What may the agent rely on?
Witold Reichhart · witold.reichhart@capgemini.com
The second essential asks what the agent may rely on, and most firms we have seen answer with a library: documents, versions, a search index. The slides that follow go beneath the library. The check at w4 cannot read a library; it reads single claims, each one statement the agent may rely on, in the knowledge sense of the word, with a standing, and the standing it needs rises with the consequence of the act. The chapter stays on what the plane needs and says nothing of how a knowledge plane is built. Where a slide shows a case, the case is chosen at the top of the slide or with C; the rest is generic.
About 10 minutes. → or the space bar moves on, and through the steps first where a slide builds in steps. Click a line on the left to go straight to that slide.
M opens the contents, F finds a word, and T switches between light and dark.
Then the explorer: the map to play with, and the thinking behind the set on the website’s last page.
“Knowledge” in an AI programme is several jobs sharing one word, run at the same time rather than one after the other. Each has a place on the map, an owner, and a market that sells it.
Intake, extraction, access: the wordings, the procedures, the lists, the reading of the law, each with where it came from stamped at entry.
Entities resolved, relations wired, the estate made queryable: the clause joined to the contract, the contract to the customer, the rule to its source.
Runs across the other three: it scores, tiers, finds and types contradictions, applies decay, and enforces who may assert what.
The bounded package the agent receives for this case, and the record of having assembled it: what w4 reads as the knowledge relied on.
Most programmes we have seen buy the first pair and call it done. The check at w4 reads only what the second pair produces.
What the agent may rely on comes from three places, with different owners, different cadences, and a different way of failing. Part three sorts the second essential by what the firm holds, the law as read, policy and procedure, the contract’s terms, and the case; this slide sorts it by where it comes from, and the third origin is the one no document holds. The case itself arrives with the act, is read at w2 for that act, and belongs to no origin.
Regulation and the firm’s own reading of it: which articles apply to this process, what they require, what the firm has decided they mean. Written by compliance and legal, dated, and owned.
The steps, the thresholds, who may decide what, which system holds the truth. Much of it is explicit and never written: a specialist could say it if asked, and nobody asks; or it is written where no index looks, in defect logs and post-mortems. It includes the procedure of the knowledge function itself: how the firm captures, checks, and retires what it knows.
How a handler treats the unusual case, why the system breaks at month end, what the supervisor looks at first. It was never explicit, so it cannot be extracted; it spreads through shared practice and the traces practice leaves on the record, and it reaches an agent through those traces and the people who take its referrals.
The first two can be written down and checked. The third enters the estate through the traces of practice and the loop, which is why the people who take the referrals are part of the second essential.
A document is a container. The unit that can be checked, dated, and owned is the single assertion inside it: a claim. Here is one, from the case chosen above, with everything the check at w4 needs to know about it.
The excess on policy 88-1201 is €250.one assertion out of wording v14 §4.2, for one contract, with one date from which it holds
The €3,400 payment of claim 4471 is €3,650 less this excess. A similarity search returns the most similar text, here the restored v13; the check needs the claim that applies, with these fields on it.
Standing is what the check reads about a claim, and it is two things kept apart: how far the evidence supports relying on it, and who asserted it under what mandate. Folding the two into one confidence score is the mistake this model was built to remove.
Assessed at curation, by rule where it can be and by a person where it cannot, and reassessed as evidence and time move. The labels here, validated, current, and uncontradicted, are bands of a warrant that the reference architecture computes on several dimensions, in The Knowledge Layer.
A fact about provenance and permission, never a score. The signed wording, the list service, the head of product, a handler’s note: each asserts with a different mandate, and the mandate does not change with how true the claim is.
A claim from an impeccable source can be poorly evidenced, and a well-corroborated claim from a junior source is no less true. The two axes stay apart so the firm can see how a claim is weak, and fix the right thing.
The set has one hard rule on the agent. It needs a second, on the basis, at the same rank. The two bind different things, are decided at different places, and must never share a vocabulary.
An agent may be operated at a level of the ladder only when the evidence level the ladder names for it is met for that agent in that deployment.
An act of consequence tier n may commit only on claims whose standing meets the bar set for n, under the authority that tier requires. The check reads the weakest premise the act rests on, never an average.
Rule one sets how far the agent may go this quarter. Rule two decides whether this act may commit now. A firm needs both, and none we found sells either.
What the agent relies on decays at three speeds in the papers, and the set adds a fourth for the lists a case turns on. Someone has to keep all of it fit to act on. That is a function of the firm, with a cadence, its stages, and its rules, and it is the part of the second essential most programmes we have seen have not staffed.
| What | Decays in | Changed by | On the map |
|---|---|---|---|
| Domain mechanicshow a claim is adjudicated, how settlement works | years | a process redesign, a migration | e2 |
| Regulatory claimsthe firm’s reading of the law | months | a regulatory change, an enforcement action | e2, read at w2 |
| Operational realityprocedures, thresholds, client configurations | weeks | a reorganisation, a staffing change, a leadership change | w2, p2 |
| Live lists and pricessanctions designations, positions, rates: the set’s addition | hours | a publication, a market move | w2, refreshed by p2 |
The knowledge function runs them, with the source owners, compliance for the reading of the law, and the platform team for p2. Curation is the continuous stage: the quality engine, where quality is assessed and decay is applied, with half to one person per domain once it runs, a domain being one product line’s wordings, one list service, or one legal register, after a first load the pilot measures: an illustration to budget against, never a benchmark.
The knowledge function is the part of the firm that knows what the firm knows. The second essential at the enterprise altitude, e2, is its charter, and most firms we have seen have not written it.
Governed Intelligence Architecture names six failures of a knowledge estate. The defences are the lifecycle’s: the slide before states them as stages and rules, and this slide gives each its name, the case it would show in, and what the check can see of it. Every dossier runs the second; situation cards 8 to 12 run the other five.
| Failure | What enters the basis | Where it would show | The defence, in the lifecycle | The field the check sees it on | The control plane sees |
|---|---|---|---|---|---|
| Pollution | a claim that was never true: an extraction read as fact, the agent’s own letter read back as a source | motor claims · card 8 | the perimeter at ingest: an extracted claim enters provisional until the source owner confirms it against the clause; an agent’s output comes back labelled and provisional | provenance, and source authority | at the act: a provisional premise is below the bar, and the check refers |
| Staleness | true once, superseded since: wording v13 after the renewal onto v14, the list of 07.10 after the act of 09.10 | every dossier | decay: the validity window on every claim, currency read at the act, the decay classes refreshed on their own clocks | the validity window | at the act: currency fails on the weakest premise, and the check refers |
| Fragmentation | two vocabularies for one thing: excess here, deductible in the acquired portfolio, and the clause is never joined to the contract | motor claims · card 9 | drift detection on the estate: one thing, one name, held by the consolidate stage; a question that returns nothing for a whole portfolio is a finding | contradiction status, which stays empty: the two claims never meet | only the effect: a package complete and uncontradicted, a settlement without the excess, caught by the sample at p6 |
| Amnesia | superseded without the chain kept: a migration carries every claim and drops the links from v12 to v13 to v14, and nobody can say what applied to a contract on 4 September | motor claims · card 10 | governed disposition: a claim leaves demoted, archived, or sealed, with its trail kept, and a migration is a disposition of every claim at once | the validity window and the governance status, as the chain | only the effect: the record names a version with its window, and the estate can no longer show why it applied |
| Cascade | one bad claim and the ten that depend on it, each looking healthy alone: a stale list behind ten green flags | held payments · card 11 | dependency-aware warrant: a derived claim carries the warrant of its weakest origin; ten copies of one claim are one claim | evidential warrant, read through to the origin | at the act, if warrant follows the dependency; only the effect if each flag is read on its own timestamp |
| Structural distortion | a hub everything touches: 1,400 claims on “payment”, and the package for one act pulls in what does not belong | held payments · card 12 | topology: the package for an act is bounded by the act’s type, the mandate, and validity, never by similarity to a word; the hub is split | none of the seven: the package’s composition, inside the knowledge plane | only the effect: the check reads a declared basis whose window closed, and never why the agent read it |
Three of the six reach the check at the act, on a field of the claim. The other three show only in their effect: a settlement without the excess, a version the estate cannot explain, a package too wide. Those live inside the knowledge plane, in the part of the second essential the check never sees, which the knowledge function keeps fit with the source owners and the platform team.
One test tells a firm which side of this chapter it stands on. For any act of the last quarter: name the claims it relied on, and their standing at the moment it committed.
Claims with provenance, standing, validity, and an owner. What the firm holds. A firm that passes the test has this.
The handlers, the agents, and the loop reasoning over that content: seeing what matters in a case and choosing an act. A property of the firm, never of a model.
Capable agents acting on governed knowledge, with every consequential act admissible at the moment it commits. What the supervisor, the customer, and the board see.
A firm that governs its documents knows where its documents are. A firm that governs its claims governs its own intelligence, and can show, for any act, what it rested on. That joint is the set’s argument, and the thinking behind it is on the website’s last page.