The Preservative AI methodology is what Ghost Ledger runs on. It is a human-reviewed advisory methodology, supported by linguistic analysis, source provenance work and structured absence-mapping. It produces maps of what an institution's own documents have learned not to hold. It does not produce verdicts. The boundary is the boundary.
The methodology audits the narratives through which institutions describe their own decisions. It looks for what is missing from the official account: the warnings that did not survive into the current statement; the constituencies that fell out of the decision record; the agency that was diffused across the document; the chronology that was collapsed to make the current position look inevitable.
The methodology is built for the moment an institution needs to know whether a material account of its own actions can survive informed scrutiny. That moment arrives before a board declares effectiveness of material controls, before a regulator opens a correspondence, before a transaction closes, before a public statement is issued, before an apology is published.
Each lens is a separate structured query against the agreed document set. Each lens is documented. Each lens produces a labelled output. The eight lenses do not all apply to every engagement; the methodology selects and weights the lenses that are relevant to the specific decision chain under review.
Can the claims in the document be traced to a dated, named, retrievable source? Where the source chain breaks, the methodology marks the claim as a provenance exception.
Where a decision was made, who made it? Where an action was taken, who took it? Where a commitment was made, who owns its delivery? The methodology flags where the named individual has changed without a documented delegation, has left, or has never been named.
Who is materially affected by the decision chain, and is that constituency represented in the decision record? The methodology cross-references the corpus against third-party sources, regulator correspondence and public-interest reporting.
What was known, when was it known, who knew it, and when did the institutional account of the matter first record the knowledge? The methodology compares the chronology in the record to the chronology in the current account.
Has the language used to describe the decision chain shifted between documents and across time, in ways that materially alter the responsibility, certainty or implicature of the account? Suffixscape tracks suffixes, nominalisations, passivisations and modal shifts.
What claims in the current institutional account, if challenged, cannot be evidenced in the corpus or in any cited source? The methodology builds a register of every such claim, with the precise wording, the precise source citation and the precise gap.
Where the institution's account of the matter differs across documents, across time, or across institutional surfaces, the methodology documents the differences, ranks them by materiality, and traces each to its origin.
The methodology asks itself: what has the methodology left out? What patterns might the methodology be missing? What assumptions has the methodology made? The reflexive self-audit is part of every engagement and is logged in the versioned audit log.
Every finding produced by the methodology carries one of four labels. The label is part of the finding. The label is preserved in the audit log. The label is preserved in the report. A finding is not a finding if it does not carry its evidence and its label.
A reviewer can point to the document, the page, the sentence, the line. The label survives the factual-accuracy review.
The institution holds the missing evidence. The label is upgraded or downgraded after the institution responds.
The institution may produce evidence that resolves the inference, or may accept it as the institution's own reading of its own record.
The methodology is bounded. The boundary is part of the boundary statement on every report.
The methodology is governed by a public protocol titled When the Finding is Ugly. The protocol is referenced in every engagement letter. The protocol is the firm's binding commitment. The protocol is published in full at /ghost-ledger/protocol.
If the methodology surfaces a material concern supported by evidence, the finding will be reported to the named client sponsor. We will not sit on a finding to preserve the relationship. We will not delay a finding to a more convenient moment in the client's annual cycle. We will not allow a finding to be re-routed to a less senior recipient to soften its impact.
The client may record disagreement. The client may record additional context. The client may record a remediation plan. The client may not purchase the disappearance of a supported finding. This is the line the firm's credibility rests on.
No case study. No quotation. No "anonymised illustration". No conference talk. The engagement exists in a closed evidence room with a controlled disclosure protocol. The protocol is in the contract. The contract is enforceable.
The work may support governance, assurance, due diligence, regulatory engagement or board representation. It is not legal advice. It is not a compliance certificate. It is not a substitute for internal or external audit. The boundary is stated on every report and on every engagement letter.
The default Ghost Ledger engagement is a six-week, board-defence-grade review of one material decision chain. The engagement is bounded. The corpus is closed at week two. The price is fixed. The deliverable is named in the engagement letter.
Scope confirmed, corpus confirmed, named sponsor confirmed, protocol acknowledged, access opened.
Corpus ingested, anonymised, indexed. Model conditions set. Eight lenses weighted for the decision chain.
Methodology applied, findings provisionally identified, evidence chains built. Mid-engagement factual-accuracy note shared with sponsor at week 3.
Material Narrative Controls Report (the Ghost Ledger) delivered. 90-minute buyer briefing scheduled. Audit log sealed in joint escrow.
The boundary is part of the boundary. The boundary is stated in every output. The boundary is published here.
The methodology is not a machine that knows the truth; a score, ranking or judgement; a certification of legal, financial, regulatory or audit compliance; a replacement for internal or external audit; a tool for assigning moral blame; a way to manufacture the missing voice; an autonomous decision system; a generalised ethical-AI service.
The methodology is a human-reviewed advisory methodology, applied under a binding public protocol, producing maps of absence tied to evidence chains, with versioned audit logs and named senior reviewers accountable for the findings.
Three ways into the work, depending on where you are.
Tell us about the decision chain under review and the decision the review must support. Every enquiry is read by a named reviewer, who replies from reports@cogniosynthesisportal.uk. Nothing written here is published.