Context Quality and Decay
Freshness, confidence, ownership, and staleness mechanics
Part IV: Context Quality and Decay
This is where most context systems quietly die.
An Operating Context System is not static. Context has a half-life. Things that were true in January may be false in March. An initiative that was well-scoped when it was created accumulates ambiguity as circumstances change and the definition isn't updated. A team structure that reflected reality six months ago may be completely wrong today.
The failure mode is insidious: the system looks like it's working because data exists. But the data is stale, and the organization doesn't know how stale. Decisions get made based on information that is plausible but wrong.
| Quality Dimension | Definition | Mechanism |
|---|---|---|
| Context freshness | How quickly does this type of context go stale? | Different decay rates per artifact type; modeled explicitly |
| Staleness scoring | A visible signal that context may not be trustworthy | Objects flagged when not touched or confirmed within a defined window |
| Confidence levels | Declared vs. inferred vs. validated context | Confidence tier visible when context is used in decisions |
| Ownership | Who is accountable for this staying true | Named owner per context object; targeted nudges when it drifts |
| Decay mechanics | What happens when no one touches something | Automatic archiving, staleness flags, or agent refresh attempts |
Confidence levels deserve special attention. Context exists on a spectrum from declared (a human explicitly said this is true) to inferred (an agent connected these dots based on signals) to validated (a human reviewed and confirmed an agent-generated connection). These are meaningfully different confidence levels, and they should be visible when the context is being used. An AI agent making a recommendation based on inferred context should treat that context differently than confirmed context.
The wrong answer on decay mechanics is doing nothing. When no one is responsible for refreshing context and nothing happens when it goes stale, the system quietly becomes a liability rather than an asset.