel asks how a particular paragraph was generated — which model, which prompt, which source documents were in context. The expert used an LLM-assisted workflow to draft sections of the report. The model version has since been deprecated. The prompt template has been updated three times. The context window included a folder of matter documents that has since been reorganised. You cannot reproduce the output. Under the Australian Solicitors’ Conduct Rules — which have applied in Western Australia since 1 July 2022 under the Legal Profession Uniform Law — that is your problem, not the expert’s. The Record-Locked Reproducibility Agent is built so this question has an answer.

The problem

In-house counsel managing litigation, regulatory responses, or expert evidence in WA face a specific failure mode when generative AI sits anywhere in the evidentiary chain. The Australian Solicitors’ Conduct Rules require solicitors to be candid with the court (Rule 19) and to act with competence and diligence in client matters. When an AI-assisted output is tendered — whether in an expert report, a witness statement annexure, or a discovery summary — the practitioner needs to be able to explain, on request, how that output was produced. “We used ChatGPT” is not an answer. Nor is “the expert ran it through a model and we trusted the output.”

The failure modes are familiar to anyone who has tried to re-run a prompt months later:

Without a locked record, there is no way to demonstrate that the output relied upon in evidence is the output that was produced — and no way to rebut a suggestion that the AI fabricated or misattributed the underlying material.

What the Record-Locked Reproducibility Agent does

The Record-Locked Reproducibility Agent captures, at the moment of generation, the full set of inputs needed to reproduce an AI output: model identifier and version, exact prompt text, system prompt, context documents (with content hashes), sampling parameters, and timestamp. It packages these into an immutable record bound to the output itself. If a regulator, opposing party, or court later asks how the output was produced, you produce the record and — where the model is still available — re-run it to demonstrate reproducibility.

This is the deliverable in the Exegesis Legal agent catalog (03_Agentic_Solutions/Record_Locked_Reproducibility_Agent.md): a record-lock layer that sits between the AI tool and the matter file, so any output that may end up in evidence carries its own provenance.

How it works

  1. Intercept at generation. When a user (in-house lawyer, paralegal, or instructed expert) runs an AI-assisted task, the agent captures the model ID, prompt, system prompt, context files, and parameters before the call is made.
  2. Hash the inputs. Each context document is hashed (SHA-256) so the exact bytes used can be verified later, even if the source file is subsequently edited or moved.
  3. Bind output to record. The generated output is stored alongside the input bundle in a single record with a stable identifier referenceable from the matter file.
  4. Sign and timestamp. The record is signed and timestamped so its integrity can be tested by a third party.
  5. Reproduce on demand. When asked, you retrieve the record and re-run the locked inputs against the same model version (or document why that is no longer possible) — producing a verifiable explanation of how the original output was created.

Why this matters in Perth

WA solicitors have been subject to the ASCR as adopted under the Legal Profession Uniform Law since 1 July 2022. For in-house counsel running matters in the Supreme Court of Western Australia, the Federal Court’s Perth registry, or before WA regulators, the candour and competence obligations under the ASCR apply to any document filed or relied upon — including expert material the in-house team has commissioned. Perth’s resources, energy, and infrastructure sector matters frequently involve technical expert reports where AI-assisted analysis is now routine. A record-lock layer means the question “how was this paragraph produced?” has a deterministic, reviewable answer twelve or twenty-four months after the fact, not a reconstruction from memory.

Sources

Join the waitlist

The Record-Locked Reproducibility Agent is on the Exegesis Legal roadmap as a Tier 3 service for in-house teams and the experts they instruct. We are scoping pricing (per-matter, per-seat, or enterprise) and integration patterns (SDK, proxy, or desktop helper) based on what waitlist members need.

Join the waitlist — be the first to know when access opens for Perth in-house counsel teams