Land source material without rewriting what was received.
ActivateDigital methodology
Public architecture
A governed architecture for product truth.
ActivateDigital resolves fragmented, differently scoped and sometimes conflicting product information into a versioned, auditable product record. The architecture is designed to operate across different product domains and attribute models under explicit evidence, identity, dependency, method and governance controls. Where the basis is insufficient, the system preserves uncertainty rather than manufacturing completeness.
PUBLIC PRINCIPLES + CONTROL MODEL / DETAILED RESOLVER IMPLEMENTATION PROPRIETARY
01 / TRUTH
Resolve what can be established.
Product truth is constructed from observations that have been bound, qualified and resolved under an attribute-specific contract.
OBSERVE→BIND→QUALIFY→RESOLVE
02 / GOVERN
Keep the basis attached.
A result remains connected to the evidence, method, state and version that produced it.
provenanceconflictsversionsrevalidation
permissionsreasons
03 / ACTIVATE
Project trusted identity.
The governed record can be shaped for multiple destinations without reconstructing product truth each time.
passportretailAPImeasurement
protectionexperiences
5control pillars
10method families
5decision states
13system laws
1governed architecture
Resolution architecture
Truth is resolved, not merged.
A product record can contain facts, claims, documents, model outputs and conflicting observations at different scopes. The system therefore separates acquisition from qualification, qualification from decision and decision from publication.
Attach observations to the product, variant, component, organisation or market they actually describe.
Determine source kind, authority, currency and whether the observation may influence the attribute.
Apply declared dependencies, approved methods and bounded inference under the attribute contract.
Return a value or refusal state with reason, provenance, versions and retained conflicts.
Expose only the permitted representation for the destination and audience.
The architecture is dependency-aware rather than uniformly linear. Different attributes can take different governed routes and can stop independently when their evidence standard is not met.
Resolution framework
Five control pillars. Ten governed method families.
The public framework describes the recurring classes of operation used to establish and maintain product truth across domains. It is intentionally one level above domain-specific attribute models and the proprietary resolver implementation.
01 / QUALIFY
Establish admissibility.
Determine what may legitimately influence the truth being resolved.
MF01
Evidence QualificationAssess whether an observation is attributable, relevant and admissible for the proposition in question.MF02
CanonicalisationConvert heterogeneous observations into controlled representations without severing them from source evidence.02 / BIND
Establish context.
Know exactly what the evidence and dependencies belong to.
MF03
Identity BindingBind observations to the correct product, variant, component, actor, scope and version before they influence a result.MF04
Dependency ResolutionConsume governed upstream truth through declared dependencies rather than independently reconstructing the same fact downstream.03 / RESOLVE
Establish supportable truth.
Use governed pathways while keeping unsupported certainty out.
MF05
Progressive ResolutionApply an appropriate governed resolution pathway and stop when the required evidentiary basis is reached or cannot be reached safely.MF06
Controlled InferencePermit derivation or model-assisted proposals only within defined decision boundaries and evidence controls.04 / GOVERN
Control the decision.
Preserve disagreement, uncertainty and the basis for every state.
MF07
Conflict AdjudicationRetain competing credible observations until governed qualification and precedence controls can resolve or escalate them.MF08
Decision State AssignmentReturn an explicit truth state rather than allowing a candidate or missing value to masquerade as certainty.05 / ASSURE
Keep truth valid.
Carry the basis forward and reconsider it when that basis changes.
MF09
Provenance BindingKeep the result connected to the evidence, transformation, method and version that produced it.MF10
RevalidationIdentify when material changes to evidence, dependencies, methods or governing conditions require targeted review.Public method families, not an internal method catalogue. Domain implementations can contain different attribute models, method implementations, evidence requirements and dependency structures beneath this framework. Those details remain controlled or proprietary.
Public resolution calculus
Each attribute is a controlled resolution function.
The expression below is a public abstraction of the operating model, not implementation syntax. It makes the important constraint visible: a product value is a function of admissible evidence, bound identity, governed dependencies, an approved method/version and system governance.
Attribute truth function
Tav = Ra(Ea, I, Da, Mav | G)
Ra is the resolver contract for attribute a. The exact resolver, route precedence, thresholds, source orchestration and rule configuration are not disclosed publicly.
state(Ta) ∈ { ESTABLISHED, CHECK, UNKNOWN, NOT_APPLICABLE, CONFLICTED }
Public semantics
Tav
Governed truth resultThe attribute result at an explicit resolver or rule version.
Ea
Admissible evidenceQualified observations whose identity, scope and evidence conditions permit use.
I
Bound identityThe product or actor identity against which evidence is interpreted.
Da
Governed dependenciesUpstream truth consumed through the contract, not independently re-solved downstream.
Mav
Approved method + versionThe named normalisation, calculation, register or model method where one applies.
G
Governance constraintsEvidence, uncertainty, conflict, version, publication and lifecycle rules.
This notation documents the architecture. It is intentionally not a disclosure of the underlying resolver implementation.
OUTPUTdefinition + semantic scope
REQUIRESdeclared upstream truth + contextual inputs
EVIDENCEtyped observations + admissibility conditions
METHODnamed method / model / register + version
PROHIBITSunsupported inference + invalid substitutions
RETURNSvalue + state + reason + provenance
AUDITroute + evidence ids + transform + versions
LIFECYCLErevalidation on material basis change
Evidence architecture
Evidence is typed before it is trusted.
Where information came from is not enough. The model retains what kind of evidence it is, what it is scoped to, when it was observed and what proposition it can support. Transforming evidence does not erase its origin.
Evidence object
Product master
Structured catalogue observations from systems such as PIM, ERP, commerce platforms or controlled uploads.
Evidence object
Technical document
Original document, revision, hash, extracted observations and identity binding remain connected.
Evidence object
Certificate
Scheme, certificate identifier, holder, scope, dates and verification state are retained separately.
Evidence object
Regulatory declaration
Declared proposition, issuing actor, scope, basis, date and any supporting identifiers remain separately attributable.
Evidence object
Operator master
Legal entity, roles, issued identifiers and registrations at organisation and market scope.
Evidence object
Merchant assertion
The exact question, stated value, actor, scope, timestamp and supersession relationship.
Evidence object
External product page
Fetched source, retrieval time, identity binding and exact structured field or retained excerpt.
Evidence object
Model execution
Method/version, input fingerprint, request payload, raw response, output and execution time.
Evidence qualification rule
Discovery is not evidence. A search result can locate a candidate source. It does not establish a product fact. A fetched source may become evidence only after the applicable identity, scope and evidence controls are satisfied.
Resolution strategy
Progressive evidence, bounded intervention.
The operating principle is minimum customer input, maximum defensible record. The system attempts to use evidence already held or obtainable through governed routes before asking a person to recreate it. Human input is treated as a typed assertion or bounded decision, not as a shortcut around the evidence model.
ROUTE FAMILYHeld evidenceConnected catalogue and product observations already under control.
ROUTE FAMILYExisting documentsTechnical, supplier, certificate and declaration evidence already available.
ROUTE FAMILYGoverned derivationDeterministic dependency or calculation routes using established upstream truth.
ROUTE FAMILYAuthoritative systemsRegisters, APIs and named official or controlled models where applicable.
ROUTE FAMILYExternal discoveryTargeted discovery used to locate exact sources, never treated as truth by itself.
ROUTE FAMILYBounded confirmationA controlled question, assertion or evidence request for the genuine residual gap.
These are public route families, not a disclosure of exact route order, evidence precedence, source weights, thresholds or orchestration.
Decision state model
Completeness is not the objective. Defensibility is.
The resolver can decline to establish a value. That is a controlled outcome, not a system failure. Uncertainty, conflict and applicability are represented explicitly so a plausible answer cannot silently acquire the status of truth.
ESTABLISHED
Evidence standard met
Supported by an accepted route and the evidence required by the attribute contract.
CHECK
Bounded decision needed
A candidate, partial truth or missing gate requires confirmation before establishment.
UNKNOWN
Not safely established
The value is absent, insufficiently supported or not safely inferable from what is known.
NOT_APPLICABLE
Positive basis required
Non-applicability is asserted only when the applicability logic itself is supported.
CONFLICTED
Credible evidence disagrees
Incompatible credible observations remain visible until an explicit rule or correction resolves them.
The ability to refuse is part of the architecture.
Missing units are not silently defaulted. Unknown shares are not fabricated to make a total reach 100%. “Not found” is not converted into “none”. A candidate is not promoted because it looks plausible. The system is designed to preserve the boundary between what is known, what is proposed and what remains unresolved.
System laws
Principles guide the design. Laws constrain the machine.
These invariants are deliberately stronger than feature descriptions. They define behaviour that a compliant resolver is not permitted to violate, regardless of product category or downstream destination.
Raw evidence is append-only.
The landed observation remains available as received. Cleaning, normalisation and correction create governed downstream representations rather than rewriting source history.
raw(t+1) never overwrites raw(t)
Identity precedes influence.
An observation cannot influence product truth until it is bound to the identity and scope it actually describes.
influence(o,Tₐ) ⇒ bound(o,I,scope)
Discovery is not evidence.
A mechanism that locates a source does not, by that act alone, satisfy the evidentiary standard for a product fact.
discover(x) ≠ qualify(x)
Proposal is not decision.
A parsed candidate, inferred mapping or model proposal remains distinct from an established value until the applicable controls are satisfied.
candidate(x) ≠ established(x)
Uncertainty must survive.
Insufficient evidence cannot be converted into an apparently normal value merely to complete a record.
¬sufficient(Eₐ) ⇒ stateₐ ≠ ESTABLISHED
Scope cannot widen silently.
A product, variant, component, organisation or market assertion cannot acquire a broader scope than its evidence supports without an explicit rule.
scope(output) ⊆ supported_scope(evidence)
Downstream consumes governed upstream truth.
A dependent resolver reads the governed result of its dependency. It does not reach around that contract and independently re-solve the upstream fact.
Rᵦ consumes Tₐ, not raw(Eₐ)
Arrival order does not determine truth.
Precedence is governed by explicit qualification and authority rules, not by which observation arrived last.
precedence ≠ f(arrival_time)
Conflict is retained until resolved.
Competing credible observations remain part of the record until a governed rule selects, rejects, supersedes or escalates them.
credible(e₁) ∧ credible(e₂) ∧ e₁≠e₂ ⇒ retain(e₁,e₂)
Every result keeps its basis.
Transformation, calculation, normalisation and publication must not sever a result from the evidence, method and version that produced it.
Tₐ ⇒ provenance(Tₐ)
Methods do not drift quietly.
Method, rule and model versions are explicit. A changed basis must be visible and capable of triggering review rather than silently changing an established answer.
result carries method@v + rule@v
Dependency change can invalidate truth.
When an upstream fact, evidence item, method, regulatory basis or merchant correction changes materially, affected downstream outputs can be invalidated and recomputed.
Δbasis ⇒ invalidate(affected downstream)
Not applicable requires a basis.
Absence of evidence is not evidence of non-applicability. The NOT_APPLICABLE state requires a positive applicability determination.
NOT_APPLICABLE ⇒ applicability_proof
Audit + lifecycle
A truth result is a record of how it was reached.
The governed output is not just a value. It carries enough structured context to explain the selected route, reproduce the basis of the decision and determine when the result has become stale.
Audit vector
Aa = <
attribute,
selected_route,
method@version,
input_evidence_ids,
source_refs,
normalisation_or_calculation,
state + reason,
resolver@version,
superseded_or_conflicting_evidence
>
attribute,
selected_route,
method@version,
input_evidence_ids,
source_refs,
normalisation_or_calculation,
state + reason,
resolver@version,
superseded_or_conflicting_evidence
>
Revalidation triggers
A governed dependency changes.
Source evidence is revised or replaced.
A method or model version changes.
The regulatory or methodological basis changes.
A merchant correction supersedes an earlier assertion.
Revalidation is targeted. A changed input should be capable of invalidating the outputs that depend on it without forcing unrelated truth to be reconstructed.
Methodology disclosure boundary
Open about the control model. Deliberately private about the resolver.
A methodology should be inspectable without making the implementation reproducible by a third party. ActivateDigital therefore separates the public architecture from controlled evidence and proprietary resolution logic.
Public
Architecture + principles
- truth states and their semantics
- evidence and provenance principles
- attribute contract structure
- system laws and lifecycle controls
- output definitions where disclosure is appropriate
Controlled
Evidence + actor detail
- source evidence and documents
- actor and supplier information
- restricted provenance detail
- assurance material shared under suitable controls
Proprietary
Resolution implementation
- exact route precedence and source weighting
- thresholds and decision parameters
- resolver orchestration and decision configuration
- model payloads, mappings and internal knowledge packs
- acceptance tests, test corpora and implementation code
Regulators and assurance partners. Appropriate technical detail can be made available where necessary for regulatory engagement, assurance or due diligence, subject to suitable confidentiality and information-access controls.
ActivateDigital.ai · Product truth methodology
Public architecture · Resolution implementation proprietary