ResearchSystems TheoryWorkflows
Structured Reasoning Procedures · Six-Step Pattern · Repeatable Inquiry

Lucid Workflows

Workflows are the procedural layer of Lucid reasoning — structured sequences that make the steps of an inquiry explicit, repeatable, and inspectable. They transform the DCR cycle from a cognitive architecture into a running procedure.

The canonical Lucid workflow is a six-step sequence organised around three reasoning phases: divergent (Observation, Hypothesis), navigational (Evidence, Interpretation), and convergent (Synthesis, Reflection). Every step preserves stance differentiation — the distinct epistemic positions that make multi-agent reasoning more than redundant parallelisation.

Position Within the Research Stack
FoundationsPhilosophical ground
TheoryCognitive architecture
Media GrammarStructural translation
InteractionInterface layer
Systems TheoryComputational infrastructure
Workflows as Epistemic Procedure

AI workflows in current practice are predominantly task pipelines — structured sequences for routing data through processing steps. The structure optimises efficiency: steps are minimised, parallelised where possible, and evaluated by throughput and output quality. The reasoning process that produces the output is incidental to the pipeline's design.

A Lucid workflow is designed for epistemic quality, not throughput. The steps exist to protect the integrity of the reasoning process — not to optimise the production of output.

This means that Lucid workflows have properties that task pipelines do not: they are stance-differentiated throughout, meaning that the distinct epistemic positions of the agents involved are maintained as separate tracks — not merged until synthesis. They are inspectable, meaning that the reasoning at each step is visible, not hidden inside opaque processing. And they include a reflection step, which is the procedural form of metacognition — a structural commitment to understanding how the reasoning proceeded, not just what it produced.

The result is a workflow that produces not just outputs but understanding — understanding that can be stored in Memory, surfaced through Interaction, and built on in subsequent reasoning episodes.

The Canonical Six-Step Workflow

Six steps across three phases. Steps 1–2 open the epistemic field (divergent). Steps 3–4 move through it (navigational). Steps 5–6 integrate and review (convergent).

01
Observation
Divergent

What is present in the epistemic field, recorded without interpretation. Observation is not passive recording — it is active, stance-aware noticing. Different stances will foreground different observations from the same material. The observation step makes this explicit: what each stance notices is tracked, not collapsed into a unified set of "what is there." The product of observation is a structured set of stance-differentiated noticings.

02
Hypothesis
Divergent

Candidate explanations, framings, and interpretive propositions generated from the observations. Hypotheses are not guesses — they are provisional structural claims about what the observed material means or how it is organised. In a multi-agent workflow, each stance generates hypotheses that reflect its interpretive orientation. The hypothesis step is the first place where divergent coverage matters: the richer the hypothesis space, the stronger the subsequent inquiry.

03
Evidence
Navigational

Structured gathering and evaluation of material that bears on the hypotheses in play. Evidence is gathered in relation to specific hypotheses — not as a general accumulation but as a directed inquiry. Each piece of evidence is evaluated under each stance: what this evidence means, whether it supports or challenges a hypothesis, depends on the interpretive position from which it is assessed. Evidence provenance — which stance evaluated it, how — is preserved.

04
Interpretation
Navigational

Meaning-making from the gathered evidence under each stance. Interpretation is explicitly stance-dependent — the same evidence base will yield different interpretations under different stances, and this is a feature of the workflow, not a defect. The interpretation step produces a structured set of stance-differentiated meaning-claims, each with its evidential basis explicit. The product is not a single interpretation but a mapped interpretation space.

05
Synthesis
Convergent

Integration across the interpretation space — the convergent move that draws together the stance-differentiated interpretations into a structured understanding. Synthesis is not averaging or majority-voting. It is the identification of what holds across stances, what the tensions between stances illuminate, and what the most structurally grounded position is given the full evidence base. Synthesis may leave some tensions unresolved — these become the basis for the next reasoning episode.

06
Reflection
Convergent

Metacognitive review of the reasoning process itself — not of the content produced, but of how the reasoning proceeded. Reflection asks: were the stances adequate to the epistemic field? Were there gaps in observation, hypothesis generation, or evidence gathering? Was synthesis premature or forced? Were avoidance patterns active? The reflection step is the feedback loop that makes workflows improve over time — its output informs both memory and the calibration of future orchestration.

Structural Properties

Four properties that distinguish Lucid workflows from task pipelines and make them genuine epistemic infrastructure.

01Repeatability

The same workflow can be applied to different epistemic fields and different content domains. The structure — six phases, stance differentiation, provenance preservation — is invariant. This repeatability is what makes workflows a genuine infrastructure rather than a one-off procedure.

02Inspectability

Every step in a workflow is visible — its inputs, the stances that processed them, the outputs. There is no black box in a Lucid workflow. This inspectability is not a logging feature; it is a structural commitment to making reasoning transparent to the humans who participate in the environment.

03Composability

Workflows can be composed — a synthesis from one workflow can become an observation in the next. The convergent output of a reasoning episode is available as a starting point for further divergent exploration. This composability is what enables the accumulation of understanding over time rather than episodic resets.

04Stance fidelity

Throughout a workflow, the integrity of each stance is preserved. Evidence, interpretation, and synthesis are not collapsed into a single perspective — the distinct epistemic positions remain traceable from observation to synthesis. This fidelity is what makes the synthesis meaningful: it integrates genuinely different positions, not a monoculture with noise.

Structural Connections
The six-step workflow is structured by DCR — steps 1–2 are divergent, steps 3–4 are navigational, steps 5–6 are convergent. The workflow is DCR made procedural. Divergent-Convergent Reasoning
Orchestration executes workflows — managing the agent configurations, phase transitions, and evidence routing that a workflow requires to run. Lucid Orchestration
Workflow outputs — interpretations, syntheses, reasoning paths — are the primary content that Memory stores and makes available to future workflows. Lucid Memory
Systems Theory
Orchestration
Workflows
Memory
← Systems Theory