▸ Zone 2 of 4 · The System

How the loop actually runs

The System zone covers the methodology layer: the Notice → Spec → Execute → Observe loop, the work ontology it produces, the spec-shaping protocol that turns intents into agent-ready specs, and the dogfood evidence from Intent building itself. It's less technical than The Build and more operational than The Hypothesis.

Intent The loop ▲ Notice Capture signals ◆ Spec Shape into contracts ● Execute Agents build ○ Observe Loop closure

The four-phase continuous loop. No sprint boundaries. Multi-speed (some loops run in minutes, some in weeks). Every transition emits an event.

▸ The four phases (with ancestors)

▲ Notice

What's the signal?

Capture friction, questions, observations, anomalies from anywhere work happens. Every signal is atomic and scored on trust dimensions.

Ancestors: Continuous Discovery (Torres), OODA Observe (Boyd), Get out of the building (Blank)

◆ Spec

What would we build?

Shape signals into agent-ready specifications with verifiable contracts. The four-persona interrogation protocol runs here.

Ancestors: Story Mapping (Patton), TDD (Beck), Shape Up (Singer), Outcomes (Seiden)

● Execute

What actually gets done?

Agents implement against contracts. Trust-gated at L0–L4 autonomy levels. Contract assertions pass or fail with evidence.

Ancestors: Continuous Delivery (Humble/Farley), DORA (Forsgren), GIST (Gilad)

○ Observe

What did we learn?

Runtime feedback turns into new signals. Single-loop: spec drift. Double-loop: questioning assumptions (still under construction).

Ancestors: PDCA (Deming), Build-Measure-Learn (Ries), Double-loop (Argyris), Three Ways (Kim)

▸ Open gap: double-loop learning

The Observe phase claims to implement Argyris's double-loop learning (questioning assumptions, not just optimizing execution). The current mechanism is single-loop only — we detect spec drift, we don't question whether the goals were right. The 328-voice registry is the latent mechanism for double-loop, framed as execution help instead of premise challenge. Work to convert this is tracked as SIG-050 / INT-013. This is an honest gap, not a capability.

▸ Deeper into the System

The Full Methodology

▸ Two-plane architecture

The work stream plane (signals → intents → atoms → events → traces) separated from the ownership topology plane (products → capabilities → teams). Bridging artifacts: specs and contracts.

v1.2 content · current · minimal framing update needed

Follow One Intent End-to-End

▸ Real SIG-010 trace

Traces a single real signal (SIG-010, our one external practitioner voice before the current discovery wave) through the full loop: signal → cluster → intent → spec → execute → observe. Real file paths, real IDs, real timestamps. The best "here's what it looks like" artifact on the site.

v1.2 content · honestly dogfooded · also the reason we flagged our own discovery gap (N=1 then; current ledger on The Proof)

The Work System Dashboard

▸ 6 operational tabs

Dashboard view of Intent's own work state — ontology, Agile→Intent translation table, signal capture pipeline, products portfolio, flow timeline. The operational home for practitioners working inside Intent.

v1.2 content · Products tab was updated 2026-04-09 with new product taxonomy · rest needs v2 framing

The Signal Stream

▸ Live signals

194 captured signals (census 2026-07-19) with trust scores, lifecycle states, clusters, and promotion history. This is Intent's own dogfood signal log — every friction we noticed while building Intent, with the full trust-scoring math applied.

v1.2 content · founding stream shown, live census 194 (2026-07-19) · external: 3 convergence signals logged, interview wave 1 still pending

The Voice Registry (352 entities)

▸ Voice catalog + filters

The full persona registry — 328 named-human thought leaders + 16 archetypes + 7 organizations + 1 peer operator persona (brien-operator). Filterable by discipline, role, industry. Each persona contributes to spec-shaping, panel-review, and archetype composition.

v1.2 content · updated 2026-04-09 with foundational tier promotions (Dunford, Gilad, Aakash Gupta)

15-Event Catalog

▸ Schemas + triggers

The 25 event types Intent emits (catalog revised 2026-06): signal.created, intent.proposed, spec.staged / .written, contract.started / .passed / .failed / .completed, the approval.* L0 gate, execution.paused / .resumed, human_input.*, observation.*, system.health, and more. The founding 15 are documented with schema, trigger, and example payload.

v1.2 content · missing the new deploy.* event family and safety-contract fields from INT-013

30-Minute Onramp

▸ 3-command install

Install Intent in 30 minutes. Three commands, real repo, real CLI. No API key for the first five steps. The conversion moment for skeptical visitors — where "is this vaporware?" becomes "oh, I can actually run this."

v1.2 content · praised by cold visitor in panel review · current and functional

The Observe Phase Narrative

▸ Loop closure mechanics

How observation closes the loop. Events become new signals. Metrics inform trust calibration. The mechanisms (current and aspirational) for turning runtime feedback into knowledge base updates.

v1.2 content · double-loop claim is aspirational · currently single-loop in practice

← Previous: The Hypothesis Next: The Build →