Skip to content
Reason Commons

For groups trying to make something better

Change things together.
Learn what actually worked.

Reason Commons helps groups agree on what to change, what to change to, and how to make the change happen — then learn from what reality actually does.

Try the demo Early access

No payment step. Just create an account and explore.

Five questions a group works through, each one writing a line into a single shared record, and a return from the last question to the first.On the left: what to change, what to change to, how will we cause it, what happened, and what did we learn — with an arrow returning from the last back to the first. On the right, one bordered record gains a row from each question in turn: claim, objective, decision, observation, standing. The loop is drawn without a frame; the record is the only enclosed object.ClaimObjectiveDecisionObservationStanding1What to change?2What to change to?3How will we cause it?4What happened?5What did we learn?one evolving recordand round again
01 / The job

Improve together.
Without losing the reasoning.

Groups do not only need better answers. They need a way to decide what matters, act on what they believe, discover when reality disagrees, and continue from what they learned.

See how it works

What to change?

“We keep missing release dates.”

The group establishes what in the current situation actually needs changing.

What to change to?

“Ownership should be explicit before work begins.”

The group agrees on a better condition and why it should improve the situation.

How to cause the change?

“Run a four-week named-owner experiment.”

Actions are connected to the reasoning that justified them and the effects expected.

What did reality say?

Expected: median delay < 48hObserved: 61h

The result changes what the group stands behind next.

02 / The commons

The reasoning survives the meeting.
And the model.

Every consequential claim keeps what it rests on, who stood behind it, what remained contested, what was expected, and what reality later did about it.

Explore the example
One evolving record
Current claim
Unclear ownership is why reviews stall.
Grounds
Three recent reviews waited for an unnamed owner.
Dissent
Maya thinks capacity may be the real bottleneck.
Decision
Run a four-week named-owner pilot.
Expectation
Median review delay falls below 48 hours.
Observation
61 hours.
Current standing
The ownership explanation is challenged.
  1. Proposed
  2. Discussed
  3. Accepted
  4. Tested
  5. Revision needed

The institution keeps the why, not just the what.

03 / How it works

Four movements.
Five supporting properties.

A simple cycle for real work, and underneath it the five things a shared record has to do so the reasoning stays cumulative, governed and portable.

Read the distinctions
The improvement cycle
  1. 01

    Orient

    What needs to change?

  2. 02

    Choose

    What should it change to?

  3. 03

    Act

    How will we cause the change?

  4. 04

    Learn

    What did reality teach us?

What the record has to do

  • Capture

    Ordinary work can enter the record.

  • Judge

    People decide what the group stands behind.

  • Remember

    Reasons, dissent, expectations and outcomes stay connected.

  • Correct

    Reality can challenge the reasoning.

  • Transport

    The record belongs to the community, not the AI or vendor.

The four are what the group is trying to accomplish. The five are what Reason Commons has to do well to support it.

04 / Human judgment

AI can help reason.
It cannot decide what “we” believe.

AI may find evidence, identify contradictions, propose causal links, suggest alternatives, and prepare changes to the shared record. People decide what the group accepts, what it will rely on, and what it will do.

Our approach to AI

AI prepares

  • Possible explanation
  • Possible consequence
  • Possible experiment

Human judgment

  • Accept
  • Revise
  • Reserve
  • Reject
  • Rely on
05 / When the question gets harder

Start simple.
Go deeper when the reasoning demands it.

Most work never needs more than the four questions. When one of them turns out to be genuinely contested, the reasoning behind it can take a stronger shape — and a novice never has to know these are here.

Read the method explainers

  1. What does better require?

    Goal Tree

  2. Why is this happening?

    Current Reality Tree

  3. What conflict keeps us stuck?

    Evaporating Cloud

  4. Will this change actually produce the future we want?

    Future Reality Tree

  5. What has to become true first?

    Prerequisite Tree

  6. What exactly will we do, and what should each step cause?

    Transition Tree

You do not begin by choosing a diagram. The structure grows out of the question the group is trying to answer.

06 / Reality gets a vote

Finishing the work
is not the same as being right.

Expectations are recorded before the answer is knowable. Automated tests can settle some. Actual use, measures, and observation settle others. When reality disagrees, Reason Commons shows which reasoning and decisions were standing on the failed expectation.

See a claim get challenged
We expected< 48hmedian review delaysealed before the pilot ran
We observed61hafter the four-week pilotrecorded against that expectation

The pilot finished. It was also wrong — and the record holds both, so “we did the work” never becomes evidence for the reasoning that asked for it.

Reality gets a vote—and the consequences propagate.

07 / MemoryIllustrative record · one loop, carried forward

Memory that answers
why.

Ordinary organizational memory can tell the next person what happened. Reasoning memory tells them why the group believed the action would work, what was disputed, and what reality later established — so nobody has to reconstruct the argument. Psychology already names three memories, and every tool you use keeps one of them: facts, routines, and a history of events.

Reasoning memory is the fourth, and it sits under the other three. It is the one nobody stores, and the only one that cannot be recovered once it is gone.

  1. what

    Semantic memory

    Facts, definitions, the state of the world.

    “Median review delay is 61 hours.”

    kept today by

    Documents, wikis, vector search

  2. how

    Procedural memory

    Skills and routines — how a thing is done here.

    “This is how we run a review.”

    kept today by

    Playbooks, runbooks, habit

  3. when

    Episodic memory

    What happened, in order, to whom.

    “On 23 March we started a four-week ownership pilot.”

    kept today by

    Logs, transcripts, chat history

  4. why

    Reasoning memory

    Claims, what they stand on, what was contested, what was predicted, and what reality did about it.

    “We chose the pilot because unnamed ownership was judged the root cause — Maya reserved on that, and a 61-hour median later challenged it.”

    kept today by

    Almost nothing. Whoever comes next derives it again.

The why contains the rest. A claim carries the facts it stands on, the action it licensed, and the version it was accepted at — so the what, the how and the when can be read back off it.

It does not run back. No amount of documents, playbooks and transcripts reconstructs the judgment that chose between them. That is why this is the layer worth making portable — and the layer nothing currently keeps.

one reasoning model · eight accepted versions

Reality disagreed — in three trees at once

One observation, median 61 hours, challenges the expected effect, the change that promised it, and the root cause both stood on. Maya’s alternative root cause is untouched — the dispute is aimed at a proposition, not sprayed over a tree — and nothing is deleted.

accepted
21 Apr 2026
state hash
413d75f16173…
chained over
7e45ee48db7a…
this digest
e0d1697b98d9…
  1. GoalGoal Tree
    necessary condition — Every review has an owner the day it opensnecessary condition — Reviewers have the capacity to take one onnecessary condition — The review standard is written downwhat success requires — Work never waits on an unnamed personwhat success requires — Review quality does not rest on one reviewergoal — Delivery is predictable
    • ProblemCurrent Reality Tree
      root cause — A review opens with nobody named on itroot cause — The booking system releases reviews in batchesintermediate cause — Reviews wait to be picked upunwanted effect — Reviews stall for daysunwanted effect — Delivery dates slip without warningobservation — Four weeks of the pilot: median 61 hours
      • Challenged · root cause · A review opens with nobody named on it
      • Added · observation · Four weeks of the pilot: median 61 hours
    • ConflictEvaporating Cloud
      shared objective — Ship reliablyneed — Protect qualityneed — Keep work movingwhat we do about it — Review every change thoroughlywhat we do about it — Clear reviews within a day
      • SolutionFuture Reality Tree
        change we would make — Name an owner when the review openswanted effect — Nothing waits to be picked upwanted effect — Delivery dates holdunwanted effect — One owner becomes a queue of one
        • Challenged · change we would make · Name an owner when the review opens
      • ImplementationPrerequisite Tree
        implementation objective — Named ownership is how reviews startmilestone — Capacity is visible on the boardobstacle — Nobody can see who is freemilestone — Reassignment is a recorded changeobstacle — Owners get named, then quietly reassigned
        • ActionsTransition Tree
          existing reality — Reviews open unassignedunmet need — Someone must own it on day oneaction — Assign an owner at open, for four weeksexpected effect — Median review delay falls below 48 hoursunmet need — The team must see the trend before the pilot endsaction — Publish the running median each Fridayexpected effect — A slipping pilot is caught in week two
          • Challenged · expected effect · Median review delay falls below 48 hours
        arrived in this versionchallenged by an observationhue says what kind of idearelationships run downward · a conflict runs both ways
        Structure only — every rectangle is an accepted idea and every line a relationship the team stands behind. Hover one to read it. Nothing is deleted when reality disagrees; it is marked.
        08 / Open work

        The project should answer
        to its own reasons.

        We are using Reason Commons to develop Reason Commons. Proposed capabilities have to earn their way in through the same cycle: identify a problem, predict an improvement, try the change, and learn from the outcome.

        Inspect the research program
        Current question
        Does preserving the reasoning behind software changes improve the quality of the next development decision?
        What we expect
        A successor can continue without reconstructing why the current design exists.
        Status
        Evidence still being gathered. An early model-only experiment found that stored reasoning maps did not improve safe continuation. Read the negative result

        Built by David Joseph, in collaboration with Rufus Pollock on reasoning, commons, and collective learning.

        09 / Early access

        Help build a better way
        to change things together.

        With AI in the reasoning, and people in charge. For teams who want decisions, actions, expectations and learning to remain connected over time.

        Sign up for early access Try the demo

        Create an account with your name, email, and password to enter the prototype. No payment step.