Principles Library / Evidence over assumption

Evidence over assumption

Evidence-Based Decisions Reduce Assumption Risk

Principle intent

Base important decisions on observed evidence rather than untested assumptions, prediction, opinion, or authority.

Read the full intent

Assumptions are unavoidable, but risk grows when they go untested, when evidence cannot reach decisions before they are locked in, or when a definition of value is allowed to outlive the conditions that created it. Change is not a planning failure. It is evidence that learning has occurred.

When it holds: what it does for you

Assumptions get tested

Wrong assumptions show up before they're locked into decisions.

Correction stays routine

Plans adjust as evidence arrives, so change doesn't turn into a disruption.

Reviews change things

Each learning cycle ends in a real adjustment, so teams stay engaged in it.

When it is ignored: from early signs to a lasting condition

Early warning signs · Visible now, cheap to fix

What happens over time · Accumulates, costly to reverse

The condition it becomes

Strategic Volatility · Primary

Every quarter starts over. Nothing compounds.

How this principle feeds it

correction arrives as abrupt reprioritization in place of deliberate change.

Customer Disconnect · Primary

Delivery is consistent. Value is not landing.

How this principle feeds it

untested beliefs about customer value replace current evidence of it.

Attribution Failure · Secondary

We know what happened. Nobody can explain why.

How this principle feeds it

a confident narrative stands in for reasoning that can be traced to evidence.

Quality Fragility · Secondary

Customers find defects before the team does.

How this principle feeds it

reviews that change no decision let the same problems repeat, which shows up partly as recurring defects.

Intent Drift · Primary

The team built exactly what was asked. It is no longer what is needed.

How this principle feeds it

governing intent stays fixed after it stops being valid.

Batch Amplification · Primary

Problems are found at release, not before it.

How this principle feeds it

work accumulates before evidence validates it, so corrections expose everything at once.

Local Optimization Bias · Secondary

Every team hits its targets. The product still misses.

How this principle feeds it

teams keep optimizing measures whose link to outcomes went unchecked, which can fragment end-to-end value.

Where is your system on this path?

Diagnose

Diagnose

AI signs are on by default. Use the AI switch to exclude them.

1. Which of these early warning signs are you seeing?

  • Decisions lean on plans, opinion or seniority more than on evidence.
  • Data that contradicts the current view gets explained away.
  • Learning only happens at milestones or after delivery.
  • Reviews get skipped or rushed when delivery pressure builds.
  • Some decisions get reopened without new evidence while others go stale untouched.
  • Backlogs and roadmaps are treated as fixed, and any change needs special justification.
  • Metrics get watched but rarely change a decision.
  • Features ship without meaningful impact.
  • AI predictions count as evidence, and outputs that contradict reality get discounted. (AI)
  • Outcome feedback gets collected but never reaches agent configuration or policy. (AI)
  • No human checkpoint exists to inspect whether agent outputs are moving toward the right outcome. (AI)
  • Agents keep executing priorities and proxy metrics whose value has quietly expired. (AI)

2. Over the last few months, have you noticed…

For each consequence: Not seen / Sometimes / Persistently / Not sure.

  • Wrong assumptions persist, and risk builds quietly behind confident narratives.
  • Reviews never change a decision, so teams disengage and the same problems repeat.
  • Decisions sit further and further from current conditions.
  • Correction stops being routine and arrives as disruption.
  • Prediction has replaced observation, and errors scale faster than learning catches up. (AI)
  • Teams keep chasing measures whose link to outcomes nobody has rechecked.

3. The condition it becomes

Based on your answers, recognize at most two conditions: No / Partly / Yes, that is us.

Strategic Volatility: Every quarter starts over. Nothing compounds.

Customer Disconnect: Delivery is consistent. Value is not landing.

Attribution Failure: We know what happened. Nobody can explain why.

Quality Fragility: Customers find defects before the team does.

Intent Drift: The team built exactly what was asked. It is no longer what is needed.

Batch Amplification: Problems are found at release, not before it.

Local Optimization Bias: Every team hits its targets. The product still misses.

Take your Diagnosis to the next level by invoking AI Diagnostic Coaching tool.

What else have you observed? Optional. Shared only if you choose to start coaching.

Start a Diagnosis session: review and edit a draft containing your note, conditions you recognized in Step 3 (including whether they match partly), warning signs and consequence ratings before sending it. No scores or principle files are sent.

See what practices will help you

Act

Act

Showing all 6 practices. Answer the Diagnose questions and the most relevant come first.

Decide where evidence really matters

Decide which decisions genuinely need empirical evidence, and write down the assumptions underneath them.

Start by asking your team: What are we assuming, and what evidence would change our mind?

Avoid · You’ll know it’s working when

Avoid

  • Mistaking the presence of data for evidence-based decisions
  • Waiting for perfect information before acting

You’ll know it’s working when

Decisions rest on evidence, and data that contradicts the current view actually gets weighed.

Get feedback in before decisions lock

Build feedback loops that reach decisions before they're locked in. Watch for drift or regression as execution speeds up.

Start by asking your team: Which decisions need new evidence before we revisit them, and which have gone too long without it?

Avoid · You’ll know it’s working when

Avoid

  • Gathering feedback without closing the loop

You’ll know it’s working when

Learning happens during delivery, while decisions can still change.

Protect a learning cadence

Hold a fixed learning cadence that delivery pressure can't cancel. Each cycle should produce a real adjustment, and a status update doesn't count.

Start by asking your team: Where is learning protected in our cadence, and what happens if that protection goes?

Avoid · You’ll know it’s working when

Avoid

  • Treating timeboxes as deadlines instead of protected learning time
  • Cancelling cycles to get more done

You’ll know it’s working when

Reviews happen on schedule even under pressure, and each one changes something.

Judge decisions by what actually happened

Judge each decision by the outcomes you observed, whatever the original intent was.

Start by asking your team: What are we observing right now, and what are we only predicting?

Avoid · You’ll know it’s working when

Avoid

  • Selecting metrics that confirm an existing belief

You’ll know it’s working when

Metrics change decisions, and shipped features show real impact.

Treat the roadmap as a hypothesis

Treat plans, backlogs and roadmaps as hypotheses. Revalidate them as conditions change, and deprioritize work whose expected value no longer holds.

Start by asking your team: Would we still choose this priority if we were deciding today?

Avoid · You’ll know it’s working when

Avoid

  • Treating roadmap priorities as permanent commitments
  • Reframing indecision as adaptability

You’ll know it’s working when

Roadmaps change when conditions do, without a fight over justification.

Check AI against real outcomes

Set a human checkpoint that compares AI predictions and agent behavior with real outcomes. Feed what it finds into agent configuration and policy.

Start by asking your team: As AI executes faster, are our feedback and evaluation loops keeping pace?

Avoid · You’ll know it’s working when

Avoid

  • Assuming AI analysis removes the need for empirical validation
  • Assuming AI speed reduces the need for learning cycles

You’ll know it’s working when

People check AI outputs against what they observe, and outcome feedback reaches agent configuration.