Skip to main content

Meta-Flow — Analytics & Feedback

Are the other six business flows learning from their own telemetry?

This flow checks whether raw events become decisions and whether decisions become better future decisions. If an event is never read by a decision, the loop is broken at the read end. If a decision is made without reading an event, the loop is broken at the decision end.

Performance connection: the meta-flow applies the Tight Five loop to the business itself: capture signals, prioritise the ones that matter, commit action, measure the outcome, and question what the next loop should improve. Its home phase is performance, because the loop only improves when decisions cite real instrument readings.

Signal​

The meta-flow tells you whether the business is closing the loop on itself. Raw events from every flow move through collection, integration, use, and decision telemetry. When it works, the body self-corrects. When it does not, every other flow degrades silently.

Data sources​

  • Collection — events, transactions, logs captured at source systems (every flow above is a source)
  • Integration — pipelines into a warehouse / lake, normalised schemas, joined entities
  • Use — dashboards, alerts, models, automations, agent-readable feeds
  • Decision telemetry — what decisions got made, who made them, what evidence they cited
  • Outcome telemetry — did the decision produce the predicted outcome? (per falsifiable-prediction pattern)

Decision gates​

  • Capture or drop event — schema valid, source authenticated, retention policy fits. Owner: data platform.
  • Promote signal to alert — threshold crossed, action available, owner named. Owner: observability + analytics.
  • Decision-vs-prediction reconciliation — at decision close, did the outcome match the prediction? Owner: any agent emitting a receipt.
  • Retire stale signal — alert nobody reads gets killed; gauge nobody decides on gets killed.

AI capabilities required​

All five capabilities operate here, because the meta-flow is what makes the other six self-correcting:

  • Perception — read every flow's telemetry
  • Decision — classify signal vs noise; promote, suppress, retire
  • Action — emit alert, brief decision-maker, surface lesson, kill a flow that fails its gauge
  • Memory — historical baselines; prediction track record; lessons compounding over time
  • Escalation — when a flow's kill signal fires, the meta-flow is the one that raises the alarm

Crypto-rail instruments​

  • On-chain receipts as immutable telemetry — every state-changing action emits a receipt; receipts indexed on chain become the queryable, tamper-evident history of the business.
  • Benchmark gauges settled on-chain — the gauge for each flow, settled periodically on chain via Nautilus TEE attestation; board + investors read the same numbers the operators read.
  • Decision markets — when a non-trivial decision is open, a futarchy market can price the predicted outcomes; the prediction becomes telemetry the meta-flow consumes.
  • Outcome-conditional payouts — pay out only when the predicted outcome was achieved; on-chain settlement removes the dispute layer.

Gauge​

% of decisions in the last 30 days that cited an instrument reading at the moment of decision. When this drops, decisions are being made on memory or vibes — the loop is broken at the read end. When it stays high, the meta-flow is doing its job.

Kill signal​

A decision is made in any flow without an instrument reading attached OR a flow's kill signal fires and no alert is raised to the named owner OR the consumed-rate of any standing dashboard drops to zero. Any one is a signal that the meta-flow is starving and the other six flows are about to drift.

Skill coverage​

Shipped: agt-save-receipt-emitter — emit structured receipts on completion. doc-audit-hermes-receipt-consumer — read unconsumed receipts, mark consumed/rejected. doc-audit-check-instrument-freshness — GREEN/YELLOW/RED freshness verdict per instrument. agt-run-falsifiable-prediction — convert any claim into a bet with check date.

Green-field: cross-flow telemetry composite that joins all six flow gauges into a single board view; on-chain benchmark settler that publishes the gauges via TEE attestation; alert-retirement agent that kills standing dashboards nobody reads; decision-quality auditor that scores decisions against the instrument-reading-cited gauge.

Full matrix: skills-matrix.

Upstream / downstream​

  • Upstream: every other flow (every flow emits telemetry into this one).
  • Downstream: every other flow (the meta-flow's outputs — alerts, lessons, kill signals — feed back into the other six). This is the loop that makes the system compound.

5P slice​

5P inspection: Principles (every decision earns its instrument reading; every gauge earns its decision), Performance (instrument-cited decision rate + flow-kill-signal MTTA), Platform (warehouse + observability + agent CLI + on-chain receipt indexer), Process (collect → integrate → use → reconcile prediction vs outcome → improve), Players (data + analytics + agent-ops + every flow owner + AI agent on perception, decision, action, memory, escalation).

Why it matters​

The first six flows answer "what is the business doing?" The seventh asks "is the business learning?" A business with all six commercial flows running and a starving meta-flow looks healthy while it degrades. A business with a strong meta-flow and weak commercial flows looks broken and fixes itself fast.

Bet on the meta-flow.

Failure modes​

This flow breaks when dashboards are treated as decoration, alerts keep firing without an owner, or receipts are emitted but never read. The misuse pattern is measuring activity instead of decision quality. A healthy meta-flow kills stale signals as aggressively as it promotes fresh ones.

Context​

Questions​

Which decision this week most needed a visible instrument reading?

  • Which dashboard or alert was ignored long enough to deserve retirement?
  • Which flow's kill signal fired without a named owner responding?
  • Which prediction should become the next calibration point?