Dashboardopencode-loreTemporal message identity, artifact exc…

Temporal message identity, artifact exclusion, and user-prose retention

Category: architecture
Confidence: 1.00
ID: 01a0a9ba-fe40-7fc6-b4c5-a4d4caa4bbec
Project ID: 6f4be9ff-ed84-4cca-a9e7-732a0b0b8677
Cross-project: No
Recalled in other projects: 0
Source session: 0IiwozJmFR9A6QyFo
Created: 2026-08-25 23:40:45
Updated: 2026-09-16 10:19:59

Content

Derive temporal_messages.id from tenant, project, session, and source rather than caller source_id; sync, vectors, distillation links, and multiple tenants require stable collision-free keys. Never expose synthetic resolution or refcheck artifacts to the model or persist them in temporal storage; they are internal probes, and the resolver must recover internally rather than throw into the request path. During oversized-message reduction, retain the leading user-prose prefix verbatim (userBlobHeadChars, default 1500 and configurable to 0) before relevance scoring. Relevance-only reduction can discard opening constraints while retaining incidental later text. Elided bulk remains searchable through temporal_messages and FTS.

Move to: