How Information Works | Decision-Relevance Threshold — When Should You Stop Gathering More Information?

You could always collect one more piece of information.

The harder question is whether it would change what you should do.

A decision-relevance threshold is the point beyond which additional information is unlikely to change the chosen action enough to justify its cost, delay or complexity.

This is the fourth pillar beneath How Information Works. The master owns information value broadly. This page owns the stopping-rule question: when is enough information actually enough?

Quick Read

Information has value when it can change a decision, improve timing, reduce risk or alter the confidence with which an action is taken. Once plausible new observations would not change the route, continued gathering may add volume without adding decision value. Strong stopping rules compare the expected value of another observation with its cost: time, money, attention and the consequence of delay. High-stakes or irreversible decisions justify more information. Fast-moving situations may justify less because information decays while you wait. The goal is not certainty. It is sufficient discrimination among the actions that are actually available.

decision options → key uncertainty → candidate observation → would result change action? → compare expected benefit with cost/delay → gather or stop → act → observe world return

More Data Can Leave the Decision Unchanged

You are choosing between two revision topics.

Topic A has failed repeatedly under retrieval.

Topic B is already strong.

Another hour of collecting old marks may not change the obvious next action.

The Best Next Question Separates Actions

Useful information is discriminating.

Ask:

If this answer were A, what would I do? If it were B, would I do something different?

If both answers lead to the same action, the question may have low immediate decision value.

Stopping Rules Depend on Consequence

Low-stakes reversible action:

act with less information and learn from feedback.

High-stakes irreversible action:

raise the evidence threshold before acting.

How Risk Works owns consequence and exposure broadly.

Delay Is Part of the Cost

Waiting for perfect information can make the decision worse.

• deadline passes;
• opportunity disappears;
• world state changes;
• others act first;
• stale data accumulate.

The first sibling, Information Half-Life, owns freshness and decay.

Missing Information Does Not Always Block Action

You may know that one field is missing.

If every plausible value still leads to the same safe action, the missingness may not need immediate resolution.

The second sibling, Missing Information, owns the meaning of absence.

Provenance Can Raise or Lower the Stopping Threshold

One strong primary record may be enough.

Ten copied secondary claims may still leave the core uncertainty unresolved.

The third sibling, Provenance Breakpoints, owns source lineage.

Confidence and Action Thresholds Are Different

You may remain only 70% confident that explanation A is correct.

If both A and the main alternative recommend the same low-cost next step, action can still be justified.

Decision readiness does not require complete belief certainty.

Diminishing Returns Appear Quickly in Some Searches

The first three credible sources establish the same stable fact.

The next twenty repeat it.

Marginal information value approaches zero.

But One Disconfirming Observation Can Be Worth More Than Many Confirmations

If a model predicts every case should behave one way, one credible exception can reveal a missing condition.

Stopping rules should therefore seek discriminating evidence, not merely a target number of sources.

Decision-Making Remains the Downstream Owner

How Decision-Making Works owns choice under uncertainty broadly.

Decision-Relevance Threshold stays upstream: it asks when information gathering has done enough work to hand the problem to the decision process.

A Practical Stop Rule

DECISION: Which topic to revise tonight?
OPTIONS: Algebra / Geometry
CURRENT EVIDENCE: Algebra fails 4/5 retrieval checks; Geometry 1/5
KEY UNCERTAINTY: Was algebra failure caused only by tiredness?
NEXT OBSERVATION: One fresh 10-minute algebra retrieval test
ACTION SENSITIVITY:
- fails again → revise algebra
- passes strongly → split time
STOP RULE: after fresh retrieval test; do not collect older marks unless result is ambiguous

A 20-Lens Decision-Relevance Audit

1. What decision is actually being made?
2. What options exist?
3. Which uncertainty separates the options?
4. What information already exists?
5. What candidate observation comes next?
6. What would result A change?
7. What would result B change?
8. If neither changes action, why gather it?
9. What is the cost of collection?
10. What is the delay cost?
11. How fast does current information decay?
12. How reversible is the action?
13. What are the stakes?
14. Is another source independent?
15. Are we gathering confirmations instead of discriminators?
16. What missing information is genuinely decision-critical?
17. What uncertainty can remain?
18. What confidence threshold is actually needed?
19. What event would reopen information gathering later?
20. Did the stopping rule allow timely action without hiding a material unresolved uncertainty?

For Primary Readers

If you already know which book you need for tomorrow, checking the shelf ten more times will not help. Good information gathering stops when the next check will not change what you should do.

For Secondary Readers

Take one decision and write two possible answers to your next question. If both answers lead to the same action, find a more discriminating question.

For Advanced Readers

Model stopping as expected value of information. Continue sampling while the expected improvement in decision quality exceeds acquisition and delay cost, with higher thresholds where actions are irreversible or loss functions are asymmetric.

Final Thought: Information Is Finished When the Next Bit Is Worth Less Than Acting

The disciplined information system does not ask for everything it could know. It asks for the next distinction that could change the route—and stops when the world is better served by action than by another pile of data.