Decay

Why Decay

A market analysis from six months ago is less reliable than one from last week. Without decay, old evidence dominates indefinitely, and the system silently treats stale analysis as current truth.

Temporal decay solves this by gradually reducing the certainty of beliefs over time. Old beliefs lose strength. Fresh evidence carries more weight. The system creates natural pressure to refresh.

How It Works

Beliefs decay toward their uninformative starting state as time passes. The longer a belief goes without fresh evidence, the less certain the system is about it.

Each type of belief decays in a way that makes sense for its structure:

• Claims. Lose directional certainty. A strong "yes" gradually weakens toward "we don't know."
• Categories. Lose concentration. A clear "enterprise is the leading segment" gradually weakens toward "could be any segment."
• Measurements. Lose precision. A tight estimate of "34% +/- 5%" gradually widens.

Configurable Decay Rates

The decay rate is configurable for your domain:

• Fast decay. For fast-moving domains like market sentiment and competitive intelligence, where information becomes stale quickly.
• Standard decay. For strategic analysis like market sizing and product positioning (this is the default).
• Slow decay. For stable domains like regulatory environments and fundamental research, where evidence stays relevant longer.
• No decay. Decay can be disabled entirely if all evidence should carry permanent weight.

When Decay Is Applied

Decay is applied at load time, when beliefs are loaded for a new turn. This means:

• The runtime always operates on time-adjusted values
• No background process is needed to age beliefs
• The same belief state loaded at different times correctly reflects the passage of time
• Storage contains the raw values; decay is computed on read

What This Means in Practice

A claim created months ago with strong evidence will show lower confidence than a fresh claim with the same evidence. The system knows the old claim is stale, not because someone flagged it, but because time has passed and the evidence has not been refreshed.

This prevents a common failure mode: agents operating on outdated analysis as if it were current. Decay creates a natural signal that says "this needs to be re-verified."