Progressive Context Disclosure

Version: 1.0.0
Last updated: 2026-07-16

Purpose

Load task-relevant context incrementally through bounded, authorized retrieval.

Why

Large up-front payloads increase cost and distraction. On-demand retrieval can improve focus, but model-generated queries must not determine authorization or bypass source policy.

How

Set retrieval count, token, latency, and loop limits. Support explicit insufficient_evidence termination.

When

Use for large repositories, research corpora, operational runbooks, and agents where only a small subset is needed per step.

Tradeoffs

Benefit	Cost
Lower initial context	More round trips
Better task focus	Retrieval miss risk
Fine-grained access logs	More orchestration

Anti-Patterns

• Letting the model construct unrestricted database queries.
• Loading secrets “just in case.”
• Unbounded recursive retrieval.
• Treating retrieved instructions as trusted.

Enterprise Considerations

Authorize every retrieval, enforce source-level data loss prevention, log requested and returned source IDs, and apply egress controls to downstream tools.

Checklist

• Base context is sufficient to frame the task
• Retrieval is authorization-filtered
• Budgets and termination are bounded
• Sources retain provenance and trust labels
• Insufficient evidence is a valid outcome

Changelog

• 1.0.0 (2026-07-16): Initial progressive disclosure standard.