Recording, for every agent action, what was done, on whose authority, under which policy, and why — retained for accountability the same way an audit trail is…
Generating a first-draft, step-sequenced procedure from a captured expert interview or a raw process description — the bridge from elicited tacit knowledge…
Checking that a generated answer is genuinely traceable to the passages it cites, rather than accepting a plausible-sounding synthesis at face value — the…
Continuous, algorithmic verification and un-verification of owned content based on behavioural, content, and analytical signals — running nightly rather than…
An agent actually carrying out the recognised action — editing a document, updating a ticket, posting to a project tool — within an explicitly defined autonomy…
Automatically drafting a candidate knowledge article from a resolved ticket, answered question, or closed incident, without requiring the person who solved it…
Prompting the person doing the work, at the natural moment — right after closing a ticket, right after answering a question — to confirm or save the captured…
Converting heterogeneous ingested content — documents, messages, tickets, code — into a structure that can actually be searched: chunking long documents into…
Recording where a piece of knowledge originally came from, what it has been derived from or superseded, and its full chain of custody — independent of whether…
Deciding when superseded or obsolete content should be retired rather than merely flagged, removing it from active retrieval while preserving it for audit or…
Capturing expertise as an ongoing habit rather than only when someone announces they are leaving — ingesting documents, recording explanations, and capturing…
Interviewing multiple people about the same process and automatically flagging where their accounts disagree — revealing team misalignment before it becomes an…
Deciding, when the same content is indexed by more than one search platform simultaneously, which platform is authoritative for a given source, avoiding…
Detecting that two live, currently published sources disagree with each other — the project plan says one date, the weekly report says another — and surfacing…
Deciding which downstream system a recognised action should land in — a document, a project tool, a ticket system, a CRM record — and dispatching it there. The…
Deciding what to formalise and promote into permanent, governed knowledge based on actual reuse frequency and demand, rather than a top-down authoring plan —…
Identifying near-duplicate or substantially overlapping content across different sources — distinct from CAP-122's contradiction detection, because duplicates…
Recognising that an incoming question or ticket matches one already resolved, and surfacing the existing resolution instead of letting the problem be solved…
Restricting who may author, edit, or even see content based on which client, matter, or engagement relationship a person is staffed to — independent of their…
Identifying and surfacing the person who is the actual authority on a given topic, so the question can be routed to them rather than requiring a document to…
Exposing verified, permission-aware knowledge to external AI systems — a general-purpose assistant, a third-party agent — so those tools query through a…
Producing a direct natural-language answer to a question, synthesised from retrieved passages and cited back to its source — rather than returning a ranked…
Defining and enforcing the checkpoint at which a proposed action requires human confirmation before it executes, independent of what the action actually is.
Applying a consistent structure — fields, format, tags — to captured content at the point of capture, rather than leaving it as an unstructured byproduct that…
Connecting to and pulling content from the systems where knowledge actually accumulates — wikis, chat, drives, tickets, email, code repositories — rather than…
Assigning an accountable owner and verifier to each piece of content, including AI-suggested routing based on who created or most recently touched the…
Ensuring retrieval only ever surfaces content the requesting person is authorised to see — normalising heterogeneous source-system permission schemes…
Surfacing relevant knowledge inside the tool where work is already happening — Slack, Teams, a handheld task app — before the person searches for it, rather…
Confirming that a worker actually completed a procedure step correctly — a photo confirmation, a check, a brief comprehension check — rather than assuming…
Version-controlling procedural content with formal change approval, tied to safety or compliance requirements rather than general content governance — a…
Combining multiple individual signals — verification status, staleness, usage, source authority, contradiction flags — into a single trust indicator surfaced…
Ordering candidate results by actual relevance to intent rather than keyword overlap alone — combining lexical matching with semantic/vector similarity, and…
Requiring and recording formal approval from a qualified, authorised role before a procedure change goes live — a named-authority gate rather than a general…
Deciding what learning from a walled engagement, project, or business unit can be generalised, sanitised of identifying detail, and promoted into a reusable…
The human-governed review cycle: a verifier is notified when content is due for review, confirms it is still accurate or updates it, and the content's verified…
Tracking which queries succeed, which fail, and which topics have no good answer anywhere in the index — turning search behaviour itself into a signal for…
Identifying content that changes frequently at its source, or has gone unreviewed for an unusually long time, and down-weighting or flagging it accordingly —…
Creating standard operating procedures and work instructions as explicitly ordered steps, rather than freeform prose — a structural requirement generic…
Deliberately creating a piece of knowledge content — a card, an article, a policy — with an accountable author, including AI assistance that drafts,…
Running a guided, AI-led interview with a subject-matter expert to surface knowledge that has never been written down — not a static exit-interview form, but…
Flagging roles or individuals whose departure would create a critical, undocumented-knowledge gap — tying retirement timelines and single-point-of-failure…
Recognising, during retrieval, trust assessment, or contradiction detection, that a finding warrants action — the origin test made concrete: the action could…
Surfacing the correct step of a procedure at the correct point in a physical task — on a handheld device, embedded in equipment, on a shop floor or in a…
Evaluating agent actions at the moment they are attempted, not after the fact — a rule resolves to block, filter, or flag for review, with the most restrictive…
Identifying content that is heavily reused but consistently needs correction or follow-up — the demand signal that something is high-value but currently…
Building a queryable representation of one person's captured expertise — from interviews, recordings, and documents — that others can consult after that person…