TOPICALAUTHORITY.ORG TAO / ROOT

Decisions Under Uncertainty

TOPICALAUTHORITY.ORGDEC / 08 · UNCERTAINTY NODE ACTIVESTATE SPACE / ROBUSTNESS / INFORMATION VALUE
DECISION INTELLIGENCE/DECISIONS UNDER UNCERTAINTY
DEC / 08 UNCERTAINTY CONTROL SYSTEM

Decisions under uncertainty.

Decision quality under uncertainty does not require pretending to know the future. It requires representing what is unknown, how it could change the choice, whether it can be reduced before the deadline and which action remains acceptable across plausible states.

CORE RULEDo not replace uncertainty with one unsupported probability. Preserve ranges, scenarios, model limits and unknowns; then match commitment strength to exposure, reversibility, time-to-harm and the value of additional information.
01 CANONICAL DEFINITION

Uncertainty is incomplete knowledge about states, effects or models that matter to a decision.

It becomes decision-relevant when plausible alternatives could change option feasibility, ranking, timing, escalation or commitment.

Decision-making under uncertainty is the disciplined selection, staging or deferral of action when relevant future states, causal mechanisms, evidence quality or outcome magnitudes are not fully known. The objective is not to eliminate uncertainty, but to control exposure while preserving value and learning.

UNCERTAINTY ≠ RISKRisk usually assumes consequences and probabilities can be represented; uncertainty may include contested or unavailable probabilities.
UNCERTAINTY ≠ IGNORANCESome unknowns are structured as ranges, scenarios or competing models even when precise likelihood is unavailable.
ACTION ≠ CERTAINTYA decision can be justified when residual uncertainty is explicit and proportionate to the commitment.
02 UNCERTAINTY TAXONOMY

Different unknowns require different controls.

Collecting more data does not solve irreducible variability, model misspecification or disagreement about what the evidence means.

01 / ALEATORY

Variability

Inherent variation across events, users, environments or time. Represent with distributions, ranges and buffers.

02 / EPISTEMIC

Knowledge gap

Uncertainty reducible through observation, testing, research or verification.

03 / MODEL

Structural uncertainty

Uncertainty about mechanisms, assumptions, functional form or omitted relationships.

04 / AMBIGUITY

Multiple interpretations

Experts, sources or stakeholders assign different meanings or probability structures.

05 / DEEP

State-space uncertainty

Plausible futures cannot be ranked reliably or the relevant model set is incomplete.

06 / TEMPORAL

Context drift

Current relationships may change before or during implementation, invalidating the original frame.

03 REPRESENTATION CONTROL

Use the weakest claim the evidence can actually support.

Precision should follow evidence, not presentation preference.

Knowledge stateValid representationUse whenDo not claimDecision control
POINT + ERROREstimate with interval and method.Model and data are defensible within scope.Exact future outcome.Sensitivity to estimate range.
PROBABILITY RANGEBounded likelihood or distribution.Frequency or calibrated judgment exists.Precision beyond calibration.Expected value plus tail test.
SCENARIOSInternally coherent alternative states.Likelihood ordering is weak or disputed.Scenario as prediction.Robustness and regret across states.
COMPETING MODELSSeveral causal accounts retained.Structure or mechanism is uncertain.One model as settled fact.Choose action resilient to model disagreement.
QUALITATIVE BANDSDefined evidence-anchored categories.Quantification would be false precision.Arithmetic distance between labels.Thresholds and explicit rationale.
EXPLICIT UNKNOWNUnresolved proposition and possible decision effect.Evidence is absent or state space incomplete.Zero, neutral or assumed safety.Buffer, monitoring, stage or escalate.
04 INTERACTIVE ACTION-CALIBRATION ENGINE

Match the action posture to uncertainty and exposure.

Adjust the four controls. The output is a transparent diagnostic posture—not an automated decision or proprietary score.

DECISION CONDITIONS

0 means low; 100 means high. The engine applies explicit rules to suggest the next governance posture.

RULE-BASED POSTURE / NOT A DECISION

STAGE THE COMMITMENT

Use a bounded pilot or reversible step that produces discriminating evidence before exposure increases.

EVIDENCE BURDENHIGH
ACTION SIZEBOUNDED
REVIEW SPEEDFAST
ESCALATIONCONDITIONAL
Interpretation: the rules prioritize reversible action when delay is costly, information purchase when uncertainty is reducible and time permits, and escalation when high exposure combines with low reversibility. Validate against the actual decision boundary.
05 SCENARIO FAN

Scenarios explore decision-relevant divergence—not decorative futures.

Each scenario needs a causal logic, trigger, option consequence and monitoring signal. It is not a prediction and should not be assigned arbitrary probabilities.

Decision divergenceInclude only futures that can change option eligibility, ordering, timing or control.TEST / DOES ACTION CHANGE?
Internal coherenceDrivers, dependencies and outcomes must be compatible inside each scenario.TEST / CAUSAL STORY
Boundary conditionsState what must be true for the scenario to remain relevant.TEST / APPLICABILITY
Early indicatorsDefine observable signals that update which scenario deserves attention.TEST / MONITORING
Response pathPredefine proceed, adapt, contain, stop or escalate rules.TEST / ACTIONABILITY
06 ROBUSTNESS & REGRET

The highest expected result is not always the strongest decision.

When probabilities are weak, examine whether an option remains acceptable across plausible states and how severe regret becomes if the assumed state is wrong.

Option postureUpside stateReference stateStress stateBreakpointRegret profileInterpretation
Full commitmentVERY HIGHHIGHLOWSEVERELow if forecast is right; extreme if wrong.Fragile / forecast-dependent.
Staged commitmentMEDIUMHIGHMEDIUMCONTROLLEDModerate foregone upside; limited downside.Robust with learning value.
External partnerHIGHMEDIUMMEDIUMCONTROLLEDControl sacrificed; exposure limited.Flexible but dependency-sensitive.
Defer under triggerLOWMEDIUMHIGHHIGHHigh missed-upside regret; low loss regret.Valid only when waiting preserves option.
07 VALUE OF INFORMATION

Buy information only when it can change the action enough to justify delay and cost.

More research is not automatically valuable. The relevant comparison is expected decision improvement minus inquiry cost, delay cost and the risk that the opportunity changes before evidence arrives.

HIGH VALUE / FAST

Investigate now

Evidence can arrive before the deadline and may change a material decision hinge.

ACTION: TEST / VERIFY / DILIGENCE
HIGH VALUE / SLOW

Stage + learn

Information matters, but waiting fully would destroy value or increase harm.

ACTION: REVERSIBLE STEP
LOW VALUE / FAST

Do not over-research

The answer is cheap, but plausible findings would not alter the chosen posture.

ACTION: RECORD + PROCEED
LOW VALUE / SLOW

Act robustly

Uncertainty cannot be reduced proportionately before action loses value.

ACTION: BUFFER / MONITOR / ESCALATE
08 UPDATE DISCIPLINE

New evidence should update the map—not erase the previous state.

Preserve the prior belief or range, the new observation, the updating rule and the resulting action change. A post-hoc narrative is not an update trail.

01 / PRIOR

Starting state

Initial range, scenario set and assumptions.

02 / SIGNAL

New evidence

Observation, provenance and error structure.

03 / RELEVANCE

Applicability

Whether signal addresses the uncertain claim.

04 / UPDATE

Revised state

Changed range, scenario weight or model set.

05 / HINGE

Decision effect

Threshold, ranking or eligibility change.

06 / RECORD

Action revision

Proceed, adapt, pause, stop or escalate.

09 FIVE APPLIED UNCERTAINTY MAPS

Different domains require different uncertainty postures.

Select a scenario to inspect the unknown, consequence, reducibility and robust action.

10 FAILURE MODES

Uncertainty is often hidden by the format of the analysis.

These patterns create confidence without increasing knowledge.

FailureWhat it looks likeWhy it failsRepairControl question
Single-point futureOne forecast presented as the expected state.Suppresses range, model and tail uncertainty.Use interval, scenarios and sensitivity.What credible states are excluded?
Probability theatrePrecise percentages without calibration or basis.Numbers imply evidence that does not exist.Expose source and use ranges or qualitative states.Who is calibrated to this reference class?
Unknown = zeroMissing effect entered as neutral.Absence of evidence becomes evidence of absence.Mark unknown and test decision impact.Could this blank change the choice?
Average-only reasoningExpected outcome hides tail or distribution.Low-frequency severe loss disappears.Show downside, threshold and bearer.Who absorbs the extreme case?
Analysis paralysisAction delayed until all uncertainty resolves.Delay itself creates cost and may not buy useful evidence.Use value-of-information and reversible action.Will inquiry change action before deadline?
Scenario decorationNamed futures with no action consequences.No decision relevance or monitoring rule.Attach option effect, trigger and response.What changes under this scenario?
11 RETRIEVAL CONTRACT

Store uncertainty as a structured state—not a confidence adjective.

RAG and analytical systems need the uncertain proposition, type, range, model basis, decision effect and update trigger beside every result.

DEC / 08 MACHINE-READABLE UNCERTAINTY

Make the unknown queryable.

This structure prevents “moderate confidence” from being retrieved without its meaning. It does not guarantee calibrated probability or correct action.

{
  "decision_id": "DEC-08-001",
  "uncertainty_id": "UNC-017",
  "proposition": "migration completes inside 12 weeks",
  "type": ["epistemic", "model"],
  "representation": {"range": "8–20 weeks", "basis": "three comparable migrations"},
  "decision_hinge": true,
  "option_effect": {"option_id": "OPT-B", "effect": "deadline feasibility"},
  "reducibility": "high",
  "information_action": "run dependency discovery sprint",
  "robust_posture": "stage commitment",
  "monitor": "critical dependency count",
  "reopen_if": "upper bound exceeds 16 weeks",
  "updated_at": "YYYY-MM-DD"
}
Methodological basis: Government appraisal guidance treats uncertainty, risk, optimism bias, sensitivity and scenario analysis as distinct controls rather than reasons for false precision. IPCC guidance also separates confidence in findings from quantified likelihood. See the UK Green Book and IPCC uncertainty guidance.
12 FREQUENT QUESTIONS

Decisions under uncertainty, clarified.

Operational answers to common uncertainty-control errors.

What is the difference between risk and uncertainty?

Risk commonly refers to uncertain consequences represented with known or estimable probabilities. Uncertainty is broader and can include unknown probabilities, competing models, ambiguous evidence, changing context and incomplete state spaces.

Should a decision wait until uncertainty is reduced?

Only when the expected value of information exceeds inquiry cost and delay, and when evidence can arrive before action loses value. Otherwise use a robust, bounded or reversible action with monitoring.

What is a robust decision?

A robust decision remains acceptable across several plausible states or models, even if it is not optimal in any single forecast. Robustness must be defined against explicit objectives, constraints and downside thresholds.

What is minimax regret?

Regret is the loss relative to the best option for the state that actually occurs. A minimax-regret posture selects the option whose worst regret across considered states is smallest. It depends on the scenario set and should not be treated as universally correct.

When should probabilities not be used?

Do not use precise probabilities when the event space is unstable, reference class is absent, expert judgment is uncalibrated or competing models cannot support one distribution. Use ranges, scenarios or explicit unknowns instead.

How is uncertainty different from confidence?

Uncertainty describes what is not known about states, parameters, models or consequences. Confidence is a judgment about support for a finding. High confidence in a broad range can coexist with substantial outcome uncertainty.

What is the value of information?

It is the expected improvement in decision quality from additional evidence, compared with the cost of inquiry, cost of delay and risk that the decision environment changes before the information arrives.

Can action reduce uncertainty?

Yes. Pilots, staged commitments, reversible containment and monitored deployment can generate evidence. The action must be designed with discriminating measures and a real next decision gate.

DEC / 08 RESEARCH NETWORK

Continue through the complete Decision Intelligence system.

Uncertainty transforms trade-offs into commitment, error cost, escalation and review rules.

DECISION INTELLIGENCE · COMPLETE OVERVIEW

Decision Intelligence

Structure choices through objectives, alternatives, evidence, uncertainty, trade-offs, commitment and review.

DECISION INTELLIGENCE / UNCERTAINTY PRINCIPLE

Do not wait for certainty. Design an action that can survive being wrong.

Represent the unknown honestly. Identify the decision hinge. Compare information value with delay. Prefer robust and reversible steps where possible. Monitor the state, preserve the update trail and escalate when exposure exceeds authority.

TOPICALAUTHORITY.ORG / DECISION INTELLIGENCEUNKNOWN → SCENARIOS → ROBUSTNESS → ACTION → UPDATE
TAO / CONTACT · DIRECT TRANSMISSION Have an asset, domain or market position to investigate? ENTER CONTACT SYSTEM →
DIGITAL ASSET INTELLIGENCE + EXECUTION
EXECUTED BY
BB DIGITALNA AGENCIJA

Investigation, consulting and execution of digital assets, premium-domain strategies, information architecture, semantic systems, websites and agreed digital growth plans.

TOPICALAUTHORITY.ORG / SEMANTIC INTELLIGENCE SYSTEM BB DIGITALNA AGENCIJA / BB.HR