MCW Constitution¶

Version 1.1 · W. Alec Akin · April 2026, amended July 2026 (see Amendments)

This document governs the canonical use of MCW framework constructs. It is not a roadmap or a philosophy — it is a set of binding invariants designed to keep cross-study comparison possible as the framework is cited, extended, and challenged.

Preamble¶

A framework contribution derives its scientific value from definitional stability. If "MCW" means something different in each paper that cites it, cross-study comparison becomes impossible and the vocabulary collapses into noise.

This Constitution encodes the rules that must hold for a work to be considered a legitimate MCW-framework contribution. Works that do not satisfy these rules may still be valuable — they simply cannot claim compatibility with the canonical framework without declaring their departures.

Article I — Definition Immutability¶

The canonical definitions of the following constructs are fixed in docs/glossary.md:

MCW (Meta-Context Window)
HCW (Human Context Window)
ACW (Artificial Context Window)
IU (Information Unit)
The six failure modes (Drift, Asymmetric State Advancement, False Alignment, Overcompression, Constraint Opacity, Repair Suppression)
The five repair operations (Re-grounding, Decompression, Re-weighting, Disambiguation, Synchronization)
The four measurement proxies (H, R, D, M)

Rule: Extensions and implementations that redefine any of these terms must declare the departure explicitly, in the form:

"This work uses a modified definition of [term]. The canonical MCW definition is [cite glossary]. The modification is: [description]. Rationale: [reason]."

Works that redefine canonical terms without declaration are not compatible with this framework, regardless of citation.

Article II — Layering Invariant¶

The OSI Layers of Understanding (Layers 0–5) define a strict accessibility ordering:

Rule: No construct may be introduced at Layer N unless an equivalent, compressed form of that construct already exists at Layer N−1. Formalization that has no intuitive layer equivalent is not a valid MCW extension — it is a new framework that happens to use MCW terminology.

Corollary: High-layer formalizations (Layer 3–4) must trace downward to Layer 0 natural language. If a researcher cannot explain the construct in plain language, the formalization is premature.

Article III — Substrate Independence¶

(Amended in v1.1 — prior text preserved in Amendments.)

IUs and all six failure modes are substrate-independent: they apply to human–human, human–AI, and AI–AI coordination systems without modification.

The MCW construct itself is canonically defined for HCW–ACW coupling. This is a declared scope restriction, not an accident of phrasing. Generalizations of MCW to human–human or AI–AI substrates are conjectured to be coherent [L1] but must be filed as declared extensions under Article V, with their own falsification conditions. The framework's sharpest substrate-specific prediction — coordination effects arising from the HCW-continuous / ACW-discrete asymmetry — is testable only in Human↔AI conditions and is deliberately protected by this scoping.

Rule: An implementation or study that restricts MCW constructs to a specific substrate (e.g., a particular model, modality, or platform) must:

Declare the restriction as a scope constraint, not a definitional revision
Not claim to test the general framework — only the substrate-specific instance
Not present substrate-specific findings as evidence for or against the general framework without explicit generalizability analysis

Anti-pattern: "We tested MCW in GPT-4 and found X" is a substrate-specific observation. It is not evidence that "MCW predicts X" in general.

Article IV — Epistemic Floor¶

All empirical claims about MCW dynamics must declare their evidence layer:

Evidence Layer	Description
L0 — Illustration	Naturalistic observation, no controls; used to demonstrate recognizability only
L1 — Practitioner observation	Extended personal use; subject to positionality bias; not generalizable
L2 — Designed pilot	Structured observation with hypothesis; single observer or small N; no inter-rater reliability
L3 — Pilot with reliability	Multi-rater pilot; inter-rater reliability reported; generalizable with caution
L4 — Controlled study	Random assignment, control conditions, validated instruments; generalizable with stated scope

Rule: No claim may be stated at a higher evidence layer than its supporting data occupies. The current framework is at L0–L1 for all empirical claims. This must be stated explicitly in any work that cites the framework as providing empirical evidence.

Anti-pattern: Citing toy experiment designs as if they were results. Experiment designs are L0 until run; pilot results are L2; replicated results with reliability data are L3.

Article V — Extension Protocol¶

The failure taxonomy, repair taxonomy, and proxy metrics are open to extension. Extensions are legitimate under the following conditions:

Declaration: The extension is presented as an extension, not a replacement or correction of canonical constructs.
Non-contradiction: The extension does not redefine any canonical term (Article I applies).
Falsifiability: The extension includes an explicit falsification condition — a predicted observation that, if absent, would count as evidence against the extension.
Traceability: The extension traces to a lower-layer description (Article II applies).

Extensions that satisfy these conditions are encouraged. The framework is designed to be challenged, refined, and built upon. Null results from well-designed studies are as valuable as positive results — and more credible.

Article VI — Anti-Capture¶

MCW is a coordination construct, not a product, platform, or capability specification.

Rule: The following claims are overclaims and are not supported by the current framework:

"System X implements MCW" (systems are initialization artifacts or optimization targets; they do not implement MCW)
"Prompt Y solves the MCW problem" (system prompts are initialization artifacts; MCW requires ongoing behavioral practice)
"Model Z is MCW-aware" (models process tokens; MCW is an emergent property of interaction, not a model capability)
"Our RAG architecture reduces MCW degradation" (this is a capability claim dressed as a coordination claim; it requires controlled study under Article IV)

MCW does not belong to Anthropic, OpenAI, Google, or any other organization. It is not a product feature and should not be marketed as one.

Article VII — Compression Invariance¶

When MCW constructs are summarized, compressed, or adapted for presentation, the following must be preserved:

Falsification conditions — any summary of an empirical claim must retain the condition under which the claim would be falsified
Epistemic level — any summary must retain the evidence layer declaration (Article IV)
Exclusion list — any summary of a core construct (MCW, IU, etc.) must retain the "what it is not" boundary; omitting exclusions converts a precise construct into a vague metaphor

Rule: A summary that drops falsification conditions is not a valid compression — it is a distortion. Authors are responsible for ensuring that cited summaries of this framework maintain these invariants.

Article VIII — Scope Boundaries¶

MCW addresses the coordination layer of human–AI interaction — the shared state that emerges through exchange. It does not address:

Capability failures — model accuracy, reasoning quality, knowledge cutoffs
Alignment failures — value misspecification, reward hacking, goal misgeneralization
Safety failures — harmful outputs, jailbreaks, adversarial inputs
Interface failures — UI/UX design, accessibility, affordance design

Rule: Extensions of MCW into these adjacent layers require explicit theoretical justification. The burden of proof is on the extension author to demonstrate that the coordination-layer lens adds value beyond existing frameworks in those domains.

Amendment Procedure¶

This Constitution may be amended in future versions. Amendments must:

Increment the Constitution version number
State the amended article and the previous text
State the rationale for the amendment
Be made by the original author or by a declared successor with author consent, until the framework is formally published under a governance structure that supersedes this document

Amendments may not remove Articles I (Definition Immutability), IV (Epistemic Floor), or VI (Anti-Capture) without creating a successor framework under a new name.

Summary Table¶

Article	Invariant	Protects Against
I — Definition Immutability	Canonical terms fixed in glossary; departures must be declared	Definitional drift across citations
II — Layering Invariant	No construct at layer N without layer N−1 equivalent	Gatekeeping formalism; inaccessible abstractions
III — Substrate Independence	IUs and failure modes substrate-independent; MCW canonically scoped to HCW–ACW coupling, with generalizations as declared extensions	Premature substrate-specificity; false generalization; canon contradicting its own definition
IV — Epistemic Floor	Claims must declare evidence layer; no overclaiming	Toy experiments cited as validation
V — Extension Protocol	Extensions declared, non-contradictory, falsifiable, traceable	Scope creep; incompatible forks
VI — Anti-Capture	MCW is not a product, system, or model feature	Vendor appropriation; marketing overclaims
VII — Compression Invariance	Summaries must preserve falsification conditions and epistemic level	Distortion through compression
VIII — Scope Boundaries	MCW is the coordination layer; adjacent layers require justification	Framework overreach; capability/alignment conflation

Amendments¶

v1.1 (July 2026) — Article III¶

Amended article: III — Substrate Independence.

Previous text: "MCW, IUs, and all six failure modes are substrate-independent. They apply to human–human, human–AI, and AI–AI coordination systems without modification." (The Rule and Anti-pattern of Article III are unchanged.)

New text: the first two paragraphs of Article III as they now stand — IUs and the six failure modes remain substrate-independent; the MCW construct is canonically scoped to HCW–ACW coupling, with generalizations as declared Article V extensions.

Amendment Rationale: the previous text contradicted the canonical MCW definition (Article I; glossary), which is phrased for HCW–ACW coupling — by that definition, a human–human system contains no ACW and therefore no MCW, so the canon disagreed with itself in two quotations. Both resolutions were drafted with full paperwork in notes/amendment-drafts/article-iii-mcw-definition-options.md; this option (B — scope restriction) was adopted because it matches the actual evidence base (every current instrument is Human↔AI), changes no frozen glossary text, and protects rather than dilutes the framework's sharpest differential prediction, the HCW-continuous / ACW-discrete asymmetry (see Related Work). IUs and the failure modes remain substrate-independent, so the two Human↔Human toy experiments remain canonical failure-mode probes.

This document is version-controlled at docs/constitution.md. Cite as: W. Alec Akin, "MCW Constitution v1.1," in Meta-Context Window (MCW) Framework, GitHub, 2026.