1 / Thesis

Intelligence is the layer between data and action.

Businesses generate signals constantly: transactions, messages, patient events, CRM changes, website activity, inventory updates, documents and human decisions. The hard part is turning those signals into the next useful action. This document sets out how OTLEN does it.

Stages in the loop
05
Fields on every event
06
Revision
2026.10

2 / Loop

Five layers between a signal and an outcome

Data enters at the bottom. The two layers in the middle are the intelligence. Action leaves through the fourth layer, and every result feeds back into the first.

Fig. 1 · Decision loop, exploded viewScale NTS

Parts list

  1. 01ObserveKnow what changed: source, time, entity, previous state, new state and a unique ID.
  2. 02UnderstandRules take the deterministic calls. Models take the messy, unstructured input.
  3. 03DecideExplicit enough to audit: the trigger, the data it needs, and the plan for low confidence.
  4. 04ActOutput lands somewhere real: a CRM update, a task, an approval, an API call or the next workflow.
  5. 05LearnEvery action creates feedback. Overrides and real exceptions tune the rules, prompts and routing.

Parts 02 and 03 form the intelligence layer. Remove either one and the loop still runs, but nothing in it decides.

3 / Stance

Who acts at each step

Deterministic work runs on its own. Judgment is supervised. Uncertainty, and any change to how the system behaves, goes to a person. The chart shows OTLEN's default design stance. It is not measured data.

Step 06 · Decide

Handle low confidence

PersonAssistedSupervisedSystem

Who acts: Person

Low confidence is a signal. A person decides, with the context attached.

StepStageActionWho actsWhy
01ObserveCapture the eventSystemNothing about recording an event needs a person.
02ObserveRecord full contextSystemSource, time, entity, previous state, new state and ID, every time.
03UnderstandApply business rulesSystemWritten rules give the same answer every time. Software enforces them best.
04UnderstandInterpret messy inputSupervisedModels read unstructured information well. Their reading is still checked.
05DecideChoose the actionSupervisedThe decision is explicit and logged. Its confidence decides whether a person reviews it.
06DecideHandle low confidencePersonLow confidence is a signal. A person decides, with the context attached.
07ActWrite back to the sourceSystemResults land in the system of record, so the source stays true.
08ActTrigger the next workflowSystemThe next step starts without waiting for someone to remember it.
09LearnRecord overridesSystemEvery override is captured as feedback, with who made it and why.
10LearnChange rules and promptsAssistedThe data suggests changes. People approve any change to how the system behaves.

4 / Specification

Operating specification

What every OTLEN system records, decides and tolerates.

Event record 6 fields

  • SourceThe system that emitted it
  • TimeWhen it happened
  • EntityWhat it is about
  • Previous stateBefore
  • New stateAfter
  • Event IDUnique

Decision record 5 rules

  • TriggerExplicit conditions
  • Required dataListed in advance
  • ConfidenceScored per decision
  • Low confidenceA person, with context
  • OverrideLogged as feedback

Where actions go 6 kinds

  • Update a recordCRM, EHR, ledger
  • NotifyThe owner
  • Create a taskWith context attached
  • Request approvalWhen policy requires it
  • Call an APIIdempotent, with retries
  • Start a workflowThe next step
Silent failures tolerated0
Decisions logged100%
Overrides recorded100%

5 / Principles

Seven rules OTLEN builds by

The stacked glyph on each rule marks the layer of the loop it governs.

01 / Observe

Know what changed

Capture every event with enough context to reconstruct it later.

02 / Understand

Rules first, models for the mess

Deterministic decisions belong to rules. Models earn their place on unstructured input.

03 / Decide

Make uncertainty visible

Low confidence is a signal. Route it to a person instead of guessing, and show them why.

03 / Decide

Audit every decision

Every trigger, input and outcome can be explained after the fact.

04 / Act

Change something real

Value appears when a real system changes. An answer on a screen changes nothing.

05 / Learn

Treat overrides as data

When a person overrules the system, record it. The rules improve next time.

01–05 / Whole loop

Start with one workflow

Instrument it. Find the exceptions. Fix the architecture. Then expand.

The test

Data becomes intelligence when it changes what the system does next.