Verification ledger
A record kept beside any document that cites sources, with one row for each claim that depends on a source: the proposition it carries, the citation as held, the primary source to check, the specific points to confirm, the falsifier that would defeat the claim, and a status of unverified, verified against the primary record on a stated date, or failed.
When this decides something for you
A document is due. It cites forty things: judgments, sections, figures, quotations. Some were found by the professional, some by a junior, some by a machine. The question a tribunal, a client or a disciplinary committee will later ask is not "did you check?" but "show me what you checked, against what, and when". A verification ledger is the answer, kept while the work is done rather than reconstructed afterwards. It is also the instrument that makes Stage 4 of the Supervised Intelligence Model something that can be finished rather than something felt to be done.
The question
How does a professional know, and later show, that every source-dependent claim in a document was verified against the primary record?
The premise
Verification that is not recorded cannot be distinguished from verification that did not happen. A document is refuted by one unverified citation; the ledger exists so that there is never a citation whose status is unknown.
How it works
The ledger has one row per claim, not per source, because one source may carry several claims and each must be checked for its own proposition. Each row has six fields.
Proposition it carries. The exact claim in the work that depends on this source: the sentence, or the number, or the quoted words.
Citation as currently held. How the source is identified in the draft at this moment, including anything supplied by a machine that has not yet been checked.
Primary source to check. Where the original is: the court's own copy of the decision, the gazette or official consolidated text of the provision, the party's own published text of an agreement, the archived original of a document. The URL is given only if it is actually known, never constructed.
Points to confirm. A list, each a tick-box: the date, the paragraph number, the exact words of a quotation, the direction of the holding (that the case decides what it is cited for, not the opposite), the currency of the provision at the relevant date.
Falsifier. What would defeat the claim if found: an earlier counter-example to a "first"; an amendment to the provision; a later decision overruling; a certified copy that reads differently. The falsifier is searched for, not merely named.
Status. One of three. Unverified: nothing has been checked. Verified: checked against the primary record on a stated date, by the person who signs. Failed: the source does not say what the draft says; the dependent text is corrected before the work cites it again. A verified status means only that the source reachable on that date was checked; it does not survive a change in the source.
Two disciplines keep the ledger honest. Sources are archived when verified, with a code that becomes the filename, and paragraph numbers are recorded at the moment of archiving, not at the moment of drafting. And no claim leaves the ledger as unverified into a final document: an unverified claim is either verified, or removed, or left in the text as a visible placeholder that prevents the document from being treated as finished.
One case
The author's paper for the Solidarity AI 2026 conference was produced with a ledger of this shape, and its notes show what the ledger does in practice. Each authority is identified with the paragraph carrying each proposition (in the Supreme Court's order of 27 February 2026 in Gummadi Usha Rani: the four non-existent decisions at paragraph 2, the High Court's finding at paragraph 4, the declaration of misconduct at paragraph 7), with the primary PDF's location on the court's server and the archive code and date of the copy held. A quotation note records that the participle in the Supreme Court's direction to the Bar Council in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd, 2026 INSC 668, is "being", not "as". The ledger also carries the falsifier for the paper's flagship claim (that the direction is the first documented instance of a court ordering a statutory professional regulator to govern AI), and the claim is stated in the paper with that falsifier's limits attached.
The rule
One row per claim, six fields, three statuses. Nothing is cited from a row that is not verified, and a row is verified only against the primary record by the person who signs.
Not to be confused with
A bibliography lists sources; a ledger tracks claims and their status. A citation checker confirms that a case exists; a ledger confirms that it holds what it is cited for, at the paragraph given, as of a date. And an AI Work Record states what the machine and the human did in producing a document; the ledger is the part of that record that concerns sources.
Provenance and method.
Where this was published
- Shivam Shukla
- AI for Indian Advocates: The Practitioner's Standard for Supervised Intelligence (2026)
- Part III
- Shivam Shukla
- The Power to Say No
- SSRN 7454698 (2026)
- production notes and archive
Changelog
- 2026-09-25, version 1. First published.