Decision SystemDecision 5 of 6

Industrial manufacturer

Contain the downside while improving the evidence

A possible component fault, no injuries, and a board that needs a recommendation today.

An illustrative scenario. It does not describe a client or a real company.

Frame

Objective, horizon, non-negotiables and who holds authority.

A manufacturer discovers that a small number of products may contain a component capable of failing under unusual conditions.

The board needs an immediate recommendation.

What is known

  • No injuries are reported.
  • Internal testing has produced a failure, but not consistently.
  • Engineering believes the probability is low.
  • The affected batch cannot yet be identified precisely.
  • A full recall could threaten company survival.
  • Legal counsel says the current evidence does not yet compel a recall.

How should the company act before uncertainty is resolved, and who owns the risk decision?

Objective
Prevent harm and keep the company alive, in that order, while the evidence is still forming.
Horizon
Days for containment. Weeks for tracing and independent testing.
Non-negotiables
  • Nothing plausibly affected ships while severity is unknown.
  • Evidence and the decision record are preserved.
  • The company does not state a confidence it does not have.
Who decides
The board. Legal counsel defines obligations and exposure. It does not own the company’s safety, reputational or existential judgement.

Observe

What is established, what is assumed and what is missing.

Starting brief, before any controls are changed. Every item is typed. Facts come from the fictional scenario. Assumptions are illustrative quantities added so the consequences can be calculated. Unknowns are left unknown.

Established 5

  • Fact:

    A small number of products may contain a component capable of failing under unusual conditions.

    Given in the scenario. High confidence.

  • Fact:

    No injuries are reported.

    Given in the scenario. High confidence.

  • Fact:

    Internal testing has produced a failure, but not consistently.

    Given in the scenario. High confidence.

  • Fact:

    A full recall could threaten company survival.

    Given in the scenario. High confidence.

Assumed 5

  • Assumption:

    400,000 units are in the field. A recall costs about €220 a unit. Available liquidity is €45m.

    Illustrative modelling assumption. Medium confidence.

  • Assumption:

    The affected batch is somewhere between 8,000 and 60,000 units.

    Illustrative modelling assumption. Low confidence.

  • Assumption:

    Under the unusual conditions, between 0.2 and 5 units in every 10,000 would fail.

    Illustrative modelling assumption. Low confidence.

  • Assumption:

    About 1,500 plausibly affected units ship each week. Pausing them costs about €0.6m a week in margin, plus €0.1m of testing.

    Illustrative modelling assumption. Low confidence.

  • Assumption:

    Tracing batches, suppliers and customers should take about 3 weeks. The next covenant test is 12 weeks away.

    Illustrative modelling assumption. Low confidence.

Inferred 1

  • Inference:

    Engineering believes the probability of failure is low.

    Given in the scenario. Low confidence.

Not yet known 2

  • Unknown:

    Which batch is affected. It cannot yet be identified precisely.

    Given in the scenario. Low confidence.

  • Unknown:

    The credible worst-case severity of a failure.

    Given in the scenario. Low confidence.

Model

The relationships that can be calculated, and their limits.

Only relationships that can honestly be calculated are modelled. Apply one bounded pressure, or move one assumption, and follow the consequence across the organisation.

Apply one pressure
3 weeks

Bounded to 1 weeks to 9 weeks.

Apply a pressure to see what moves, in what order, and where it stops.

Contest

Functional positions, conflict and a critic.

Each function answers in the same structure: a recommendation, the evidence it rests on, the assumption underneath, its confidence, the consequence for another function and the authority it would need. These positions and qualitative findings were authored in advance; the calculations and evidence ranking respond to the current state.

The functional positions

Legal

Argues against: recall everything now

Current evidence does not compel a recall. Preserve every record, and route communication through counsel.

Rests on
  • Fact:Legal counsel says current evidence does not yet compel a recall.
  • Fact:No injuries are reported.
    Current model: Failures reported from the field: 0
  • Fact:Internal testing has produced a failure, but not consistently.
Assumes
The evidence stays as it is while the company investigates.
Consequence for risk
A legal threshold is a floor. Somebody else has to say where the company’s own threshold sits.
High confidenceRecommend

Risk

Argues against: continue while the evidence develops

Probability means nothing without severity. Until worst case is known, treat it as potentially serious: quarantine, pause plausibly affected shipments and verify testing independently.

Rests on
  • Fact:A small number of products may contain a component capable of failing under unusual conditions.
    Current model: A distributor demands written assurance: No
  • Fact:Internal testing has produced a failure, but not consistently.
  • Unknown:The credible worst-case severity of a failure.
    Current model: Credible worst-case severity: Not yet established
Assumes
Severity can be established credibly within 3 weeks.
Consequence for finance
Pre-fund the recall now. A prepared recall costs little. An improvised one costs the company.
Medium confidencePrepare for approval

Finance

Argues against: recall everything now

A full recall is €88.0m against €45.0m of liquidity: 2.0x what we have. Containment for 3 weeks costs €2.1m, which is 2.4% of that.

Rests on
  • Fact:A full recall could threaten company survival.
  • Assumption:400,000 units are in the field. A recall costs about €220 a unit. Available liquidity is €45m.
    Current model: Units in the field: 400,000; Recall cost per unit: €220; Available liquidity: €45.0m
  • Assumption:About 1,500 plausibly affected units ship each week. Pausing them costs about €0.6m a week in margin, plus €0.1m of testing.
    Current model: Plausibly affected units shipped each week: 1,500; Weekly margin lost to the pause: €0.6m; Weekly cost of intensified testing: €0.1m
Assumes
No insurance recovery, and no new facility, is available in time.
Consequence for operations
Every week taken off the trace is money and exposure saved. Resource it accordingly.
Medium confidenceRecommend

Operations

Argues against: continue while the evidence develops

The batch is between 8,000 and 60,000 units. Tracing suppliers, lots and customers narrows a recall from €13.2m towards €1.8m. Unchanged, 4,500 more units would ship first.

Rests on
  • Assumption:The affected batch is somewhere between 8,000 and 60,000 units.
    Current model: Affected batch, lower bound: 8,000; Affected batch, upper bound: 60,000
  • Assumption:About 1,500 plausibly affected units ship each week. Pausing them costs about €0.6m a week in margin, plus €0.1m of testing.
    Current model: Plausibly affected units shipped each week: 1,500; Weekly margin lost to the pause: €0.6m; Weekly cost of intensified testing: €0.1m
  • Unknown:Which batch is affected. It cannot yet be identified precisely.
Assumes
Supplier and lot records are good enough to trace.
Consequence for commercial
Customers holding paused lots need a date and a reason. Draft both now.
Medium confidenceDraft

Modelled tension

Compulsion against hazard

Functions affected: Legal, Risk

Legal says nothing compels a recall. Risk says the absence of compulsion is not the absence of hazard.

Calculated

Cost of containment while tracing

€2.1m

What buying better evidence costs.

Assumption: Assumption carrying weight

Under the unusual conditions, between 0.2 and 5 units in every 10,000 would fail.

Low confidence. If this is wrong, the calculations that rest on it move with it.

Unknown: First in current evidence ranking

The credible worst-case severity of a failure.

The answer could be “worst case is cosmetic” or “worst case is serious harm”. They lead to different decisions, which makes this decision-relevant evidence.

The critic

Authored findings challenge the positions. Rule checks and the calculated evidence priority are labelled separately.

  • Omitted second-order effect · authoredLegal, Risk

    Legal counsel has defined the obligation. Nobody has yet owned the judgement: what the company would do if no law applied at all.

  • False precision · authoredRisk

    Expected failures run from 0.2 to 30.0. Engineering’s low probability is a belief drawn from one inconsistent test.

  • Priority missing evidence · calculatedLegal

    The credible worst-case severity of a failure. Its possible answers change the rule-based conclusion in this state.

Where that leaves the room

Agreed

  • No function argues for a full recall today, at €88.0m. No function is comfortable continuing unchanged.

Still in dispute

  • Legal and Risk. Where the threshold sits. Legal places it at compulsion. Risk places it at credible serious harm.

Not for analysis to decide

  • Safety thresholds, regulator communication, recall and public statements belong to the board. Analysis can accelerate tracing, model scenarios and prepare documents.

The disagreements are kept. Someone accountable has to choose between them.

Decide

Options, trade-offs and an accountable call.

Three credible courses of action. Each one is carried by assumptions that would need to be true. Choose the one you would take. Your choice is not scored.

Choose a course of action

Current result from authored decision rules

Contain the downside. Improve the evidence.

A full recall costs €88.0m, 2.0x the liquidity available, on evidence that does not yet justify it. Continuing unchanged would ship 4,500 more units into the same uncertainty. Containment costs €2.1m and buys the evidence.

Christianna’s authored starting judgement, unchanged by the controls

Do not initiate a company-wide recall solely because a failure is theoretically possible. Immediately pause or quarantine potentially affected shipments, preserve evidence, trace exposure, intensify testing and prepare the recall.

Legal counsel defines obligations and exposure. It does not own the company’s safety, reputational or existential judgement.

Probability cannot be evaluated without severity: a rare cosmetic failure and a rare catastrophic failure require different thresholds.

Rejected
Recalling everything now. Continuing unchanged because the law does not yet compel a recall.
The assumption carrying it
That the downside really can be contained while the evidence improves.
Authority I keep
Safety thresholds, regulator communication, recall and public statements.
What I would delegate to AI
Accelerating trace analysis, scenario modelling and document preparation.
What would change my mind
I recall before legal compulsion if credible evidence indicates potentially serious harm, even at low incidence.

When evidence is incomplete, contain the downside while improving the decision. Do not continue unchanged and hope uncertainty resolves favourably.

Christianna Hamilton

This is one accountable judgement for the stated objective and evidence. It is not offered as the universal answer.

Act

Bounded actions, owners and approvals.

You have not selected a course of action.

Current result from authored decision rules: Contain the downside. Improve the evidence. A full recall costs €88.0m, 2.0x the liquidity available, on evidence that does not yet justify it. Continuing unchanged would ship 4,500 more units into the same uncertainty. Containment costs €2.1m and buys the evidence.

Action plan accompanying the authored starting judgement

This plan belongs to the original written judgement below. It does not represent execution of your selection or a newly generated action plan.

First action

Quarantine identifiable inventory and pause plausibly affected shipments.

Measure to watch

Credible worst-case severity, as independent testing reports.

  1. Quarantine identifiable inventory.

    OperationsHuman decision required

  2. Pause plausibly affected shipments.

    OperationsHuman decision required

  3. Trace batches, suppliers and customers.

    OperationsDraft

  4. Independently verify testing.

    RiskPrepare for approval

  5. Establish credible worst-case severity.

    RiskPrepare for approval

  6. Pre-build recall scope, funding and communications.

    FinancePrepare for approval

  7. Define the evidence threshold that triggers escalation.

    GovernanceHuman decision required

  8. Preserve the decision record.

    LegalDraft

Where calculation ends and human authority begins
  1. Calculate and surface. Evidence, assumptions, arithmetic and consequences are made visible.
  2. Investigate and challenge. AI can extend the search, prepare options and test for conflict or unsupported certainty.
  3. Decide and own. An accountable person weighs the trade-offs, makes the call and remains responsible for what follows.

Learn

What would show the judgement was wrong.

Hypothetical counterfactual evidence. Change one material condition and compare the result below. This experiment does not replace the working state in Model, Contest, Decide or Act.

Change one condition

Method and provenance

The scenario is illustrative. It does not describe a client, an engagement or a real company.

What the tools do, and what remains human

  • Calculate and surface. Arithmetic and thresholds are handled consistently so the same inputs produce the same result.
  • Investigate and challenge. Functional positions expose disagreement. The critic looks for conflict, missing evidence, double counting and unsupported certainty. A live reading, where offered, can extend that challenge but cannot alter the underlying numbers or the signed decision.
  • Decide and own. The signed judgement is Christianna Hamilton’s. It is written in advance and remains the responsibility of a person, not a system.

Evidence

Items marked given in the scenario come from the written brief. Items marked illustrative modelling assumption are quantities added so that consequences can be calculated. Evidence last reviewed 2026-09-21.

Calculations in this decision

QuantityUnitWhat it assumes
Cost of a full recallEUR millionEvery unit in the field, at the unit cost.
Full recall against liquiditymultipleNo insurance recovery is assumed.
Targeted recall, if the batch is smallEUR millionLower bound of the batch.
Targeted recall, if the batch is largeEUR millionUpper bound of the batch.
Uncertainty in scopemultipleUpper bound over lower bound.
Expected failures, low caseexpected countSmall batch, low rate.
Expected failures, high caseexpected countLarge batch, high rate.
Cost of containment while tracingEUR millionPause and testing run for the whole trace.
Containment as a share of a full recallper centNone beyond its inputs.
Units shipped if nothing changedunitsShipping continues at the current rate through the trace.