Every source is time-windowed: evidence that was true when it was gathered can be stale by the time a decision relies on it.
Why it matters when the plan changes
Organisational evidence decays at different speeds. A reporting line can change overnight, a strategy document can be superseded, and an interview reflects the week it was held. When a judgement mixes sources without tracking when each was current, an old answer quietly shapes a new decision. Even supervisory guidance ages: SR 11-7, the model-risk guidance banks followed from 2011, has been superseded and replaced by SR 26-2.
The tension is that the freshest source is not always the best. A recent interview may be more current and less reliable than an assessment taken a year earlier, because it captures a mood rather than a pattern. A validity window therefore depends on what the source measures, and the judgement has to say which sources it relied on and how current each one was when it was used.
In practice
A forecast prepared before a change of ownership relied on a strategy document, the organisation chart and interviews with the leadership team. After the transaction, the chart and the strategy are both outside their windows, while the behavioural evidence still holds. The reforecast keeps the assessment evidence, replaces the other two, and records which sources changed.
Evidence
Model-risk guidance treats ongoing monitoring as checking whether a model still performs as expected given potential changes, including in data relevance.
Board of Governors of the Federal Reserve System, OCC and FDIC, Revised Guidance on Model Risk Management (SR 26-2) (2026)SR 26-2 supersedes and replaces SR 11-7, the guidance on model risk management issued in 2011.
Federal Reserve, SR 26-2, Revised Guidance on Model Risk Management (2026)
What it cannot tell you
A validity window is a judgement about how fast a source ages, not a measured property of it. Two reviewers can set different windows for the same interview or document, and a source inside its window can still be wrong, so the window limits staleness without guaranteeing accuracy.
How Atlas reads it
The validity gate is part of how the forecast is formed: every document is time-windowed, and stale context is kept out of the reading rather than blended into it. The audit record is meant to make each material claim traceable to the assessment, document and model versions it relied on, so a later reviewer can see what the judgement knew.
Questions
It depends on the source. Relatively stable behavioural patterns age slowly. Plans, organisation charts, ownership and dependencies can age with a single decision. Interviews and short check-ins reflect the moment they were taken. The window is set by how quickly the thing described changes, not by a fixed period.
It is either excluded from the judgement or used only with an explicit note that it may be stale. Blending it in silently is the failure the window exists to prevent, because an out-of-date source can carry the same apparent weight in a judgement as a current one.
Model-risk supervision offers one. The Federal Reserve's 2026 guidance treats ongoing monitoring as checking whether a model still performs as expected given potential changes, including in data relevance. The guidance itself shows the point: SR 26-2 superseded and replaced SR 11-7, issued in 2011.
Not every one. A transaction usually changes authority, priorities and the plan, which puts strategy documents, charts and ownership maps outside their windows. Evidence about how people behave under pressure changes more slowly and may still hold, which is why each source is re-dated rather than all discarded.
The author of the judgement, with the reasoning recorded so a reviewer can challenge it. Supervisory guidance on model risk from 2026 expects similar documentation. A later reviewer should be able to reconstruct which sources were used and when each was current, which makes the call defensible after the fact.