A weak signal is an invitation to investigate, not a forecast
Horizon scanning examines diverse sources for early signs of change, connects fragments into testable emerging issues and routes decision-relevant findings into monitoring, foresight and action.
Organisations are often surprised not because there was no early evidence, but because scattered observations never became a shared question. A laboratory result, regulatory consultation, unusual investment pattern and change in public language may each look minor in isolation. Scanning creates a disciplined place to collect and compare them.
The danger is symmetrical. A team can overlook unfamiliar developments, or amplify novelty into alarm. Good horizon scanning therefore records provenance, distinguishes signals from trends and events, clusters cautiously, uses disconfirming search and defines escalation pathways before attention is captured by a dramatic story.
Chapter contents
- 1. Start with a decision-facing scan question
- 2. Design a diverse and traceable source portfolio
- 3. Record weak signals as observations before interpretations
- 4. Distinguish signals, trends, drivers, events and wild cards
- 5. Cluster signals without manufacturing a story
- 6. Worked example: turn scattered signals into an emerging issue
- 7. Map drivers, actors and cross-domain interactions
- 8. Assess novelty, plausibility, impact and decision relevance separately
- 9. Search deliberately for disconfirmation and absence
- 10. Protect the scan from popularity, recency and automation bias
- 11. Define escalation pathways and thresholds before urgency arrives
- 12. Maintain signposts, feedback and institutional memory
Choose a reading route
- Route 1: Frame the scan — decision, horizon and domain boundaries
- Route 2: Collect signals — sources, weak signals and observables
- Route 3: Build issues — clustering, drivers and worked synthesis
- Route 4: Challenge meaning — bias, alternatives and disconfirmation
- Route 5: Escalate and learn — thresholds, ownership and revision
How to use this guide
Start with the chapters that match your immediate decision, then return to the worked example and audit chapters. Each chapter separates a model from the evidence needed to use it responsibly. A claim is not strengthened merely by being technical; it is strengthened when assumptions, observations, alternatives and limits can be inspected.
| Layer | Question | Minimum record |
|---|---|---|
| Concept | What exactly is being represented? | Definitions and boundary |
| Method | How will the reasoning be carried out? | Steps, inputs and assumptions |
| Transfer | Would it work in a changed setting? | A deliberately altered case |
| Audit | What could prove it wrong? | Evidence, limits and revision trigger |
Chapter 1. Start with a decision-facing scan question
The idea
Scanning becomes useful when it serves a defined decision, horizon and audience. A universal search for “the future” produces noise and unowned observations.
A disciplined method
Write the focal decision, time horizon, geographic and system boundary, stakeholders, excluded domains and the actions that new evidence could change.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which present choice the scan is intended to inform. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A school network scans five years ahead for changes affecting assessment, staffing and learner support, rather than collecting every education headline.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the decision from long-term facilities to next-term staffing and rebuild the horizon and source mix.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Narrow framing can miss cross-domain change, while broad framing overwhelms analysts.
Evidence to retain: scan brief, horizon, boundaries, decision owner, exclusions and review date
Common misconception: A scan should begin with the largest possible keyword list.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 2. Design a diverse and traceable source portfolio
The idea
Weak signals often appear outside familiar channels. Diversity across research, regulation, patents, investment, practice communities and affected groups reduces institutional blind spots.
A disciplined method
Create source categories, justify coverage, record access dates and search methods, and add deliberate peripheral sources rather than relying only on prestige or popularity.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which communities and evidence systems are systematically absent. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A technology scan combines peer-reviewed papers, standards drafts, procurement notices, user forums and disability advocates.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Remove one high-volume source and test whether the issue map changes or merely becomes less noisy.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Source diversity does not guarantee representativeness, and inaccessible or non-digitised evidence remains undercounted.
Evidence to retain: source register, rationale, query history, language coverage and access constraints
Common misconception: More sources automatically produce a more complete horizon.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 3. Record weak signals as observations before interpretations
The idea
A weak signal is an early, incomplete indication of possible change. It should be captured as a dated observation with context, not immediately promoted into a trend or prediction.
A disciplined method
Use a signal card containing claim, source, date, domain, novelty, possible implications, alternatives, confidence and follow-up question. Quote sparingly and link to the original.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is what was directly observed and what the analyst inferred. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Three municipalities test a new cooling standard. The observation is the trials; nationwide adoption is only one possible implication.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Discover that two reports copy the same press release and reduce apparent corroboration.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Signals can be deliberate publicity, temporary experiments or artefacts of measurement and media attention.
Evidence to retain: original source, publication and event dates, independent-source count and analyst note
Common misconception: A surprising headline is a weak signal by definition.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 4. Distinguish signals, trends, drivers, events and wild cards
The idea
These categories answer different questions: a signal hints, a trend describes direction, a driver exerts influence, an event occurs, and a wild card challenges assumptions through low-probability high-impact change.
A disciplined method
Label each item provisionally, state the evidence for the label and permit reclassification as frequency, duration and causal understanding improve.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which temporal and causal claim the label implies. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A single prototype is a signal; sustained falling sensor cost is a trend; regulation is a driver; a ban is an event.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Turn the prototype into repeated adoption data and document when the classification changes.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Categories overlap and should organise inquiry rather than create false precision.
Evidence to retain: classification history, observations, frequency, causal hypothesis and revision reason
Common misconception: Anything important but uncertain should be called a wild card.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 5. Cluster signals without manufacturing a story
The idea
Related signals can reveal an emerging issue, but clustering can also force independent observations into a fashionable narrative.
A disciplined method
Group by mechanism, affected system and consequence; preserve unclustered items; compare alternative clusters; and record why each signal belongs or does not belong.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is whether the proposed issue explains relationships better than coincidence. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Changes in battery chemistry, recycling rules and grid procurement form a possible storage-resilience issue, while a celebrity product launch remains peripheral.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Remove the most vivid signal and see whether the cluster still has a coherent mechanism.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Analyst expectations and keyword taxonomies can determine clusters before evidence does.
Evidence to retain: signal membership, similarity rule, rejected links, alternative clusters and facilitator notes
Common misconception: A larger cluster is necessarily a stronger emerging issue.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 6. Worked example: turn scattered signals into an emerging issue
The idea
A transparent synthesis shows how observations become a bounded issue statement, indicators and questions for deeper foresight.
A disciplined method
Use five fictional signals, identify a shared mechanism, map affected actors, compare two alternative explanations and draft an issue statement with confidence and next evidence.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which evidence justifies escalation beyond routine monitoring. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
New repair rules, modular product trials, spare-part marketplaces, technician shortages and warranty changes suggest an emerging repair-capacity transition.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Remove the regulatory signal and assess whether commercial experimentation alone supports the same issue statement.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: The fictional case demonstrates synthesis, not a forecast that the transition will occur.
Evidence to retain: signal cards, cluster map, alternative explanations, indicator list and confidence rationale
Common misconception: Five related observations prove a stable trend.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
| Signal | Possible meaning | Alternative |
|---|---|---|
| Repair rule consultation | Policy opening | Routine review |
| Modular trials | Design shift | Limited pilot |
| Parts marketplace growth | New ecosystem | Temporary demand |
| Technician shortage | Capacity constraint | Local mismatch |
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 7. Map drivers, actors and cross-domain interactions
The idea
An emerging issue develops through incentives, capabilities, institutions and constraints. Driver mapping moves beyond counting mentions toward mechanisms and leverage points.
A disciplined method
Identify political, economic, social, technological, legal and environmental drivers; add actors, feedbacks, dependencies and countervailing forces.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which mechanism could amplify or suppress the issue. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Cheaper sensors accelerate adoption, privacy rules constrain deployment and workforce standards change implementation speed.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change one driver direction and trace which implications reverse or remain robust.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Driver maps are hypotheses and can disguise uncertainty behind arrows.
Evidence to retain: driver evidence, actor interests, feedback assumptions, dependencies and contested links
Common misconception: A trend extrapolation explains why the change is happening.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 8. Assess novelty, plausibility, impact and decision relevance separately
The idea
A signal can be novel but irrelevant, plausible but low impact, or high impact but weakly evidenced. Separate ratings prevent one dramatic dimension from dominating.
A disciplined method
Use ordinal ratings with written rationales, compare raters, state uncertainty and prioritise follow-up based on the decision rather than a single composite score.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which dimension makes the issue worth attention. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A credible standards revision has moderate impact and near-term relevance; a speculative breakthrough has high impact but low evidence.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the decision owner and observe how relevance changes while evidence does not.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Scales are judgement aids, not calibrated probabilities; averaging can hide disagreement.
Evidence to retain: rating definitions, rationales, dissent, evidence quality and sensitivity analysis
Common misconception: Multiplying novelty and impact produces an objective priority score.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 9. Search deliberately for disconfirmation and absence
The idea
Scanning teams should seek evidence that an issue is stalled, local, duplicated or based on one promotional source. Absence is informative only when collection coverage is adequate.
A disciplined method
Write competing explanations, assign targeted searches, check source independence and look for non-adoption, failed pilots, counter-regulation and resource constraints.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which observation would reduce confidence most. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A supposed adoption wave rests on vendor announcements; procurement data show few completed deployments.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Improve coverage in a previously unsearched language and reassess whether silence was real or a collection gap.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Negative evidence can be hard to find and non-reporting does not prove non-occurrence.
Evidence to retain: disconfirmation queries, coverage map, failed cases, independence checks and confidence changes
Common misconception: A lack of headlines shows the issue is not developing.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 10. Protect the scan from popularity, recency and automation bias
The idea
Search rankings, social sharing and automated summaries overrepresent accessible, recent and repeated material. Novelty can be a property of the feed rather than the world.
A disciplined method
Rotate source sets, cap duplicate coverage, sample low-volume channels, expose recommendation effects and require human inspection of original documents before coding.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which signals are visible because of platform selection. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A dashboard shows twenty articles that all cite one report; deduplication reduces them to one evidential origin.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Change the ranking algorithm and compare which issue clusters appear or disappear.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Human review has its own language, expertise and institutional biases.
Evidence to retain: deduplication graph, platform settings, sampling method, reviewer notes and error audit
Common misconception: An AI summary makes a signal reliable because it combines many documents.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 11. Define escalation pathways and thresholds before urgency arrives
The idea
Escalation connects scan output to deeper analysis, scenario work, collection or decision review. Predetermined criteria reduce arbitrary alarm and silent neglect.
A disciplined method
Assign issue owners, define monitoring, analytical and decision thresholds, specify evidence packages and time limits, and include a de-escalation or archive route.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is what combination of evidence changes the organisational response. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
Two independent deployments plus a regulator timetable trigger a structured assessment; one vendor announcement remains on watch.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Accelerate the timetable while evidence quality stays unchanged and test whether urgency or confidence drives escalation.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Thresholds can be gamed or become stale as the environment changes.
Evidence to retain: threshold definitions, trigger evidence, owner, action, dissent and decision date
Common misconception: Every weak signal should go directly to senior leaders.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Chapter 12. Maintain signposts, feedback and institutional memory
The idea
A horizon-scanning system learns by revisiting signals, recording false alarms and misses, and linking emerging issues to scenarios and decisions. Old records prevent repeated rediscovery.
A disciplined method
Set review cadences, update indicator status, record issue merges and closures, compare earlier expectations with outcomes and preserve reasons for decisions.
A useful working record makes the sequence visible: first state the question and unit of analysis; next record the observations and their provenance; then apply the stated rule; finally compare the result with plausible alternatives. For this chapter, the decisive choice is which evidence should reopen a closed issue. Keeping that choice explicit prevents a later reader from mistaking a convenient convention for a fact about the world.
Worked example
A dismissed supply-risk signal returns after a standards change; the archived rationale shows which assumption no longer holds.
Work the example twice. On the first pass, follow the stated assumptions. On the second, reverse one assumption, remove one piece of evidence and ask whether the conclusion, ranking or boundary changes. This small stress test matters because polished outputs can hide sensitivity. The objective is not to force certainty; it is to locate the evidence on which the conclusion actually turns.
Changed-context transfer check
Transfer the system to a new team and test whether records explain prior classifications without oral history.
A transfer succeeds only when the structure of the problem survives the change. Match roles, dependencies and measurement conditions rather than surface vocabulary. Record what was preserved, what changed, and why the method still applies—or why it does not. If the new context changes the data-generating process or the people bearing the burden, re-specify the model instead of importing the old answer.
Limits, evidence and revision
Limit: Retrospective evaluation is affected by hindsight and changing objectives.
Evidence to retain: versioned signal cards, signposts, decisions, outcomes, lessons and reopening criteria
Common misconception: Deleting weak signals that went nowhere improves the quality of the archive.
A gold-standard conclusion is therefore conditional. It should name the observation period, excluded cases, uncertainty and the event that would trigger review. Negative findings and unresolved contradictions belong in the record. They protect future readers from repeating a hidden assumption and make correction possible without reconstructing the entire analysis from memory.
Practical evidence check
Can another reader recover the input, reproduce the intermediate decision and identify at least one observation that would change the result? If not, the chapter has produced an explanation but not yet an auditable finding.
Field workbook
Create twelve signal cards for one bounded question using at least five source types. Deduplicate their origins, form two alternative clusters, score dimensions separately and define one monitoring, one analytical and one decision threshold.
- Write the decision question in one sentence and name the decision owner.
- List observations separately from interpretations, with dates and provenance.
- Record at least two plausible alternatives before choosing a preferred account.
- Run the changed-context test and note which assumptions fail.
- State the limit, confidence and next review trigger in plain language.
| Layer | Evidence | Failure signal |
|---|---|---|
| Collection | Diverse traceable sources | Feed or prestige bias |
| Signal | Observation separated from inference | Headline becomes forecast |
| Synthesis | Alternative clusters and drivers | Narrative forced early |
| Escalation | Threshold and owner | Urgency without evidence |
Continue through the verified ecosystem
- Scenario analysis — plausible futures and morphological spaces
- Indications and warning — signposts, thresholds and surprise
- Assumption-based planning — signposts and adaptive actions
- Key assumptions check — support, confidence and reversal
- Change analysis — baselines, drivers and alternatives
Frequently asked questions
What is horizon scanning?
A systematic, ongoing examination of diverse sources for early signs of change that may affect future decisions.
Is a weak signal a prediction?
No. It is an incomplete observation or indication that needs investigation and comparison with alternatives.
When does a signal become an emerging issue?
When several observations and mechanisms support a bounded change question with decision relevance, while uncertainty remains explicit.
Why search for disconfirming evidence?
It tests whether a cluster is independent, persistent and mechanism-based rather than repeated promotion or analyst expectation.
Should every signal be escalated?
No. Most should remain monitored, merged, reframed or closed under transparent criteria.
How should AI support scanning?
It can help retrieve, cluster and summarise, but source inspection, provenance, deduplication and accountable judgement remain necessary.
Authoritative sources and further reading
- UK Government Futures Toolkit — official horizon-scanning and foresight methods
- UK Analysis Function: Horizon scanning and futures thinking — signals, PESTLE scanning and government analysis
- OECD: Building capacity in technology horizon scanning — systematic practice and capability
- OECD Observatory of Public Sector Innovation: Futures and foresight — horizon scanning in strategic foresight
- OECD: Using foresight to anticipate emerging critical risks — horizon scanning, weak signals and emerging-risk analysis
This guide supports anticipatory analysis. It does not predict the future, assign calibrated probabilities to weak signals or justify surveillance, alarm or action without proportionate evidence and authority.