Point
Atlas can be reduced or rejected
Project investment and format coherence do not justify keeping Atlas when measured value fails to exceed semantic and maintenance cost.
- Posture
- asserted
- Lifecycle
- active
Atlas is not an end in itself. A valid format and a passing release gate show that the project can state and reproduce its contract; they do not show that the contract deserves adoption. Negative product evidence is allowed to remove a Map, simplify a Point model, defer an application, or end the project.
The decision must compare observed gains with authoring, review, training, validation, integration, and maintenance costs. Sunk cost, historical effort, visual metaphor, or self-hosting scale cannot override that comparison. Rejection criteria are recorded before the portal study so an unfavorable result remains actionable.
Rejection boundaryFailure against explicit kill criteria is a reason to reduce or remove Atlas despite prior investment.
Useful differenceMeasured benefit must exceed semantic and maintenance cost before Atlas counts as a useful difference.
Evidence collection is useful only when a neutral or negative result can remove structure, defer a surface, or end the project.
The project records stop conditions before the portal study and separates protocol readiness from outcome claims. This protects learning from sunk-cost reasoning and makes simplification an expected response when cross-context benefits do not exceed authoring and maintenance burdens.
Costs and failure modesNegative results must count against continued semantic and maintenance investment.
Learning agendaPrecommitted stop conditions turn neutral or negative evidence into an actionable change of direction.