[2217173240]

The planner overview filtered running work out of the latest 300 intents, so newer finished work could hide older active work. Facts and findings were also reported as the size of a latest-1000 list. Query running intents before applying the display limit, return their full count and truncation flag from one SQL snapshot, and reuse the existing Stats query for complete fact/finding counts.

Keep failed reads explicitly unknown through null totals, incomplete and named query_errors while retaining successful sections. A code-owned planner notice explains unknown values and truncated lists. The overview remains a collection of independent reads; this change does not establish natural-language goal correctness or alter task completion, approval, inheritance or digest rules.

This is a narrow current-architecture adaptation of RuochenPan / RC-CHN's 3a99ec0a, AGPL-3.0. The running list/count uses one SQL window query instead of separate source queries. Attribution and behavior are documented in docs/fork-integrations.md and docs/evidence-provenance.md.

Validation:

• Full agent and db package tests passed with -race -count=1 against separate fresh PostgreSQL 16 databases; the live-model review test was explicitly excluded.
• Actual PostgreSQL regressions cover older running work beneath hundreds of finished intents, more than 300 running intents, more than 1000 facts/findings, and selective query failures with other sections still available.
• Independent review and focused real PostgreSQL tests passed; source file hashes are bound to the review record. Existing cold-graph failure behavior remains covered.
• Production prompt rendering distinguishes known zero/empty values from unknown and truncated state. No model, paid-provider or real-target requests were made.

Exact-HEAD quality and CodeQL checks are required before normal merge.