Canonical · L4 Primitive·Shipped
Token bloat vs. governed memory
Token bloat is the accumulation of more text than the system can use coherently. Governed memory is the promotion of reviewed lessons into durable guidance. Growing the window does not solve drift.
Plain definition
Token bloat is the accumulation of more text than the system can use coherently.
Governed memory is the promotion of reviewed lessons into durable guidance.
Why it matters
Long histories preserve more text without preserving operational meaning. Without a review-and-promotion path, the model re-interprets the same provisional truths every turn.
What it is not
- A summarization technique.
- A retrieval optimization.
- An argument against long-context models.
- A prompt-compression trick.
Where it appears in Ubiquity
CGG addresses token bloat by reducing the need to keep every old line alive. It distills reviewed lessons into future-session guidance.
Ladder context
Demand ladder
L1 Pain→L2 Contrast→L3 Category→L4 Primitive→L5 Branded
← Up the funnel
Cross-session lesson compoundingDown the funnel →
Claude Code context governanceAdjacent transitions
Related terms
Frequently asked
- What is token bloat?
- Accumulation of more text in the context window than the system can use coherently.
- Why is token bloat a problem?
- Because more text is not more meaning. The model can keep re-interpreting provisional truths it should have promoted to durable guidance.
- How is governed memory different?
- Governed memory routes lessons through human review and promotes approved lessons into durable guidance the next session inherits.
- Does a larger context window solve this?
- No. A bigger window is not a governed context.