Lake Oswego, Oregon

A practical school for learning how to use AI.

Loading the accepted plan.

What we have so far

This is the working plan as it stands now. Saved answers are decisions we have accepted. Open questions are the next pieces to work through. Needs review means a newer answer changed something nearby.

Financial model

The P&L calculates when all required variables have evidence-backed values.

Next question

Answer one thing, then review the update.

This page walks top to bottom through the current interview step. Technical graph details stay available, but the main job is to answer the question in front of you.

Business graph

See what depends on what.

Browse by business domain, then inspect claims, assumptions, tests, evidence, and downstream effects.

Validation work

Tests and evidence.

Define the threshold before collecting evidence. Evidence can support, refute, or remain neutral.

Validation backlog

0

Define a test

Add evidence

Plan node

Related test

Source type

Direction

Source label

Citation URL

Public evidence summary

Quality, 1-5

Independence, 1-5

Numeric value

Unit

Visibility

Observed date

Private notes

Evidence register

Public summaries remain visible; private notes require builder access.

Audit trail

What changed and why.

Graph decisions lead. Legacy Loopus runs remain available as provenance.

Graph decisions

Legacy planning runs

How to use AI School

Turn an idea into a business plan you can defend.

This workbench keeps the whole business in one connected model. It repeatedly finds the most important uncertainty, asks for evidence, records a decision, and updates everything that depends on it.

Start here

The human and the engine have different jobs.

The engine organizes

It stores connected claims, ranks unresolved assumptions, calculates blast radius, and propagates approved decisions into the plan and P&L.

The human judges

You decide whether a test is fair, whether evidence is credible, and whether the current claim should be accepted, revised, split, or killed.

The graph and loop

The entire process in one picture

Boxes are graph nodes. Labeled arrows are relationships. The return arrow is the repeated hardening loop.

Current plan Connected business claims

depends on

Highest-risk assumption The next claim that could break the plan

verified by

Test Method + success threshold

Evidence Sources, results, and numeric facts

supports a decision

Accept Revise Split Kill

propagates into

PlanUpdated claims

P&LUpdated numbers

BacklogUpdated work

Repeat with the next highest-risk assumption

First cycle

What to do, in order

One complete cycle should change the graph, reduce uncertainty, or make the next piece of work more specific.

1
Read the current plan
Open the business domains and understand what is already claimed. Notice which items are hypotheses, supported, adopted, stale, or blocked.
2
Review the next uncertainty
Confirm that the selected assumption is consequential. Read its risk factors and the claims that would be affected by changing it.
3
Define the test before the result
Write a method and a success threshold that could genuinely change the decision. Avoid tests that can only confirm what you already believe.
4
Record evidence honestly
Add the source, result, direction, quality, and independence. Neutral or refuting evidence is useful; do not turn context into proof.
5
Make one explicit decision
Accept, revise, split, or kill the assumption. Read the impact preview before applying it so downstream changes are intentional.
6
Inspect what changed
Check the decision log, stale claims, financial variables, and newly selected risk. The audit trail should explain the new graph version.

Workbench map

Where each tab fits

Plan

Read the current business model and financial projection.

Next

Work the single highest-value uncertainty and make a decision.

Graph

Trace dependencies, hardness, and validation history.

Tests

Define thresholds and record evidence or numeric facts.

Activity

Audit decisions, graph versions, runs, prompts, and feedback.

Guide

Learn the workflow and look up the terms used throughout the workbench.

Method

Inspect the technical engine, ranking factors, and planner prompt.

Glossary

Terms used in the workbench

Showing all terms.

Filter glossary

Graph: The connected source of truth for the business. It stores claims and the relationships between them.
Node: One item in the graph, such as a claim, assumption, test, evidence record, action, constraint, or financial variable.
Edge: A typed relationship between two nodes, such as depends on, supports, refutes, constrains, or decomposes into.
Claim: A statement the current business plan presents as part of its model. A claim may still be weak or stale.
Assumption: A claim that must be true for part of the plan to work but has not yet been sufficiently verified.
Dependency: A relationship where one claim cannot be trusted unless another node is sufficiently reliable.
Hardness: The current evidence state of a node. Hardness is separate from how detailed or deeply decomposed the node is.
Hypothesis: An active, testable statement that is not yet supported strongly enough to use as a settled fact.
Testing: A node with a planned or running validation test.
Supported: A node with enough weighted supporting evidence to increase confidence, but not necessarily a final human decision.
Adopted: A node a human has explicitly accepted into the working business plan.
Stale: A node that must be reviewed again because something it depends on changed.
Contradicted: A node where weighted refuting evidence is stronger than supporting evidence.
Blocked: A node that cannot proceed because a required dependency was killed or otherwise invalidated.
Superseded: A node replaced by a newer statement or by multiple child nodes after a split.
Killed: A node intentionally removed from the active plan while remaining in history for auditability.
Test: A predefined method for collecting information that could change the status or wording of an assumption.
Success threshold: The result required for a test to count as successful. It is written before evidence is gathered.
Evidence: A cited observation, interview, experiment, sale, quote, or public fact attached to a graph node.
Evidence direction: Whether evidence supports a claim, refutes it, or remains neutral.
Quality: A 1–5 assessment of how directly and reliably an evidence record measures the claim.
Independence: A 1–5 assessment of how separate the evidence source is from the people or incentives behind the claim.
Risk score: An explainable priority based on impact, uncertainty, urgency, dependency reach, reversibility, and test cost.
Blast radius: The downstream nodes likely to change or become stale if the selected assumption changes.
Accept: Adopt the current statement. This requires evidence or a recorded override reason.
Revise: Replace the current statement with a more accurate or more testable version.
Split: Replace one statement with two or more narrower statements that can be tested independently.
Kill: Remove the statement from the active plan and block claims that require it.
Impact preview: A review of expected graph, plan, backlog, and financial changes before a decision is applied.
Propagation: The automatic process that marks downstream claims stale or blocked after an upstream decision.
Projection: A human-readable view generated from the graph, such as the Plan, validation backlog, evidence register, or P&L.
P&L: The profit-and-loss projection calculated from evidence-backed financial variables.
Graph version: The numbered state of the graph after an applied decision. A decision increments the version.
Provenance: The record of where a node or decision came from, including artifacts, prompts, evidence, and prior versions.
gatedThumbs: The thumbs-up or thumbs-down feedback gate used to judge legacy run steps, outputs, and prompt proposals.

Graph-hardening method

Evidence, decision, propagation.

The loop is now a traversal operator over a versioned business graph.

The hardening cycle

1

Select riskChoose the consequential unresolved assumption with the best testing leverage.

2

Define testState method and success threshold before evidence arrives.

3

Record evidenceGrade direction, quality, independence, and recency.

4

DecideAccept, revise, split, or kill.

5

PropagateUpdate dependent claims, plan projections, backlog, and P&L.