C2 · Publication Volume 12

Decision Gates

continue, modify, pause, drop or drill decisions and information gain

Learning goals

After this lesson, you should be able to define a gate before results are known, distinguish continue, modify, pause, drop and intrusive-test actions, calculate a simple gate example, incorporate information gain and non-technical constraints, and write a decision record that remains auditable after outcomes are known.

A decision gate is a controlled comparison between evidence and a predeclared rule. It is not a meeting at which a persuasive narrative wins. Good gates expose which evidence is required, what uncertainty is tolerable, which actions are available and which constraints can veto an otherwise attractive technical proposal.

Actions and maturity states

Continue repeats or extends the current strategy because its observation model remains adequate. Modify changes hypothesis, geometry, method or scale while retaining a testable opportunity. Pause preserves the target pending a resolvable dependency such as season, licence, access, data lineage or prerequisite interpretation. Drop retires the current target or concept under explicit conditions. Intrusive test advances to a higher-disturbance observation only when its incremental discrimination justifies the burden and all requirements are satisfied.

These actions are not a universal corporate workflow. They describe evidence states. A dropped geometry may leave a regional concept alive. A paused target is not low potential. A modified target receives a new version so that its success is not credited retrospectively to the original prediction.

Maturity should reflect what has been tested: regional permissivity, coherent footprint, target geometry, direct geological test or follow-up. Do not infer resource, reserve or economic maturity from exploration stage names. Formal reporting categories require applicable rules and qualified work outside this tutorial.

Gate charter and minimum evidence

A gate charter states the decision question, target version, hypotheses, actions, required inputs, quality thresholds, necessary gates, scoring or utility method, non-technical constraints, decision authority by role, dissent rule and review trigger. Freeze it before opening the result dataset whenever possible.

Minimum evidence is not “all planned data delivered.” It is the evidence needed to distinguish actions. Require adequate location, recovery, quality-control performance, coverage and detection power. If a critical input fails, the result may be indeterminate and trigger modification or pause rather than a forced technical decision.

Gate criteria should include explicit contradictions. For example: advance only if two independent evidence groups support the geometry and no adequately tested necessary condition fails. Avoid rules composed solely of positive thresholds; they can ignore a decisive negative observation.

Information gain, burden and proportionality

A decision gate combines technical evidence, information gain and constraints into explicit actions
A decision gate combines technical evidence, information gain and constraints into explicit actions

Expected information gain can be measured as change in entropy, expected utility or discrimination among hypotheses. A result can be scientifically informative even when it lowers target rank. A failed test that retires a broad class of geometry may have high portfolio value.

Compare the proposed next test with alternatives, including no action. Acquisition burden includes synthetic cost units, physical disturbance, exposure to hazards, environmental footprint, time dependence and opportunity cost. Safety, legal permission and consent are constraints, not negative utility values that strong geological scores may compensate.

Proportionality matters. Use the least intrusive method that can answer the decision question, but do not perform a low-impact survey merely because it is easy if it cannot resolve the uncertainty. When no feasible test has sufficient detection power, pause or retire rather than collect decorative data.

Stopping rules, exceptions and governance

Stopping rules can be evidence thresholds, failed necessary conditions, exhausted discriminating tests, unacceptable quality, constraint failure or dominance by another portfolio option. Include rules for repeated indeterminate results. Unlimited additional work after every failure makes the gate ineffective.

Exceptions may be justified by new evidence or a documented change in objective, not by discomfort with the outcome. Record the original rule, exception reason, authorising role, new prediction and consequence. Reviewers should be able to distinguish a legitimate model revision from criterion drift.

Separate technical recommendation from authorisation. The teaching dossier records roles, not named people. A real program must follow current governance, legal and professional requirements. Disagreement is preserved with its evidence basis; consensus is not evidence quality.

Worked synthetic example

A fictional target has three actions: stop, conduct a low-disturbance survey, or conduct an intrusive test. Two states are modelled: coherent target H with probability 0.40 and no coherent target N with probability 0.60. Teaching utilities before acquisition burden are:

| Action | H | N | |---|---:|---:| | Stop | 0 | 0 | | Low-disturbance survey then optimal action | determined by test | determined by test | | Intrusive test now | +10 | −5 |

Intrusive testing now has expected utility 0.4(10)+0.6(-5)=1.0. A low-disturbance survey has P(+\mid H)=0.80, P(+\mid N)=0.25 and burden 0.5 units. A positive result has probability 0.47 and posterior 0.32/0.47=0.681; a negative result has posterior 0.08/0.53=0.151. Intrusive testing is preferred after a positive result because 0.681(10)+0.319(-5)=5.215; stopping is preferred after a negative result.

Expected utility of the staged strategy before survey burden is 0.47(5.215)=2.451; after burden it is 1.951, exceeding 1.0. The gate therefore selects the low-disturbance survey, conditional on adequate detection and all safety, legal and environmental constraints. If permission for that survey is unresolved, the target pauses; the model must not represent lack of permission as a geological penalty.

Interpretation workflow

  1. Define target version, decision question, actions and maturity state.
  2. Freeze hypotheses, minimum evidence and quality thresholds.
  3. Apply necessary geological and non-technical gates.
  4. Check whether the result is adequate, negative or indeterminate.
  5. Compare action consequences under each remaining hypothesis.
  6. Estimate whether another test could change the preferred action.
  7. Include burden and seek the least intrusive adequate test.
  8. Apply continue, modify, pause, drop or intrusive-test rule.
  9. Record dissent, residual uncertainty and exception handling.
  10. Preserve the decision-time evidence snapshot for later review.

Practice and review

  1. Recalculate the synthetic staged strategy if survey burden becomes 1.6 units.
  2. Write a gate in which a coordinate-control failure produces “pause,” not “drop.”
  3. Give an example of a negative result with high portfolio information value.
  4. Explain why safety and legal permission should not be offset by a high geological score.
  5. Draft an exception record for a revised geometry after an adequate negative test.

Sources and further reading