Repositioning to Modeled Information Format

ADR-010: Repositioning to Modeled Information Format

Status

Accepted

This decision shipped in MIF v1.0.0. It is the positioning that ADR-009 (OKF compliance as a superset) presupposes: that decision describes how MIF’s independent identity relates to OKF, and that independent identity is established here.

Context

Background and Problem Statement

Prior to v1.0.0, MIF stood for Memory Interchange Format. Its identity was memory-centric: the format existed to serialize and interchange the persistent memory of AI agents, and its concept files carried a .memory.md infix with .memory.json sidecars. The cognitive vocabulary (semantic / episodic / procedural memory, the forgetting curve, decay) was the spine of the format itself, not an application of a more general model.

In practice the mechanisms MIF defined were not memory-specific. A typed relationship graph, provenance and trust tiers, validity windows and freshness, contradiction and supersession semantics, and a Markdown-canonical / JSON-LD-derived projection are all general knowledge-modeling concerns. AI memory is one important consumer of those mechanisms, not their definition. Naming the whole format after a single application understated its scope and coupled the format’s identity to one domain’s vocabulary.

The architectural question for v1.0.0 was therefore one of identity and scope: what is MIF, fundamentally? — a memory format that happens to be general, or a general content model whose first worked application happens to be memory.

Current Limitations

• The “Memory Interchange Format” name advertised a narrower scope than the format actually delivered, discouraging adoption for non-memory knowledge modeling (catalogs, documentation graphs, domain ontologies).
• Memory-specific vocabulary was woven into the core, leaving no clean seam between the general model and a domain’s interpretation of it.
• Without a designated extension mechanism, every new domain would have had to fork or reinterpret the core rather than layer on top of it.
• The repositioning ADR-009 depends on — MIF as an independent specification that is OKF-compliant by superset — had no documented identity to point to.

Decision Drivers

Primary Decision Drivers

1. Scope honesty: the format’s name and identity must match what it actually models — general agent-readable knowledge, not only memory.
2. Extensibility seam: there must be a clean boundary between the general model and any single domain’s interpretation, so domains layer on rather than fork.
3. Independent identity: MIF must have a stable, domain-neutral identity that the OKF-superset relationship (ADR-009) can presuppose.

Secondary Decision Drivers

1. Adoption surface: a general content model invites a broader set of adopters than a memory-only interchange format.
2. Proof of the extension model: the repositioning is only credible if at least one real domain profile demonstrates that the seam works.

Considered Options

Option 1: Keep “Memory Interchange Format” / memory-centric identity

Description: Retain the original name and keep AI memory as MIF’s defining identity. General-purpose use remains an implicit, undocumented possibility.

Technical Characteristics:

• Cognitive vocabulary stays in the core as first-class identity.
• No formal profile layer; domain reuse is ad hoc.

Advantages:

• Established name with existing recognition; no rebrand cost or churn.
• Narrow, immediately legible scope — adopters know exactly what it is for.
• No need to disentangle memory vocabulary from the core.

Disadvantages:

• The name understates the format’s actual generality, suppressing non-memory adoption.
• No clean seam between general model and domain interpretation; new domains must reinterpret or fork the core.
• Provides no domain-neutral identity for the OKF-superset relationship to rest on.

Risk Assessment:

• Technical Risk: Low to keep; the format works.
• Schedule Risk: Low. No work required.
• Ecosystem Risk: Medium. Misnaming caps the addressable use cases and entangles core evolution with one domain’s vocabulary.

Option 2: Reposition to “Modeled Information Format” with AI memory as a profile (chosen)

Description: Re-expand the acronym to Modeled Information Format. MIF’s identity becomes a general, opinionated, OKF-compliant content model for agent-readable knowledge. AI memory is demoted from MIF’s identity to its first domain profile (profiles/ai-memory/), which builds on the core without changing it.

Technical Characteristics:

• The core specification frames mechanisms in neutral knowledge-management terms.
• A domain profile is a normative overlay: it adds entity types, traits, defaults, and rationale on top of the mif-base model, and everything in the core applies unchanged.
• The AI-memory profile keeps the original cognitive-memory rationale (e.g. the forgetting curve) as a domain interpretation of a neutral core mechanism (e.g. validity windows and freshness).

Advantages:

• Name matches actual scope; broadens the adoption surface beyond memory.
• Establishes a clean, reusable profile seam — domains layer on, never fork.
• Gives MIF the domain-neutral identity that ADR-009’s superset relationship presupposes.
• Ships a worked profile that proves the extension model rather than asserting it.

Disadvantages:

• One-time rebrand cost: name, narrative, and migration of existing bundles.
• Requires disentangling memory vocabulary from the core into the profile.

Risk Assessment:

• Technical Risk: Low. The mechanisms are unchanged; this is a re-layering and renaming, validated by an existing profile.
• Schedule Risk: Low. Delivered within the v1.0.0 release.
• Ecosystem Risk: Low. The profile model preserves all memory use cases while opening the format to new domains.

Option 3: Generic format with no reference profile

Description: Generalize MIF to a domain-neutral content model but ship no worked domain profile — leave AI memory (and every other domain) as future, externally-defined extensions.

Technical Characteristics:

• Core model only; the profile mechanism is described but unexercised.

Advantages:

• Maximally neutral core, unburdened by any single domain’s vocabulary.
• Smallest core surface to maintain.

Disadvantages:

• No proof the extension model actually works — a profile seam that nothing uses is unvalidated.
• Abandons MIF’s strongest, most-developed use case (AI memory) instead of carrying it forward as the canonical example.
• Leaves adopters without a reference to model their own profiles on.

Risk Assessment:

• Technical Risk: Medium. An unexercised extension seam may not survive contact with a real domain.
• Schedule Risk: Low.
• Ecosystem Risk: High. Adoption stalls without a worked example; credibility of the “general model” claim is unproven.

Decision

MIF is repositioned as the Modeled Information Format: a general, opinionated, OKF-compliant content model for agent-readable knowledge. AI memory is not MIF’s identity; it is MIF’s first domain profile, located at profiles/ai-memory/ and layered on top of the core mif-base model.

The core specification defines the general model — the .md / JSON-LD knowledge representation, namespaces, the temporal model, embeddings, provenance, and typed relationships — in domain-neutral terms. A domain profile is a normative overlay that adds entity types, traits, defaults, and rationale; everything in the core applies to the profile unchanged. The AI-memory profile re-interprets neutral core mechanisms (e.g. validity windows and freshness) with their original cognitive rationale (e.g. the forgetting curve), demonstrating the seam works.

This is the independent identity that ADR-009 presupposes: MIF is OKF-compliant as a superset because MIF is its own specification with its own scope — a scope this repositioning makes explicit.

Consequences

Positive

1. Scope-accurate identity: the name and abstract describe a general content model, matching what MIF actually delivers.
2. Reusable profile seam: domains extend the base without forking; the AI-memory profile is the reference pattern.
3. Foundation for ADR-009: the OKF-superset relationship now rests on a stated, domain-neutral identity.
4. Preserved memory use case: nothing about AI memory is lost — it is carried forward as the canonical first profile.

Negative

1. Rebrand churn: the acronym expansion, narrative, and existing bundles required migration (handled by the v1.0.0 migration tooling).
2. Two layers to keep coherent: the core must stay domain-neutral while the profile carries the domain vocabulary, an ongoing editorial discipline.

Neutral

1. Cognitive vocabulary relocates, it does not disappear: semantic / episodic / procedural and the forgetting-curve rationale move from the core’s identity into the AI-memory profile’s interpretation of neutral mechanisms.

Decision Outcome

The reposition-to-general-with-a-first-profile approach achieves the primary drivers: scope honesty (the name now matches the model), a clean extensibility seam (profiles/ai-memory/ layers on mif-base), and an independent identity that ADR-009 builds on. Mitigations:

• The positioning is stated normatively at the top of SPECIFICATION.md (the “AI memory is the first domain profile of MIF, not its identity” callout) and carried into the v1.0.0 CHANGELOG.md entry and the migration guide.
• The extension model is proven, not merely asserted, by the shipped profiles/ai-memory/ profile, which opens by restating that AI memory is a profile and not MIF’s identity.

Related Decisions

• ADR-009: OKF Compliance as a Superset (Pinned OKF v0.1) — builds on the independent, domain-neutral identity that this repositioning establishes.
• ADR-001: Cognitive Triad Taxonomy — the semantic/episodic/procedural taxonomy becomes the base content model on which the AI-memory profile sits.

Links

• docs/okf-conformance.md — the pinned OKF v0.1 interoperability surface that MIF’s general content model fills (see ADR-009).

More Information

• Date: 2026-06-18
• Source: SPECIFICATION.md L15 and the L24-L27 callout; profiles/ai-memory/SPECIFICATION.md; CHANGELOG.md v1.0.0 entry; MIGRATION.md.
• Related ADRs: ADR-009, ADR-001

Audit

2026-06-18

Status: Compliant

Findings:

Finding	Files	Lines	Assessment
Acronym re-expanded to “Modeled Information Format” as the project identity	SPECIFICATION.md	L15	compliant
Callout: “AI memory is the first domain profile of MIF, not its identity (see profiles/ai-memory/)“	SPECIFICATION.md	L24-L27	compliant
AI-memory profile exists as a normative overlay on mif-base	profiles/ai-memory/SPECIFICATION.md	L1-L19	compliant
v1.0.0 changelog records the reposition and the memory-to-profile demotion	CHANGELOG.md	L10-L12	compliant
Migration guide states MIF is repositioned and AI memory is now a profile, not its identity	MIGRATION.md	L5-L6	compliant
ADR index names the project “Modeled Information Format (MIF)” (remediated 2026-06-18; previously stale)	adr/README.md	L7	compliant

Summary: The repositioning to “Modeled Information Format” and the demotion of AI memory to MIF’s first domain profile are stated normatively across the specification, the AI-memory profile, the changelog, and the migration guide. The ADR index (adr/README.md L7), which previously still named the current project “Memory Interchange Format (MIF)”, was corrected to “Modeled Information Format (MIF)” during this same change set. (The historical CHANGELOG.md 0.1.0-draft entries that name “Memory Interchange Format” are correct in their historical context and are not flagged.)

Action Required: None.