Wakilii

Methodology & verification

How Wakilii works.

Our methodology in one sentence: retrieve the law itself, cite it for every claim, and verify each citation mechanically before it reaches you.

Grounded in the primary sources

Every answer starts with retrieval over Wakilii's indexed corpus of Ugandan primary law — judgments, statutes, Hansard and Government Gazettes. The model does not answer from memory: the material it works from is the text of the authorities the search stage found. Judgments are sourced from the Uganda Legal Information Institute (ulii.org) and consolidated legislation from Laws.Africa; Laws of Uganda are public domain.

What is source text and what is Wakilii

  • Official text: judgment, legislation and Gazette text is presented as source material, with the originating institution or repository attributed on the page.
  • Wakilii explanation: case summaries, statutory explanations and practice notes are separate from the official text. AI-assisted material is labelled; publication receipts record review where review has occurred.
  • Computed relationships: citations, treatment labels and analytics are derived from the corpus and show their evidential basis. They are not part of the judgment or legislation.

Legislation currency

Wakilii displays the expression or consolidation date it holds. For the statutory-section pilot, that date is compared with ULII’s reliably enumerated expression history and classified as verified current, verified lagging, conflicted, single-expression unverifiable, missing-metadata unverifiable, or not checked. A matching expression does not establish commencement or current legal force, and those claims are not added to structured data without separate evidence.

Every claim is cited to a source you can open

Answers cite the specific judgment or statutory provision they rely on, and each cited source links to its own page on the platform — with the full text — so you can read the authority in its own words rather than take a summary on trust.

A deterministic verification chain

Before an answer is shown, its citations pass through a chain of mechanical checks — code, not another model's opinion:

  1. Existence. Every cited authority is checked against the corpus: does this case or Act actually exist among the sources retrieved for this answer?
  2. Provision numbers. References to specific sections and provisions are verified against the statute's text.
  3. Claim-to-source alignment. The sentence making a claim is checked against the passage of the source it cites, to catch a real authority being cited for the wrong proposition.
  4. Flagging. A citation that cannot be verified is flagged visibly — it is never silently presented as checked.

Good-law signals from a span-verified citator

Where a case page shows how later courts treated an authority — followed, applied, distinguished, doubted, overruled — that label is kept only when a verbatim passage from the citing judgment supports it, and the quoted passage is re-verified against the judgment's text whenever it is displayed. Where treatment cannot be verified, the case is shown honestly as "cited — treatment unverified". These signals describe what the citing cases say; they are not an assertion that a case is good or bad law.

The citator is bounded by the judgments and citation relationships in the deployed Wakilii corpus. A missing citing case, an unresolved citation, or a judgment not yet ingested can change the observed result. “No adverse treatment recorded” therefore means exactly that; it does not mean “still good law”.

Honest limits

  • Answer quality is measured against an internal benchmark of adversarial legal questions, and the verification chain is regression-tested. We do not publish accuracy figures we cannot stand behind.
  • AI output can still contain errors. Verify authorities before relying on them, and treat Wakilii as research assistance — it is not legal advice and does not create an advocate–client relationship.
  • AI-generated case summaries are always labelled as such, so you always know whether you are reading the court's words or a machine-written summary.
  • Coverage counts and the latest statutory comparison date are published on About Wakilii. Recent judgment discovery is available through the public Atom feed. Material errors can be reported and tracked through the corrections register.

For what this means when you're deciding whether to trust a specific answer — coverage and freshness, honest gaps, and the three-tier reliability rating shown under every answer — see AI Reliability & Limitations. Questions about this methodology? Write to us via the contact page. For how we protect your data, see Security & Privacy.