Framework

Framework

🇷🇺 Читать на русском


Ceiling-Residual Framework

An integrated analytical methodology connecting the Institutional Ceiling diagnosis with the Governance Residual operational response.


Section 1: What the Framework Establishes

First: AI governance architectures reach a structural ceiling — the product of three limits interacting multiplicatively. When all three operate simultaneously, corrective capacity collapses below the threshold at which meaningful enforcement remains viable.

Second: After the correction window has closed, governance residual remains. Partial and asymmetric, it operates through insurance, litigation, infrastructure adjacency, and dependency management.

Third: The depth of agency transfer constrains the extent of the residual. Institutions that have already transferred significant decision-making authority to AI systems have less residual available.

A fourth question remains open: what would institutions designed from these constraints look like from the outset? This constitutes the next research programme — AI-Native Institutions. Its development is documented as inquiry proceeds, not as conclusions arrive.


Section 2: Core Concepts

Definitions below are aligned word-for-word with the Analytical Glossary v1.0 (DOI: 10.5281/zenodo.20741338), the canonical terminology standard for this platform. Glossary entry numbers in parentheses.

Institutional Ceiling (97) — The limit above which an existing governance architecture is incapable of functioning at a given level of technological complexity — regardless of intentions or resources.

Governance Residual (107) — The partial, asymmetric control mechanisms that retain operational leverage after formal governance capacity is lost — including insurance underwriting, litigation, and procurement conditions. Not a governance alternative but the empirical condition peripheral jurisdictions must work within once the correction window has closed.

Correction Window (11) — The last moment of reversibility; the brief phase during which institutions retain the capacity to alter an AI system’s architecture before operational dependencies render it effectively ungovernable.

Sovereign Override (15) — The situation in which state authority cancels or ignores previously established constraints on an AI system in pursuit of a priority — including strategic or security — objective.

Material Predetermination (105) — The constraint on sovereign governance choices imposed by the physical configuration of the AI technology stack — chip fabrication, energy infrastructure, data centres — before any policy deliberation begins. One of the Three Structural Limits. Physical assets precede institutional possibilities.

Institutional Mismatch (106) — The categorical incompatibility between existing governance architectures — built for slower, territorially bounded, physically verifiable systems — and the speed, cross-jurisdictional operation, and opacity of advanced AI. One of the Three Structural Limits. Governance institutions govern yesterday’s world with yesterday’s instruments.

Agency Transfer (9) — The gradual migration of decision-making authority from human actors to algorithmic systems, driven by automation, growing dependency, and institutional incentive structures.

Human Reversal Capacity (24) — The possession by operators of retained skills and expertise sufficient for manual management of critical processes in the event of an AI system’s failure.

Governance Theater (5) — The institutional-scale performance of governance activity by state or corporate actors without real capacity to alter system behaviour. Distinct from Performative Control (#6), which operates at the level of individual actions rather than strategic posture.

Point of No Return (114 · addendum, pending Zenodo v1.1) — the threshold of Agency Transfer Depth (95) beyond which the cost of restoring Human Reversal Capacity (24) exceeds what an organisation can practically absorb. Distinct from Correction Window (11): the Correction Window is the interval during which reversal is still possible; the Point of No Return is the threshold at which that interval ends.

Full glossary of 113 canonical terms (114 including this addendum) in Section 4.


Section 3: Institutional Self-Assessment — Ten Operational Questions

Ten questions to assess an institution’s position relative to its own correction window. This is the operational self-assessment layer of the Framework — the underlying methodology of INVEXI’s Agency Transfer Audit (ATA). The conceptual screening filter for identifying Governance Theater in any system (not limited to self-assessment) is the separate Diagnostic Protocol in the Analytical Glossary, Section: Part Two.

  1. What percentage of decisions with material operational consequences are currently made or substantially shaped by AI systems without independent human verification?

  2. How long would it take to restore fully manual operation of core functions if AI systems became unavailable? Has this been tested?

  3. What proportion of staff who supervise AI-mediated processes could execute those processes independently if the AI system were removed?

  4. What contractual rights does the institution have to audit, modify, or halt the AI systems it deploys?

  5. Is the institution currently able to switch to an alternative AI system provider without material disruption?

  6. How many critical operational processes depend on data infrastructure controlled by a single external provider?

  7. What formal mechanisms exist to report AI system failures to regulators? How many reports have been submitted in the past 12 months?

  8. Has the institution assessed the gap between declared AI system behaviour and observed operational behaviour?

  9. What governance authority does the board exercise over AI deployment decisions?

  10. In the event of an AI system failure causing material harm, through what mechanisms would the institution establish its liability position?


Section 4: Full Glossary

113 canonical terms in English, Russian, and Uzbek (Latin script). The Khodjaev Framework, Version 1.0 · June 2026.

Read the full Glossary → · Download PDF · DOI: 10.5281/zenodo.20741338

Substack X.com LinkedIn