ADRIAN BERTINO-CLARKE is the founder of FIA Labs and the creator of LIA Apex, a governed legal intelligence workspace being developed with miscarriage-of-justice work in mind. LIA Apex is designed to help lawyers, investigators, innocence advocates, journalists, academics, and review bodies organise complex case records, test prosecution theories, challenge expert evidence, identify unresolved doubt, and preserve reviewable governance logs. It is not an innocence machine or a substitute for lawyers; it is a tool for making legal reasoning more disciplined, transparent, and accountable. 

 Wrongful convictions rarely begin with a single dramatic failure.

They emerge through accumulation. A doubtful identification is treated as stronger than it is. A forensic opinion is allowed to sound more certain than the science permits. A confession is accepted without adequate scrutiny of vulnerability, pressure, contamination, or context. Disclosure gaps remain buried. An early theory takes hold and begins to shape how every subsequent fact is interpreted.

By the time a conviction is challenged, the case may no longer look like a collection of open questions. It looks like a settled record. And the task of unpicking it becomes extraordinarily difficult.

That is one of the central reasons I have been building LIA Apex.

LIA Apex is not a general legal chatbot. It is being built as a governed legal intelligence workspace — a system specifically designed to help lawyers, investigators, innocence project volunteers, journalists, family advocates, academics, and review bodies examine legal and evidentiary problems with greater discipline and transparency.

Its defining feature is not speed or fluency. It is reviewability.

 

The Record Behind the Reasoning

 One of the most significant features being built into LIA Apex is its governance-log architecture.

That may sound technical, but its significance is practical — and its design reflects something that matters deeply in legal work: not every reasoning record should be exposed.

Privilege, confidentiality, litigation strategy, work product protection, client instructions, and sensitive case analysis all carry legal and ethical weight. A serious legal AI system should not casually turn privileged reasoning into a discoverable trail. LIA Apex is being designed with that concern built in from the start.

For that reason, LIA Apex is being designed with two types of governance logs.

The first is an internal governance log. This is the fuller matter-level record available to the practitioner or authorised user. It can help show what material was reviewed, what assumptions were made or flagged, what uncertainties were identified, what claims were verified or left pending, what alternatives were tested, what challenges were raised, and what decisions the human user made after reviewing the system’s output.

The second is a discoverable governance log. This is a more limited proof layer, designed to protect privilege and confidentiality while still demonstrating the integrity of the process. It need not expose strategy, private instructions, privileged analysis, or the substance of confidential advice. Instead, it can show process indicators: whether evidentiary validation was complete or pending, whether alternative hypotheses were considered or left open, whether expert claims were challenged, whether uncertainty was flagged, and whether the forensic choices made were reasonable and defensible at the time.

That distinction matters considerably.

Courts, parties, regulators, insurers, and disciplinary bodies may not always need to see the full reasoning file. But they may legitimately need to know whether AI-assisted legal work was prepared with disciplined evidence review rather than unsupported assertion.

A judge may ask at a pre-trial conference how AI was used in preparing an argument or reviewing evidence. A party may seek disclosure of non-privileged AI-assisted process records, or a court may require a limited assurance record showing whether evidentiary validation, alternative hypothesis testing, and challenge steps were properly carried out. In a leave application, a court may want to understand whether a proposed ground has genuine evidentiary substance or remains speculative. In a professional negligence claim, a practitioner may need to demonstrate that the advice given, the evidence relied upon, and the forensic choices made were reasonable at the time.

LIA Apex is being built for that world.

The point is not to make every internal thought discoverable. It is to make legal AI use more accountable without destroying the protections that allow lawyers to do their work properly. If AI is used in legal reasoning, the system should be able to show that the process was disciplined, validated where necessary, and appropriately challenged — while still respecting the boundaries of privilege, confidentiality, and professional responsibility.

This is not surveillance for its own sake.

It is reviewability without recklessness.

From Conversation to Continuing Matter-Based Review

LIA Lite, the first version, is mainly conversational. LIA Apex is matter-based. That shift matters.

A wrongful conviction review is rarely a single exchange. It is a long, evolving process: chronologies are built, evidence maps are drawn, fresh evidence arrives, expert opinions are challenged, and a reviewer may return to the same material repeatedly over months or years.

LIA Apex is designed to support that reality. A case becomes a persistent workspace where analysis accumulates rather than disappearing into a series of disconnected chats. This is particularly important for miscarriage-of-justice work because the problem is usually cumulative. Individual weaknesses may seem minor in isolation. But several weaknesses, properly mapped, can reveal a pattern: tunnel vision in the investigation, unreliable identification, overclaiming by experts, inadequate disclosure, poor defence work, or a prosecution theory that was never rigorously tested.

Users can begin in plain language. A family member might ask, “How do I organise this case?” A journalist might ask, “What needs verification before publication?” An advocate might ask, “What are the strongest possible grounds for review?” From there, the system can move into structured modes: verification, challenge, chronology construction, expert evidence analysis, disclosure gap mapping, or legal research.

That accessibility matters. Many miscarriage-of-justice cases begin outside elite legal settings. They begin with a family member, an advocate, a journalist, a student, or a local lawyer who suspects something is wrong but does not yet have a disciplined framework for that concern.

Challenge as a Core Justice Function

Many wrongful convictions involve early narrative lock-in. A theory forms, facts are interpreted to fit it, contradictions receive less weight, and alternative explanations are sidelined. The case becomes coherent because the theory has trained everyone to see it that way.

LIA Apex is being designed to resist that tendency.

Its challenge function actively asks the kinds of questions that should be asked before any theory hardens: What facts contradict the preferred account? What evidence is assumed rather than proved? What alternative explanations remain viable? What claims depend on unsupported inference? What disclosure gaps matter? What would rigorous cross-examination explore? Where does consistency with guilt get mistaken for proof of guilt?

That capability is valuable well beyond the defence. It is useful for prosecutors who want to test whether a case is genuinely safe before proceeding. It is useful for conviction integrity units reviewing old cases. It is useful for journalists and researchers trying to avoid overstating claims.

The value is not that LIA Apex “knows” the answer. It does not. The value is that it keeps doubt, alternatives, and assumptions visible long enough for human judgment to do its proper work.

Forensic and Medico-Legal Evidence

A large proportion of wrongful convictions turn on expert evidence. Often the problem is not outright fabrication but overstatement: possibility presented as probability, association treated as causation, general population-level research applied too confidently to an individual case.

LIA Apex is being built with medico-legal and causation discipline specifically in mind. It helps users distinguish between possibility, association, mechanism, probability, and legal causation — distinctions that matter acutely in cases involving forensic pathology, shaken baby allegations, DNA transfer, toxicology, injury timing, fire investigation, digital evidence, and accident reconstruction.

It surfaces questions like: What does the study actually show, and what does it not show? Were confounders properly controlled? Is the expert applying general research to the specific case correctly? Is the opinion stronger than the underlying data allows? What alternative explanations remain open?

This is one of the clearest ways AI can assist justice without displacing legal judgment. It can help prevent scientific uncertainty from hardening into courtroom certainty before that process has been properly examined.

Dual-Engine Architecture for Safer Reasoning

LIA Apex is being developed with a dual-engine approach rather than a single generative model.

The purpose is not to make the system more impressive. It is to make the reasoning safer.

The intended architecture uses a staged process: intake and routing, primary synthesis, independent validation, and final arbitration. One pass produces an analysis. An independent engine then challenges it — testing unsupported claims, identifying missing assumptions, and flagging alternative explanations. This creates deliberate friction before the output reaches the user.

A single coherent explanation can be dangerous in legal work. A case theory can sound persuasive while resting on fragile inference. An appellate ground can look attractive until the record is tested. A medico-legal claim can seem strong until the causation chain is examined in its parts.

Dual-engine validation is intended to introduce that internal challenge before the lawyer or reviewer receives the result. The goal is not to replace the practitioner. It is to give them a better-tested starting point.

Practical Applications

For lawyers, LIA Apex is designed to support the work that often determines whether a case is safe: early case assessment, evidence review, chronology construction, appeal issue identification, expert-report analysis, legal research planning, argument stress-testing, forensic cross-examination preparation, client explanation, and drafting review memoranda.

A defence lawyer can use it to test whether the prosecution theory actually follows from the evidence. An appellate lawyer can use it to identify possible grounds and separate stronger issues from weaker ones. A solicitor advising a family can use it to turn years of documents into a structured review plan. A prosecutor can use it to check whether a case has become overly dependent on assumptions. A journalist can use it to identify what needs legal or expert verification before publication.

For innocence projects, student clinics, and review commissions, LIA Apex could help create more repeatable workflows for disclosure gap analysis, expert-evidence reliability review, fresh evidence assessment, case triage, and pattern detection across cases.

The common thread is not automation. It is disciplined assistance. LIA Apex should help lawyers and reviewers think, test, organise, and verify. It should not encourage them to delegate responsibility. That is a design principle.

What LIA Apex Is Not

LIA Apex is not an answer machine. It is a governed legal reasoning workspace for cases where the reasoning itself may one day need to be reviewed.

It is important to be clear about the limits.

LIA Apex will not declare guilt or innocence. It will not replace lawyers, experts, investigators, judges, or review commissions. It will not produce confident legal conclusions from incomplete material. It will not turn a family’s hope into false certainty. It will not give institutions cover for superficial review. It will not expose privileged strategy or confidential legal reasoning merely because AI was used in the preparation process.

Its purpose is to make legal and evidentiary reasoning more transparent, disciplined, contestable, and accountable to the actual evidence. Not to automate justice. Not to manufacture conclusions. Not to serve as a shortcut around the hard work that serious cases require.

 

The Larger Purpose

Wrongful convictions are failures of evidence, but they are also failures of reasoning. Doubt was not preserved. Alternatives were not tested. Expert claims were overstated. Disclosure gaps were never exposed. A theory became too coherent too early, and the system moved from suspicion to conviction without enough friction at the right points.

LIA Apex is being built to introduce some of that friction.

Not delay for its own sake. Not endless doubt. Not a machine second-guessing every lawyer or judge. The right kind of friction: the kind that keeps evidence, inference, uncertainty, and professional responsibility clearly visible before action is taken.

That is the connection between this project and the broader work I have been developing in From Advice to Authority. AI systems in high-consequence domains should not simply produce outputs. They should preserve the conditions under which human beings can reason responsibly.

In law, that responsibility is especially serious. A flawed legal conclusion can cost money, reputation, liberty, family life, and sometimes life itself. If legal AI is going to be used in that world, it must be built with more than efficiency in mind.

It must be built for truth, evidence, uncertainty, challenge, accountability, and justice.

COMING SOON