🔭 OFEK — Project Manual
How to launch and manage a foresight project, in ten steps. For analysts and project admins. The reader's guide covers the everyday review screens.
The theory in one page
OFEK implements the Houston Foresight Framework (Hines & Bishop, University of Houston), as applied to qualitative-analysis software in Loxton's Using MAXQDA for Strategic Foresight (2021). The framework is a six-step loop. Each step has its own tabs in OFEK:
| Step | Question it answers | Where in OFEK |
|---|---|---|
| 1 Framing | What world are we studying, for whom, and why now? | Overview, Logbook, Documents, Stakeholders, Corpus Voice |
| 2 Scanning | What evidence of change is out there? | Signals, Source finder, TZOFIM Watches, Radar, Timeline |
| 3 Forecasting | Which forces of change do the signals add up to? | Codes, Analytics, Leaders |
| 4 Visioning | What futures follow, and which one do we want? | Visioning (assumptions, futures wheel, what-if, four scenarios) |
| 5 Planning | What should we do about it? | Reports (Foresight Report, simulations) |
| 6 Acting | Did it work, and what changed since? | PDSA board, living reports, the next review cycle |
Four ideas carry the whole method. A signal is one piece of evidence about the future (an article, a report, a speech), tagged by STEEP direction (Social, Technological, Economic, Environmental, Political), by horizon (H1 within about two years, H2 mid-term, H3 far future) and by four 0–5 scores (impact, plausibility, novelty, credibility). Signals are coded into a taxonomy; codes that recur across many signals become forces of change. Forces are combined into four archetype futures: Baseline, Adjusted Equilibrium, Transformation and Collapse. The rule that makes it research rather than opinion: every sentence in a scenario or report is traceable to the signals that support it. The AI does the reading, tagging and drafting; the analyst frames, reviews, corrects and decides.
1Create the project
Workspace → New project. The name is the research question ("The future of X", "How Y is perceived abroad"). The domain description is one paragraph describing the world the study covers. Write it carefully: it is handed to the AI in every later step (classification, scoring, scenarios, reports). The Houston taxonomy (STEEP, H1–H3, system patterns, Diamond's five factors) is pre-seeded as codes.
2Frame it
Logbook: the key questions, why now, and every decision you take (this is the project's memory). Stakeholders: three to five groups and why each matters. Documents: upload the framing material (PDF, DOCX, TXT, a URL, or a pasted interview transcript) into the internal corpus. Corpus Voice then shows the recurring phrases of your own documents next to those of the outside world.
3Build the collection layer
Signals → Source finder → Find sources with AI. Enter the research question. The AI proposes news watches per language, RSS feeds (each URL live-checked), a press backfill by date range and institutional document indexes. Untick what you do not want, press Create selected, then Run all watches & sources now for the first harvest.
4Let the robots run
The scheduler runs every hour and respects each watch's frequency and daily cap. Check the Harvest run log on the Source finder page after the first day: a watch that brings noise gets tighter keywords, a source that brings nothing gets paused (Pause / Resume on the Watches page). Nothing needs to be run by hand after this, but ▶ Run now is always there.
5Review the queue, every day
Signals → Review queue. Each unreviewed item shows its lede, analysis, STEEP, horizon, event kind and score. Press ✓ Approve if it is real, relevant evidence, ✗ Reject if it is noise. Only approved signals count in the Radar, the Timeline, the analytics and the scenarios. Foreign-language signals: pick a language and press 🌐 Translate (the translation is kept for everyone), or open the original in Google Translate. Ten minutes a day keeps the corpus clean.
6Read and correct
Click a signal title. You see the lede, the AI analysis, the badges, the four scores with the AI's reasons, and the full text. Change what is wrong in Metadata & memo and Scores: the human value is stored beside the AI value, never over it. 🤖 Deep analysis extracts summary, claims and commitments, entities, stance, quotes and numbers. Code a segment attaches a code to an exact passage, with an optional weight.
7Build the code system
Codes: add or rename codes in the code families. Corpus Voice → Taxonomy builder → AI card sort clusters the corpus phrases into candidate codes; accept the ones that make sense. Domain map and Ishikawa turn the codes into hot topics with their key events and causes. Every code memo should say what the code includes and excludes: that is what keeps two analysts coding the same way.
8Analyze
Analytics holds the book's methods: code frequency, the co-occurrence heat map, the code map, Coding query, Lexical → synthetic codes, Entities & commitments, Corpus Q&A (answers only from the project's evidence, every sentence cited), the Entity network, the Before/After tracer and the Stance scan. Which future? tells you which archetype the approved signals point to. Leaders maps directed sentiment between leaders. The Radar and Timeline are the pictures you show to others.
9Vision
Visioning. Start with the Assumption register (add the assumptions the domain lives on; the taboo / missing-voice scan finds the ones nobody states; validate vs signals checks each against the evidence). Spin the Futures Wheel on one force of change to see its ripple effects. Vary it in the What-if simulator to see the picture with a force weakened or strengthened. Then Compose the four scenarios, one archetype at a time; read them as an editor, and remove any sentence that has no signal behind it.
10Report, act, maintain
Reports. Generate Foresight Report assembles framing, forces, assumptions, scenarios, action plan and a cited appendix; Generate Perception Scan is the media-image variant. Publish freezes a version, with a public share link and PDF export. Simulate rebuilds a published report under a shock and shows the changes as tracked changes. Decisions go to the PDSA board (plan, do, study, act). Then keep the loop alive: review the queue daily, re-run the report monthly, and log each decision. Team members are invited by the platform owner in Console → Users, with a role: admin (everything), analyst (collect, code, generate), viewer (read, ask, translate).