Dashboardopencode-loreConversation cache layout, anchoring, T…

Conversation cache layout, anchoring, TTL, and LTM pin invariants

Category: architecture
Confidence: 1.00
ID: 01a0a580-e978-7f3f-ae47-e459c38286e4
Project ID: 6f4be9ff-ed84-4cca-a9e7-732a0b0b8677
Cross-project: No
Recalled in other projects: 1
Source session: 0t5Rvmq9fQlJArxjW
Created: 2026-08-24 14:23:21
Updated: 2026-09-15 14:38:04

Cross-Project Recalls

ProjectHitsLast recalled
cli 1 8d ago

Content

Use the 1h conversation-cache TTL. The distilled prefix occupies transformed positions [0,1]; durable knowledge deltas stay near the tail, and orphan removal never touches the prefix. Meta passthrough remains uncached. Keep CACHE_ANCHOR_MIN_MESSAGES >= 2 * CACHE_ANCHOR_STEP: below that, the quantized anchor index remains zero and the intermediate anchor never fires. Persisted ltmCacheText must equal ltmPinText; the pin is baseline-only metadata and is never sent upstream. ltm.forSession() must not bust system[2] when the selected entries are unchanged, except for byte-identical re-anchoring. Independently rendering cache and pin text, moving deltas forward, or caching passthrough destabilizes bytes and can cause prefix drift or cache spirals.

Move to: