Information Typing Model

Raw to Knowledge classifies every extracted answer using Robert Horn's Information Mapping taxonomy. This model governs how answers are validated, stored, and retrieved.

Horn Types

FACT

A declarative assertion about the state of the world. Facts carry the highest governance obligations:

Requires an assertion subtype (see below)
Requires assertion grammar (subject, predicate, object)
Requires a human validator with domain authority over the claim
Flagged as requires_review when confidence < threshold

Example: "The platform supports localized help sites when locale-specific branches are configured."

CONCEPT

A definition or explanation of what something is. Concepts answer "what is X?" questions.

Example: "A portal target is a configured output destination that defines how content is published to a specific audience."

PROCEDURE

Step-by-step instructions for how to accomplish a task. Requires validated, tested steps before approval.

Example: "To configure a portal target: 1. Open Portal Settings. 2. Select Add Target. 3. Assign a locale code."

PROCESS

A description of how the system behaves — as opposed to how a user acts. Often uses passive voice with a system subject.

Example: "When a content map snapshot is published, the system automatically validates DITA constraints and generates the output package."

PRINCIPLE

Normative guidance — best practices, design decisions, or policies. Must not be treated as a factual claim.

Example: "You should always version-lock portal target configurations before a major release."

REFERENCE

A list, table, or lookup resource. High noun-phrase density; low verb content.

Example: "The platform supports the following output formats: HTML, PDF, Markdown, and custom templates."

TROUBLESHOOTING

A problem–cause–solution pattern. Routes to technical validators.

Example: "If the portal rebuild fails, check the error log for publishing errors and verify the content map is valid."

RECOMMENDATION

Context-specific advice given in the moment. Distinct from Principle because it applies to a specific customer situation.

Example: "In your case, I recommend using conditional text rather than branching given the small team size."

OPEN_ISSUE

Unresolved or uncertain — "I don't know", TBD, escalated. Never publishable. Always requires review. Routes to the appropriate specialist queue.

Example: "The retention policy for deleted content is still TBD pending legal review."

Assertion Subtypes (FACT only)

When the primary type is FACT, a second classifier determines the assertion subtype. Six subtypes are implemented in Milestone 2.

Subtype	Definition	Example signal
`existence`	Asserts that something exists or is available	"is available", "there is", "out of the box"
`capability`	Asserts that something can perform an action	"supports", "can", "allows", "enables"
`configuration`	Asserts a default, setting, or configurable value	"defaults to", "configured", "set to"
`state`	Asserts the current runtime or lifecycle state	"is currently", "triggered by", "status"
`constraint`	Asserts a limit, requirement, or restriction	"requires", "must", "maximum", "prevents"
`temporal`	Asserts a time-bound or version-scoped fact	"as of", "since", year pattern, "version X.Y"

Assertion Grammar

For FACT candidates, the AssertionGrammarExtractor produces structured fields:

Subject   → The entity the assertion is about
Predicate → The relationship or property asserted
Object    → The value or entity on the receiving end
Scope     → Boundary within which the assertion holds
Condition → Prerequisite, if any
Temporal  → Date or version range of validity

Populated grammar enables downstream contradiction detection (Milestone 7): two facts with the same Subject + Predicate + Scope but different Object are flagged as conflicting.

Governance Matrix

Type	Subtype required	Grammar required	Validator authority	Auto-approve eligible
FACT	Yes	Yes	Domain SME	No
CONCEPT	No	No	Content strategist	Yes (low confidence)
PROCEDURE	No	No	Technical writer	No (must be tested)
PROCESS	No	No	Engineer	No
PRINCIPLE	No	No	Principal/Architect	No
REFERENCE	No	No	Content strategist	Yes
TROUBLESHOOTING	No	No	Support engineer	No
RECOMMENDATION	No	No	Consultant	No
OPEN_ISSUE	No	No	— (not publishable)	No