Comprehension Debt
What this is
Comprehension debt is the gap between what an execution pattern generates and what any human actually reads, understands, and stands behind. Volume-generating patterns — fan-out, generate- and-filter, tournaments, loops — can produce diffs, drafts, and artifacts faster than reviewers can absorb them. Unread output is not value; it is deferred risk.
This is a review discipline, not a runtime control. It is the governance counterweight that makes the volume-generating patterns in the Execution Pattern Library admissible at all.
What every volume-generating pattern must declare
Any pattern that generates output volume must declare, before it runs:
- who reads the diffs or outputs;
- what percentage must be reviewed;
- which gate proves quality;
- when to pause the workflow;
- when to reduce scope;
- when to require human review.
If none of these can be answered, the pattern is generating debt by design and should be rejected or downgraded at selection time (see Execution Vehicle Selection).
Comprehension debt is not context rot
These are related but distinct failure modes, and they require different controls. Context rot is degradation inside one agent session: as history grows, the agent contradicts itself, forgets constraints, or drifts from the goal — the agent’s working memory decays. Comprehension debt is degradation on the human side of the system: the agents may be perfectly coherent while producing more artifacts than any reviewer reads, so human understanding of the generated work decays. Context rot is fixed with checkpoints, re-anchoring, and session restarts; comprehension debt is fixed with declared readership, review percentages, quality gates, and the discipline to pause or reduce scope. A workflow can have zero context rot and still be drowning in comprehension debt.
Review shape
comprehension_debt_review:
execution_id:
pattern:
period:
artifacts_generated:
artifacts_reviewed_by_human:
accepted_change_rate:
unread_artifact_risk: low | medium | high
reviewer_capacity:
debt_signals:
- repeated_confusion
- unknown_generated_area
- fragile_gate
- high_output_volume
decision: continue | reduce_scope | pause | retire
Related
- Execution Pattern Governance — the layer that demands this declaration
- Execution Pattern Library — which patterns generate volume
- Context-Rot Controls — the in-session counterpart failure mode
- Agent Harness Contract — the per-task envelope the reviewed work runs under