- 28 October 2026Working note
The Value of What the Model Knows: A Working Note on Governed Knowledge as a Variable in the Economics of AI Tasks
Reichhart and Gelas · v0.3
What an agent is permitted to rely on enters the economics of an AI task as a variable of its own, priced beside the token.
ZenodoDOI 10.5281/zenodo.23726953open, CC BY 4.0
- 28 October 2026Paper outline
The Value of What the Model Knows: Governed Knowledge as a Variable in the Economics of AI Tasks
Reichhart and Gelas · v0.3
The paper the working note above prepares.
ZenodoDOI 10.5281/zenodo.23726794metadata public, files restricted
- 28 October 2026Paper outline
The Mandate as a Unit of Risk
Reichhart and Gelas · v0.3
The paper the mandate note below prepares: the mandate as the unit at which a firm sets, measures, and narrows its appetite for what an agent may do.
ZenodoDOI 10.5281/zenodo.23726616metadata public, files restricted
- 28 October 2026Working note
A Risk Appetite Where the Act Happens: A Working Note on the Mandate under Which an AI Agent Acts
Reichhart · v0.1
Firms write delegated authorities for their staff and run their agents on entitlements. The mandate is the join between the two: a versioned authority over one class of acts, which says what the agent may do, what evidence it needs, when the control plane may refuse, and who answers when it does.
ZenodoDOI 10.5281/zenodo.23726164open, CC BY 4.0
- 17 September 2026Paper
Ontology Governance in Claim-Level Architectures: The Bootstrap Problem, Versioning, and Cross-Domain Transfer
Gelas and Reichhart
The type system beneath the claim graph: how it bootstraps, how its versions are bound into manifest replay, and how it transfers across domains, governed by the same mechanism as the claims it types.
ZenodoDOI 10.5281/zenodo.22813139also on SSRN from 21 September 2026SSRN 10.2139/ssrn.7477438
- 21 June 2026Paper
The Knowledge Layer: A Reference Architecture for Delegated AI Action in Regulated Institutions
Reichhart and Gelas
The claim as the unit, its standing computed on two axes of evidential warrant and authority, a consequence-scaled gate before the act, and a replayable manifest after it: the record an institution can put in front of a supervisor.
ZenodoDOI 10.5281/zenodo.20784662v2, CC BY 4.0The architecture page
- 3 May 2026Paper
From Autonomy to Initiative: Enterprise AI's Real Endgame
Reichhart and Gelas
Three conditions under which an agent stops waiting for instruction and starts perceiving what matters, and where governance relocates when it does.
SSRNDOI 10.2139/ssrn.6702239mirrored on Zenodo
- 3 May 2026Paper
Governed Intelligence Architecture for Institutional AI
Gelas and Reichhart
The governed intelligence lifecycle, from ingest to apply, with the immune function that defends the knowledge against pollution, staleness, fragmentation, amnesia, cascade, and distortion.
SSRNDOI 10.2139/ssrn.6701941mirrored on Zenodo
- February 2026Paper
Build the Medium: Why Organizational Intelligence Is Mechanism, Not Metaphor
Reichhart and Gelas
Ten traditions, from cell biology to social systems theory, converge on what an organization needs in order to be intelligent. Capability is the smaller half of it.
SSRNDOI 10.2139/ssrn.6230858mirrored on Zenodo
- February 2026Paper
The Predictive Organization: Architecture for Enterprise Intelligence
Gelas and Reichhart
Map, Physics, Player: a tripartite architecture that couples neural perception with symbolic reasoning over claims-based knowledge.
SSRNDOI 10.2139/ssrn.6230780mirrored on Zenodo
- February 2026Paper
Dynamics Blindness: When AI Is Locally Correct and Globally Non-Compliant
Reichhart and Gelas
A model answers correctly at the point it is asked and breaks a rule three systems away. The gap is architectural, and no amount of prompting closes it.
SSRNDOI 10.2139/ssrn.6230758mirrored on ZenodoThe explainer
The spine
Read in the order they were written, the papers make one argument. Enterprise AI fails because of dynamics blindness: a model is correct where it is asked and non-compliant three systems away (Dynamics Blindness). The resolution is architectural, a tripartite structure over claims-based knowledge (The Predictive Organization). The architecture is the right one because ten traditions, from cell biology to social systems theory, converge on what a living system needs in order to be intelligent (Build the Medium). The practitioner's version is a governed lifecycle with an immune function (Governed Intelligence Architecture for Institutional AI). At sufficient depth the architecture produces initiative, agents that perceive what matters through immersion rather than instruction (From Autonomy to Initiative). The Knowledge Layer is the reference architecture that makes a delegated act admissible at the moment it commits (The Knowledge Layer). Ontology governance keeps the type system beneath the claim graph governable, versionable, and transferable (Ontology Governance in Claim-Level Architectures).
The working notes and outlines of October take the step above the architecture: the mandate as the unit at which a firm sets its appetite for what an agent may do, and the value of what the model is permitted to know. The control plane page says where they belong.