Documentation
How Helen’s Foundry works
Helen’s Foundry keeps the work you do connected to the reasons you do it. Artifacts form a chain across four tiers. Checks find where the chain breaks. You decide what to do in Reconcile, and agents read the same chain before they act.
The context chain
Everything in Helen’s Foundry is an artifact. Each artifact type belongs to one of four tiers: Orient, Directional, Structural, and Execution. Each tier gives the next one its reason to exist. A ticket exists because of a project. A project advances a goal. A goal advances the strategy.
There are two ways into the tiers. Planning starts at the mission and works down through strategy and goals. Discovery starts at the domain model and works through personas and problems. Both meet in the Execution tier, where projects serve personas and address problems.
Orient
Tier 1What world we create and the territory we operate in
Directional
Tier 2How we win and what outcomes define progress
Structural
Tier 3Who we serve, what they need, and how work flows
Execution
Tier 4What gets delivered, tracked, and shipped
- Mission
- Why the company exists. It rarely changes, and every other artifact traces back to it.
- Vision
- The future you build toward. It changes when the market or your ambition changes.
- Domain Model
- The territory you operate in: the market, its key concepts, and its boundaries.
- Personas
- A research-backed description of a core user, placed inside the domain model.
- Problem Definitions
- The problem you solve and who has it. Projects address problems.
Findings and Reconcile
Checks run against your workspace and report findings. A finding is evidence to inspect, not an error. A finding is a conflict only when two artifacts make claims that cannot both be true. The check reference lists every check.
- Reconcile
- The page where findings and proposed changes wait for a decision. Each row names the action that applies to it.
- Ratify
- An authorized person grants authority to an artifact’s premise, the text the artifact rests on. Agents cite ratified premises as authority. An edit to the premise invalidates the ratification, and you re-ratify to restore it.
- Renew validation
- Confirms that a ratified premise is still current. It resets the validation clock and does not change authority.
- Apply fix, acknowledge, or dismiss
- Apply fix makes the correction a check proposes; the finding resolves when the next check confirms it. Acknowledge records that you know about a finding without claiming a fix. Dismiss records that a finding does not apply.
Examples
A child judged to contradict the parent it serves
A goal says “Reduce setup time for small teams.” A project under it plans a mandatory 40-field setup form. A language model judged that the project works against its goal.
What to do: Reframe the project, move it under a goal it serves, or close it. If the judgment is wrong, dismiss the finding.
Read about this checkCurrent premise needs ratification or validation renewal
Your team ratified the strategy in March. In June, someone edited its premise. The March ratification does not cover the new text.
What to do: Read the new premise. If it is correct, re-ratify it. When only time has passed and the premise is unchanged, renew validation.
Read about this checkActive work building on an abandoned or obsolete ancestor
A project closed as obsolete. Two tickets under it are still in progress, so they build on work the team stopped.
What to do: Close the tickets if the work is no longer necessary. Otherwise, move them to a live project.
Read about this checkAgents and Helen
Coding agents connect over MCP. Before an agent starts a ticket, it reads the ticket’s context chain: the project, goal, and strategy above it, and the personas and problems it serves. See Connect an agent and the MCP tool reference.
Helen is the assistant built into the product. She reads the whole graph to find contradictions and raises findings, so the reason for a decision stays with it.
Reading workspace health
Insights summarizes the health of your graph: how much execution work traces back to live strategy, how work splits between people and agents, and where intent and delivery separate. Use it to find where to look, not as a score. The Insights glossary defines each number. A healthy workspace shows these signs:
- Every ticket traces through a project to a goal.
- Active personas and problems have live projects connected to them.
- Findings get a decision instead of accumulating in Reconcile.
- Ratified premises stay current, and validation is renewed on cadence.