Framework · Fix Governance

01 · Premise

The premise

Enterprise governance for thirty years has been a discipline of declaration. A system pronounces a fact (this is the master record, this is the permitted action, this is the classification) and the rest of the stack is asked to comply.

This worked because the actors downstream were human. People noticed when the catalog was stale and overrode it. People escalated the ambiguous case. The governance system was the starting point for human judgment, not a replacement for it.

Agentic execution removes the human safety net. A declaration is now the decision, made at the latency of an API call, with no afternoon to think about it. The premise of declarative governance has expired. What replaces it is a control plane that adjudicates across sources rather than picking one to trust.

02 · Architecture

The control plane

FIQ resolves two questions on every agentic transaction. The first is about the entity. The second is about the action. Both must be answered, in-flight, with a confidence score, before execution proceeds.

Engine 01

Entity Resolution

"Is this the right entity?"

Federated identity and context resolution across source systems. No persisted master record. No nightly reconciliation. Multi-source adjudication at query time, producing a resolved entity plus a confidence score derived from inter-source agreement.

Zero-copy
Always-live
Confidence scoring
Source systems remain authoritative

Engine 02

Token Governance

"Should this action execute?"

Pre-execution decision layer for agentic actions. Combines policy, identity, budget, history, and lineage into a single emission: allow, allow-with-conditions, route-to-human, or deny. Logged per transaction as auditable evidence.

Monitor → Report → Block
Threshold → Limits → Chargeback
Pre-execution, not post-hoc audit

Both engines compose into a single answer to a single question:

"Should this agentic transaction proceed?"

03 · What it is not

Anti-positioning

Category-defining work is recognized as much by what it refuses to be as by what it claims. FIQ is not in any of the categories below, and the distinctions are load-bearing.

Not MDM. There is no canonical master record. The premise of mastery is rejected.
Not ETL or iPaaS. No data movement. No pipelines. Source systems are queried at decision time, not copied.
Not a data warehouse. No persisted copy of source records. Storage is operational, not analytical.
Not a system of record. Sources remain authoritative. FIQ is the adjudicator across them.
Not a vendor MCP connector. Vendor MCPs are tools FIQ governs across, not products FIQ replaces.
Not a catalog. Catalogs answer "what is this?" FIQ answers "should this happen?"

04 · Strategic frame

Open thesis, closed implementation

The strategy that produces a category and a defensible product at once is the one Stripe, Tailscale, and dbt Labs used. The thesis is published in public. The reference architecture is open. The implementation is not.

This site is the thesis layer. The implementation is Tomosu. The intentional separation is the same one that gave Stripe a decade of category-shaping essays before any competitor understood what was actually being built.

For the deeper thesis on federation as a primitive, distinct from agentic governance, applicable to data and identity at the architectural level, the companion publication is federatediq.ai.

05 · Reading order

Where to start

The framework above is the spine. The essays explore each part of it in operational detail.

For the principle behind Engine 01's confidence scoring, start with N° 001 · Disagreement Is the Signal. For the distinction between catalogs and control planes, read N° 002 (forthcoming). For the historical frame on why MDM failed and why federation succeeds where it failed, read N° 004 (forthcoming).