№ 03 · CHAPTER THREE · THE PATH
From feature to substrate. Governed all the way.
A tool advantage can be neutralised by a competitor buying the same tool. A substrate advantage compounds with every loop cycle run.
Governance is not a feature. It is the substrate. Every action carries its policy gate, its evidence trail, and its escalation path before it ever runs.
CANON · ARCHITECTURE CONTRACT
Safety is not an overlay.
- Source sovereignty. Each data class is explicitly copied, normalised, referenced, federated, retained, exported, or deleted under policy.
- Tiered autonomy. High-consequence decisions require human review by policy. The four operational states are bounded, not assumed.
- Transparency by architecture. Every action logs its full reasoning chain. What was perceived, inferred, decided, and on what evidence.
- The Red Button. The canonical requirement is a global execution pause within one processing cycle. A millisecond SLO requires production evidence.
- Compliance as code. Your values and the regulations you operate under are versionable, testable, and enforced before any action is taken.
- Auditable end-to-end. The target runtime records every material transition with actor, input, policy, model, time, hash, and outcome lineage.
TARGET · CUSTOMER-SPECIFIC PROGRESSION
Progression by evidence. Never by calendar alone.
Every tenant begins in Observe. Movement to Advise, Co-Execute, or Absorb is a target progression governed by measured outcomes and can regress when quality or calibration deteriorates.
- Trust Contract. Action classes, policies, evidence measures, owners, downgrade rules, and the Red Button are agreed.
- Observation. Read-only sources build STATE and confidence. No recommendations or execution.
- Advising. Recommendations are surfaced with evidence. Humans remain sole executors.
- Co-Execution. Drafts or staged actions require explicit approval at execution time.
- Absorption. Only bounded action classes with sustained evidence may execute within policy.
DEMO · FIRST-SCENARIO SPECIMEN
Start where two systems disagree about one customer.
The bounded first scenario joins Sales and Finance around one question: should a €240k expansion progress while €186.4k remains overdue under a different customer name?
- 1
- TENANT
- 2
- SOURCE CLASSES
- 1
- CANONICAL CUSTOMER
- 1
- GOVERNED DECISION
The website version uses fictional records. A customer Observe engagement replaces them with approved read-only source data and measured identity coverage.
TARGET · FIRST-SCENARIO BUYING MODEL
Install one scenario. Prove one loop.
The first purchase is a bounded Observe scenario, not the complete platform. It creates the evidence required to decide whether progression is justified.
- Scope. Name one tenant, domain, question, source set, decision class, and accountable owner.
- Connect. Provision one or two read-only sources and approve data-class handling.
- Resolve. Build the Identity Graph slice and measure entity coverage with a declared denominator.
- Observe. Surface one evidence-bearing belief or anomaly without creating a write path.
- Review. Verify findings, data lineage, policy scope, and commercial value with the customer team.
- Decide. Stop, extend Observe, or approve the next trust state based on measured evidence.