← Back to the method library
Futures Research Ecosystem · interactive overview
Plan → Collect → Analyse & Report

One ecosystem · three phases

From research question to foresight

A modular ecosystem of methods and AI agents that carries a Delphi study from the first framing decision to validated scenarios and stakeholder reports — with humans in the loop at every interpretive turn.

Phase 1 · Planning

Delphi Planning Ecosystem (DPE)

Design the study as research: orientation, phenomenon, panel and question architecture.

Output: Planning Handoff Package
→
Phase 2 · Collection

Pronoia — facilitation

Argue, don't just vote: iterative, anonymous expert panels. Numbers and reasons.

Output: stances + arguments + metadata
→
Phase 3 · Analysis

Delphi Analysis Ecosystem (DAE)

A 21-checkpoint pipeline interprets the material under quality gates.

Output: findings, scenarios, reports
Learning loop: every study feeds a shared Case Library — lessons return to the next plan (planning M6 & analysis CP-19/20).
3
Main phases
21
Analysis checkpoints (CP-0…CP-20)
12+
Method agents (Max* family)
10+
Quality gates & tests
27
Report templates
How to use this page: navigate with the tabs or the ◀ ▶ arrow keys. Inside each phase, click the pills to reveal details. Everything runs offline in the browser — safe to project. The companion user guide (fi/en/sv) is the most up-to-date reference.

Phase 1 · planning core (M1a) + satellites

Planning — research design as research

The planning core walks from orientation to a quality gate in staged steps, each ending in a STOP point for the researcher. The result is a Planning Handoff Package the analysis pipeline (DAE CP-0) ingests directly.

Six research orientations

N · Normative

“Should it happen?” — desirable futures, value-laden theses tested dialectically. Leans to consensus.

S · Strategic

“How do we make it happen?” — means, actions and pathways toward a chosen goal.

R · Radical

Provocative framings that break the regime's assumptions and protect deviance — no forced convergence.

E · Emergent

“What is already emerging?” — abductive theses built from weak signals rather than fixed claims.

D · Combination

A blended design mixing dialectical orientations within one study. Default for multifaceted studies.

D+ · Four-path

Extended combination running intentional and emergent framing paths in parallel.

The core sequence — orientation → phenomenon → panel → questions

After the orientation is locked, three staged steps build the study and end at a quality gate. Click each:

B3 Phenomenon — what is actually being studied
The phenomenon is the boundary of the inquiry — planning's most critical scoping step, deciding what is in and what is out. It is sharpened by the optional pre-analysis (Futures Wheel = consequences, Relevance Tree = goal–means) and frames everything downstream: a contested, multi-actor phenomenon calls for a Delphi, not a forecast.
B4 Panel — a designed epistemic space
The panel is not a random set of experts but a designed epistemic space whose coverage and blind spots are known before any data exists (arguments = theses × panel members). The orientation sets the type:
N/S/R/D → expert matrix (≥ 3×3)E → observer matrix (≥ 2×3)D+ → two parallel panels
The optional perspective layer (position + view) yields the Perspective Health Score (PHS); argument coverage is read as a 7×3 = 21-cell map (the Epistemic Coverage Index). Orientation-specific quality gates must pass before collection.
B5 Question architecture — the theses
The theses are the collection instrument. Their structure follows the orientation — dialectical (N/S/R/D: a claim taken a position on by probability & desirability) or abductive (E: open, exploratory). A per-thesis P/D tag can override the default (probability → consensus, desirability → dissensus). At least three theses; the instrument is locked when Round 1 opens so every round's data stays comparable.

Satellite modules around the core

M1b Pre-analysis — Futures Wheel & Relevance Tree
Offered at a gate before the phenomenon is fixed:
  • Futures Wheel — 1st/2nd/3rd-order consequences, STEEP+ classified impact network.
  • Relevance Tree — goal–means hierarchy with priority weights and success criteria.
M1c Simulation — the Argument Simulator (Route B)
Generates orientation-aware synthetic arguments for the theses — to stress-test questions before a real panel, or to run a fully simulated Delphi. Output is DAE-ready, always with explicit GenAI disclosure.
M5 Perspective — MaxPerspective (Hautamäki)
Perspective-realist health check of the planned panel: projected aspect map, aspect diversity (AD%), perspective diversity index (PDI), Perspective Health Score (PHS, 0–12) and supplementation advice — before fieldwork starts.
M6 Learning — the planning learning loop
Every planning process is documented as a Case Card with Learning Notes; retrospectives after execution and a pre-survey of past lessons when a new study begins.

Phase 2 · panels & rounds

Data Collection — argued expert judgment

Delphi collects two kinds of evidence at once: positions (probability & desirability) and arguments (the reasons behind them). The panel works anonymously and iteratively on Pronoia — the AI-native data-collection platform built by the Metodix community as the successor to eDelphi.org.

What makes Pronoia different

Round 0 — AI orientation chat

Before Round 1, an AI chat orients and prepares each panelist in a personal dialogue — clarifying concepts, surfacing their perspective and warming up the argumentation.

AI-assisted facilitation

The facilitator gets AI support between rounds: argument summaries, feedback synthesis and suggested probing questions. Suggestions are drafts the facilitator owns and audits.

Rich visual analytics

Pulse, group comparison and per-thesis traffic lights detect differences between subgroups and support targeted panelist prompts.

Native DAE integration

Panel data flows into the analysis pipeline (CP-0) with no export/import friction — first analyses can begin while the panel still runs.

The phases inside a round

Every numbered round runs the same four phases — deliberately separating “form your own view” from “engage with others”.

1a Response

Answer each thesis independently. Others' answers stay hidden until you commit — capturing genuine, un-anchored judgement.

1b Dialogue

Read the panel, comment and reply with the result graph visible — exchanging reasoning, not just numbers.

Revision

Update your position in light of the dialogue. The change is tracked as a revision, so movement can be measured.

Closed

The round is locked (read-only) — freezing a comparable snapshot before the next round opens.

Four iteration modes (delphi_mode)

Argumentative — staged, argument-first (default) Classical — blind feedback + optional dialogue Real-Time (RTD) — one continuous, always-open round Barometer — recurring waves, trend over time
The visibility ladder — blind · after-submit · open
What a panelist can see is set per phase — the main lever against anchoring. Blind: others' answers hidden for the whole phase (max protection). After-submit: no anchoring before you commit, then the panel and aggregate graph open (the eDelphi norm). Open: the panel is visible throughout. A policy preset encodes the stance; individual phases can still be overridden per round.
privacy Per-study AI mode — privacy by design
Each study sets an AI mode — off / assist / full — enforced on every AI call. With AI off, no panel data leaves the system; the mode and GenAI disclosure travel with the data through Planning → Pronoia → DAE. Group reporting is k-anonymous (k ≥ 3) and AI panelists are always labelled “AI”.

The iteration

Round 0AI chat orients each panelist
→
Round 1positions + arguments
→
AI feedbackdistributions, group gaps, prompts
→
Round 2…nrevisions, counter-arguments
→
Closeconsensus & mapped dissensus

Three routes into analysis

Route A
Authentic panel

A real expert panel. The gold standard: genuine stakes, perspectives and surprises.

Route B
Simulated panel

The Argument Simulator generates synthetic, orientation-aware argumentation — always with GenAI disclosure downstream.

Route C
Hybrid

Real panel data complemented by simulation — to pre-test theses or fill perspective gaps.

What flows to phase 3: numeric stance distributions, full argument texts, panelist metadata and the Planning Handoff Package — received and validated by DAE checkpoint CP-0.

Phase 3 · DAE — Delphi Analysis Ecosystem

Analysis & Reporting — the 21-checkpoint pipeline

An orchestrated pipeline of 8 phases and 21 checkpoints (CP-0…CP-20). AI agents do the heavy analytical lifting; quality gates route every interpretive step to human validation. Red-zone phases require a fresh session and human sign-off. Click a phase:

CP-0–2 Foundation — validate & quantify
MaxPanelist validates the panel (PQS); MaxQuant builds the quantitative base (distributions, clustering, ECI). Python-backed.
CP-3–5 Content trilogy — what the panel argued
  • MaxDA (CP-3) — dialogue analysis: the moves peers make on each other's comments (adopt · qualify · extend · challenge · reframe).
  • MaxAA (CP-4) — argument analysis: Toulmin structure + the epistemic 7×3 coding (knowledge base × function).
  • MaxDiscourse (CP-5) — critical discourse analysis: power, frames and what stays unsaid.
CP-7–10 System trilogy — depth · dynamics · purpose
  • MaxCLA (CP-7) — Causal Layered Analysis, the depth dimension: litany → systemic → worldview → myth.
  • MaxMLP (CP-8) — Multi-Level Perspective, the dynamics dimension: landscape / regime / niche and transition paths.
  • MaxCATWOE (CP-9) — Soft Systems, the purpose dimension: who transforms what, for whom, under which worldview.
  • System synthesis (CP-10) integrates the three; adaptive panel supplementation (CP-7b) fills thin coverage.
CP-11–13 Futures trilogy — signals & tensions
  • MaxWS (CP-11) — weak signals: early/marginal change cues, PESTLE-classified.
  • Signal validation (CP-12) — which signals hold up under scrutiny.
  • MaxTension (CP-13) — tension map: structural, well-understood dissensus that feeds scenario axes.
CP-14 Grand synthesis — integrate every lens
MaxGrandSynthesis: cross-trilogy integration, keystone validation, stability triad, polyphonic synthesis and falsification. RED ZONE.
CP-15–18 Report — scenarios & stakeholder reports
MaxScenarios (six scenario paths) + MaxReports (27 templates, audience-specific, with data provenance). RED ZONE.
CP-19–20 Continuity — case library & learning
Stores the run and its lessons; feeds the improvement queue and the shared case library.

⚙ DA-Quality — the engine running underneath

Smoke & adversarial tests

Each analysis is challenged before acceptance.

QVG — validity gate

When AI validity is insufficient, the work is routed to human coding.

Staged κ agreement

Human–AI coding agreement measured and staged (lock vs candidate codes).

Data provenance (DPP)

Every claim tagged to its source: panel / analysis / researcher / hypothesis.

Falsification (FLC)

Popper-style attempts to break the synthesis before publishing it.

Phase gates & RED ZONE

Gate validator between phases; high-interpretation zones force session splits and human sign-off.

Why it is built this way

Foundations & the learning loop

The ecosystem is not a bag of techniques — each layer operationalises a distinct epistemic tradition.

🗣
Dialogue
Socrates

Knowledge emerges through structured questioning — the Delphi conversation itself.

⚖
Argumentation
Toulmin

Claims, grounds, warrants and backings — the skeleton of argument analysis (MaxAA).

🎭
Polyphony
Bakhtin

Many voices kept distinct in synthesis — no premature merging into one “average”.

🤝
Communicative rationality
Habermas

Fair, domination-free discourse as the panel ideal — and a lens for discourse analysis.

🔨
Falsification
Popper

Findings must survive attempts to break them — built into the Grand Synthesis gates.

🔭
Perspective realism
Hautamäki

Every panel sees from somewhere — perspective health is measured, not assumed (M5, CP-14c).

Method lineage inside the System Trilogy

Inayatullah → CLA

Causal Layered Analysis: litany → systems → worldview → myth. The depth dimension (CP-7).

Geels → MLP

Multi-Level Perspective: landscape, regime, niche and transition pathways. The dynamics dimension (CP-8).

Checkland → CATWOE

Soft Systems Methodology: who transforms what, for whom, under which worldview. The purpose dimension (CP-9).

The learning loop — the ecosystem improves itself

PlanM1a–M6
→
CollectPronoia / Simulator
→
Analyse & reportDAE CP-0…CP-18
→
Case Librarypublish · CP-19/20 · M6
↺
Next studystarts smarter
Two parallel memories: the planning side keeps Case Cards and retrospectives (M6); the analysis side keeps run-level continuity and an improvement queue (CP-19/20). Both publish to a shared case library.