Diagram Studio · Published

The VVFL Loop Gauge

A picture of the cycle that can prove whether the cycle is closing.

Checked-in projection of the loop inventory audited 2026-07-14.

§01

Wired, dormant, and open loops

The typed snapshot is the first provider boundary; a DB-native registry can replace it later without changing this surface.

Verified loop inventory

Is the value–verification–feedback loop closing?

21 executable loops projected from docs/_AGENTIC/loops/01-reality.md, audited 2026-07-14. This checked-in reading can later be replaced by a DB-backed provider without changing the diagram.

WIRED
11
DORMANT
5
OPEN
5
Values, beliefs, and control systems rendered as a feedback loop: experience phenomena, form intent, decide how to act, hone capability, observe outcomes, test expectation, adjust or amplify what works, then feed better questions back into the ideas hopper. Prioritise: Filter for attention: choose the important and actionable before attention scatters. The filter is carried by the negative space between the visible instruments.VVFL thought loop stationsCapture: Experience phenomena: notice the stories, worries, signals, collisions, and connections entering the idea hopper.Commit-Action: Turn intent into motion: decide how to act and focus capability on the outcome you mean to produce.Question-Learn: Close the loop: ask why reality differed, adjust or amplify what worked, and feed sharper questions back into the hopper.Measure: Observe outcomes: compare what happened against expectation and what matters.Learning returns to sharpen the next capture
  • WIRED

    Runs with a real consumer

  • DORMANT

    Built or proven, but unconsumed

  • OPEN

    The cycle exists, but does not close

Value loops

Touch a buyer or money · 9 loops

  1. RFP closed-loop

    WIRED

    venture → RFP pursuit → Q&A → proposal

    Runs with real consumers; no scheduler.

    Evidence
    • libs/agency/src/lib/loops/sales-rfp-workflow/
    • run-closed-loop.use-case.ts
    • agent-api POST /rfp/closed-loop/run

  2. RFP → contract → settlement

    OPEN

    accept quote → award/sign → settle → spawn fulfilment

    Wired to Stripe intent, but open at capture.

    Evidence
    • libs/application/src/lib/use-cases/sales__contract/
    • StripePaymentAdapter
    • apps/dreamineering/drmg-sales

  3. Settlement receipt loop

    OPEN

    payment webhook → value_transaction ledger

    Webhook omits organisationId and writes nothing.

    Evidence
    • apps/dreamineering/drmg-sales/src/app/api/webhooks/stripe/route.ts

  4. Outcome pricing

    DORMANT

    anchor × demand × capability → PricedOutcome

    Built with zero consumers; quotes are hand-typed.

    Evidence
    • libs/agency/src/lib/algorithms/pricing/
    • price-outcome.use-case.ts

  5. Receipt calibration

    OPEN

    settlement receipts → re-rank/re-price

    Both ends are built but wired to nothing.

    Evidence
    • libs/application/src/lib/use-cases/valued-outcomes/calibrate-from-receipt.use-case.ts

  6. Valued-outcomes triage

    WIRED

    demand-value-score → value-triage → ranked next work

    Seed is hand-authored; receipts do not feed back.

    Evidence
    • drmg plan outcomes
    • triage-valued-outcomes.use-case.ts

  7. Sales-dev outreach

    WIRED

    research → outreach draft

    Runs through agent-api; no recurring trigger.

    Evidence
    • libs/agency/src/lib/loops/sales-dev-outreach-workflow/
    • agent-api sales-dev module

  8. Marketing strategy

    WIRED

    brief → strategy

    Runs with a real consumer.

    Evidence
    • generate-marketing-strategy.use-case.ts

  9. Lead-magnet / vibe-marketing / SEO-content / marketing-loyalty / AI-transformation

    DORMANT

    content/distribution cycles

    Agency-internal only.

    Evidence
    • libs/agency/src/lib/loops/*

Delivery loops

Turn a plan into shipped work · 5 loops

  1. drmg planning loop

    WIRED

    create → advance → log-trace/decision → session → receipt

    DB-native with a complete verb set.

    Evidence
    • libs/drmg-cli/src/lib/planning/
    • planning_* tables

  2. runAgentLoop control loop

    DORMANT

    budgeted sense → act → verify with pause/resume

    No consumers outside libs/agency.

    Evidence
    • libs/agency/src/lib/pipeline/control-loop/loop-runner.ts

  3. Task dispatch

    DORMANT

    typed dispatch payload + parent verification contract

    Contract exists; runtime dispatcher does not.

    Evidence
    • get-task-dispatch.use-case.ts
    • .claude/rules/dispatch-contract.md

  4. Ralph goal loop

    WIRED

    Stop hook re-injects prompt until promise met

    The one self-perpetuating loop.

    Evidence
    • .claude/hooks/stop-ralph-loop.sh
    • .claude/ralph-loop.local.md

  5. Project-plan / chat-stream / coach-conversation

    WIRED

    generation workflows

    Runs with real consumers.

    Evidence
    • libs/agency/src/lib/loops/*
    • compositions

Improvement loops

Raise the platform floor · 7 loops

  1. SilentWitness

    OPEN

    friction events → daily digest → curator triage

    Capture and processing are wired; promotion to controls is manual.

    Evidence
    • .claude/hooks/silent-witness.sh
    • process-silent-witness.ts

  2. Compound step

    WIRED

    Stop gate → forced log-trace/log-issue → WSJF controls backlog

    Runs as a Stop gate.

    Evidence
    • compound-step-gate.cjs
    • .claude/rules/flow-routing.md

  3. Health reports

    WIRED

    CI cron → scanners → auto-PR

    Runs every three days.

    Evidence
    • .github/workflows/health-reports.yml
    • docs/code-reviews/

  4. Scheduled proof lanes

    WIRED

    daily integration/e2e/security/codeql scans

    Runs on a daily schedule.

    Evidence
    • .github/workflows/*.yml

  5. Eval loops

    DORMANT

    corpus + LLM judge + L4 workflow and skill evals

    No schedule and gates nothing.

    Evidence
    • libs/agency/src/lib/algorithms/eval-runner/
    • *.workflow.eval.ts

  6. Agent memory / evolution loop

    OPEN

    decision → outcome → lesson → memory

    Memory writes fail; lesson promotion is manual.

    Evidence
    • docs/agency/evolution-loop.md
    • agent-etl.handlers.ts

  7. @claude CI loop

    WIRED

    @mention on PR/issue → bounded agent run

    Runs when a human triggers it.

    Evidence
    • .github/workflows/claude.yml