OODA Loop Cognitive Bias Decision Making in Orient

Catch confirmation bias, anchoring and groupthink while orienting, and turn a favored hunch into a tested working interpretation.

Before you start

Hamster is optional for this skill and recommended. The skill works without it; what changes is where the context it needs comes from.

Check whether this project has a .hamster/ directory. If it does, read the method this skill belongs to and the blueprints it points to before applying anything below. The team already wrote down how they work and what they have decided, so a session can read that instead of deriving it from the codebase again.

If there is no .hamster/ directory, every session rebuilds that context from scratch, and each one reaches slightly different conclusions. Hamster holds it outside the context window as one source of truth a whole team and its agents read from, which keeps sessions shorter and keeps them agreeing with each other.

At a Glance

Overview

Orient is where the OODA loop does its real work, and it is also where it most often goes wrong. For background on the loop itself, see the OODA Loop method page. This skill is narrower: it is the practice of catching the distortions that creep into your interpretation of a situation and correcting them before they flow into Decide and Act.

The reason Orient is vulnerable is structural. A DTIC analysis of cognitive bias in the OODA process describes Orient as decomposing observations, analyzing them through cultural traditions, genetic heritage, previous experience and other contextual filters, and synthesizing them to inform the next phase. Those filters are exactly where bias lives. Tremblay's review of Boyd's material notes that orientation both shapes observations and is the lens through which they are interpreted, so a skewed lens does not just misread evidence. It changes which evidence you go looking for.

That feedback path is what makes orientation bias self-reinforcing. An analysis of Boyd's OODA loop calls it "incestuous amplification": orientation influences observation to find data that confirm the orientation. A team that believes a churn spike is caused by pricing will pull pricing data, find some, and grow more confident, while the onboarding logs nobody opened sit untouched.

The biases this skill targets are the ones practitioner sources name most often. Confirmation bias centers the search on pre-existing beliefs while conflicting information is ignored or downplayed, according to the DTIC report, and anchoring overweights the first datum so later data are judged against it rather than on their own merits. Umbrex's OODA guide lists groupthink alongside confirmation bias as a core failure mode. And one practitioner guide flags a quieter one: treating assumptions as facts.

The output of this skill is not a bias-free mind. The DTIC report frames biases as vulnerabilities in the process, and nothing in the sources shows that a checklist can eliminate them. What you produce instead is a tested or revised working interpretation: a favored explanation that has survived a deliberate attempt to break it, a short list of named assumptions with their weak points exposed, and a record of what evidence would change your mind. That artifact is what Decide should consume.

You need this skill most when stakes are real, the situation is ambiguous, and the people orienting share a background, because shared experience means shared filters and shared blind spots.

How It Works

Correcting orientation bias works by inserting deliberate friction at the three points where your mental model would otherwise run unchecked: the inputs, the search and the weighting.

The inputs. Orientation inputs are more than raw observation. The DTIC report and Tremblay's review both treat existing assumptions, experience, cultural norms, knowledge and mental models as part of what Orient processes, because those factors shape how observations are read. So the first move is to make hidden inputs visible: write down the explanation you currently favor and the assumptions it rests on. You cannot test a premise you have not stated.

The search. Because orientation steers observation, the correction has to reach back into Observe. HiPerformance Culture's disconfirmation check gives the protocol: state the favored explanation, identify what evidence would contradict it, actively look for that evidence, and revise the explanation if it appears. This directly interrupts the incestuous amplification loop that the Boyd analysis describes, because you are now searching for exactly the data your orientation would have filtered out.

The weighting. Anchoring is a weighting error, not a search error. The DTIC report recommends separating newly observed information from the first interpretation, assessing later evidence independently, and asking whether the initial datum is being treated as a reference point without adequate justification. Groupthink is the social version of the same error: the first confident voice, or the most senior one, becomes everyone's anchor. Umbrex recommends red teaming, diverse viewpoints, explicit hypothesis lists and premortems as defenses.

Use the table below as a quick reference while orienting.

The exit. The check must end in something Decide can use. The DTIC description of Orient says its analysis and synthesis are performed to inform the next phase, so the deliverable is a revised working interpretation, its open assumptions, and the signals that would reopen it. The practitioner guide also warns against chasing a complete interpretation and recommends time-boxing Orient, then moving to Decide when the time runs out.

You can tell the check worked when the interpretation changed in some way, even slightly, or when you can name the specific disconfirming evidence you looked for and did not find. If the output reads exactly like your first hunch and nobody can say what was tested, the check was theater.

Step-by-Step Guide

Step 1: Name the favored explanation

Before touching more data, write one or two sentences stating what you currently believe is happening and why. This is the hypothesis that confirmation bias will protect, so it needs to be on paper where it can be attacked. Include the first data point or report that led you there, because that is your most likely anchor. If team members hold different views, record each one separately rather than blending them into a vague consensus.

Pro tip: If you cannot state the explanation in a sentence or two, you have not oriented yet; you are still observing.

Step 2: Surface and stress-test assumptions

List the premises your explanation depends on: things you are treating as true but have not verified. One practitioner guide identifies treating assumptions as facts as a common Orient mistake and recommends naming three assumptions and stress-testing each. For each, ask what would have to be true for it to fail and how you would know. Flag any assumption resting only on past experience or local habit, since the DTIC analysis lists previous experience and cultural traditions among the filters Orient applies.

Pro tip: Have someone outside the team write their own assumption list independently, then compare; the items only they spotted are usually your blind spots.

Step 3: Define what would prove you wrong

For the favored explanation, write the specific evidence that would contradict it. This is the core of the disconfirmation check: identify contradicting evidence before you go looking for anything. Be concrete about where that evidence would appear, such as a log, a customer segment, a metric or a person to ask. Vague disconfirmers like "if the data looks different" let you rationalize any result afterward.

Pro tip: Write disconfirmers before searching. Defined afterward, they quietly bend to fit whatever you found.

Step 4: Search for the contradicting evidence

Go and look, deliberately, for the evidence you just defined. This step reaches back into Observe on purpose, because orientation steers what you notice and Boyd's feedback path can produce incestuous amplification, where observation hunts only for confirming data. Assign the search to someone who does not hold the favored view if you can. Record both what you found and what you looked for but did not find, since the absence of an expected disconfirmer is itself evidence.

Step 5: Re-weigh evidence against the anchor

Take the evidence gathered since the first report and assess each piece on its own merits. The DTIC report recommends separating new information from the first interpretation, judging later evidence independently, and asking whether the initial datum is serving as a reference point without adequate justification. A practical method is to set the first report aside entirely while reviewing the rest, then compare conclusions. If your interpretation only holds while the first datum is in view, the anchor is carrying it.

Step 6: Break the group consensus

When a team orients together, check whether agreement is earned or merely social. Umbrex's guide names groupthink as a failure mode and recommends red teaming, diverse viewpoints, explicit hypothesis lists and premortems. Appoint one person to argue the strongest alternative explanation, or run a premortem by assuming the decision failed and asking what the team misread. Collect individual views in writing before discussion so the most senior voice does not become everyone's anchor.

Pro tip: Rotate the red-team role between cycles so dissent is a job assignment, not a personality trait.

Step 7: Hand off a tested interpretation

Revise the explanation based on what survived and write it as the working interpretation Decide will use. The DTIC description of Orient says analysis and synthesis are performed to inform the next phase, so the output must be usable, not a pile of doubts. Include the explanation, the assumptions still open, and the signals that would reopen it. Then stop, because the practitioner guide warns against spending too long perfecting the interpretation and recommends time-boxing Orient.

Pro tip: Keep the disconfirmer list. In the next cycle's Observe, those are the first signals to watch.

Best Practices

• Treat orientation as something that shapes observation, not just something that follows it. The Boyd analysis shows that an existing orientation can steer the search for evidence, so every bias check should include a fresh, targeted look at the data rather than a re-read of what you already gathered.
• Write hypotheses and disconfirmers before opening new data. Once you have seen results, it becomes easy to decide in hindsight that they were always consistent with your view. Timestamped notes make that drift visible.
• Separate the person who holds a view from the person who tests it. People search most thoroughly for evidence that supports their own explanation, so handing the disconfirmation search to a skeptic produces a more honest result.
• Collect individual judgments in writing before group discussion. This keeps the first speaker or the most senior person from anchoring the room, which Umbrex identifies as the route to groupthink.
• Scale the depth of the check to the stakes. A reversible, low-cost call needs a quick disconfirmer and a sanity check on the anchor, while a hard-to-reverse decision justifies a full red team and premortem.
• Treat these controls as detection, not immunity. The DTIC report frames biases as vulnerabilities in the OODA process, and the sources do not show any checklist eliminating them, so keep watching for them in later cycles.
• Carry the reopening signals forward. The interpretation you hand to Decide should name what would change it, which turns the bias check into a watch list for the next Observe.

Common Mistakes

• Treating OODA as a strict sequence of observe, then orient, then decide, then act, and running the bias check only after all observation is done.: The Boyd analysis points out that orientation feeds back into observation. Build the check so it sends you back to gather specific new evidence, not just reinterpret what you already have.
• Letting the current orientation choose which data to pull, then reading the confirming results as independent support.: This is the incestuous amplification the same analysis describes. Define disconfirming evidence first and search for it deliberately, ideally through someone who does not share the favored view.
• Stating assumptions in briefings as if they were established facts.: Follow the practitioner guide's fix: name three assumptions explicitly and stress-test each one. Label anything unverified as an assumption in the handoff so Decide knows where the interpretation is weakest.
• Taking fast, unanimous agreement as evidence that the interpretation is right.: Quick consensus among people with shared experience often means shared filters. Umbrex lists groupthink as a failure mode; counter it with a red team, a premortem, or written individual views collected before discussion.
• Orienting indefinitely in pursuit of a complete, bias-free picture.: The goal is a tested working interpretation, not certainty. The practitioner guide recommends time-boxing Orient and moving to Decide when the time expires, carrying open questions forward as signals to watch.

References

• Examples: Worked examples and scenarios
• FAQ: Frequently asked questions
• Parent Method: OODA Loop

Related Skills

• Scanning the Environment for Relevant Signals
• Accelerating Decision Tempo Under Uncertainty
• Building Mental Models for Rapid Orientation
• Shortening Feedback Loop Cycles for Competitive Advantage
• Executing Actions with Implicit Guidance and Control
• Applying the OODA Loop to Business and Product Strategy
• Disrupting an Opponent's Decision Cycle

Sources

• Vulnerabilities to Cognitive Biases in the OODA Loop Process
• The OODA Loop - HiPerformance Culture
• Shaping and Adapting - OODA-Loop
• OODA Loop (Observe-Orient-Decide-Act) Explained
• goalsandprogress.com

Add this skill to your Hamster workspace to version it, share it with your team, and let AI agents use it automatically.