Proposal: optional `durability: durable | volatile` frontmatter field for GC/compaction hints

[chirag127]

Proposal

Optional frontmatter field durability: durable | volatile.

Meaning

• durable — concept should persist across bundle cleanups. Fundamental facts, standing decisions, invariants.
• volatile — concept is time-boxed; safe to prune during garbage collection or context compaction. Session summaries, migration notes, in-flight state.
• absent — no assertion; consumers treat as durable by default (backward-compatible).

Why not covered by existing fields

• confidence (Proposal: graded confidence — an epistemic-reliability axis, distinct from integrity (#140) and citation (#92/#94) #151) — epistemic grade. Orthogonal: a high confidence fact can still be volatile (correct-but-time-boxed).
• lifecycle from Proposal: Knowledge Maturity Model — combining confidence, evidence, lifecycle, and verification gates #160 — describes verification transitions (proposed → active → expired). Orthogonal: a volatile concept can be active its entire lifetime.
• status: active|deprecated|draft — describes publication state. Orthogonal.
• timestamp — records when written; consumers can't distinguish "old but durable" from "old and expired."

Concrete gap

An agent doing context compaction over a bundle needs to answer: "if I evict this concept, do I lose institutional knowledge or ephemeral state?" Currently the only signal is guessing from type / path / age.

Adopter data

828 concept files at chirag127/workspace/knowledge/ ship this field optionally as of 2026-07-03. Public bundle: knowledge.oriz.in.

Proposed SPEC.md addition (§4.1 Frontmatter, Extensions block)

durability: durable | volatile   # optional; controls GC/compaction eligibility. Absent = durable.

Backwards-compatible — no existing bundle breaks. Consumers ignoring the field keep prior behavior.

Not-in-scope

• Prescribing WHEN to run GC.
• Prescribing tooling to detect volatility.
• Overlapping with lifecycle (Proposal: Knowledge Maturity Model — combining confidence, evidence, lifecycle, and verification gates #160) — orthogonal axes.

Thanks for the great work on OKF!