[scbrown]

Summary

sd count --assignee X --priority P put both filters in OPTIONAL { … } FILTER(…) after the work-item type triple. Quipu evaluates a pattern that holds a type triple in written order, so every scoped count scanned every seed before narrowing to the cohort. On a copy of the live board, sd count --assignee franklin --priority 1 --by status sends two such queries and took about 10.7s; the live command is killed at 5s, which blocks the burn-down that needs this count.

• An explicit --assignee becomes VALUES ?assignee { <principal/X> "X" } ?s <agent> ?assignee, and a single non-default --priority becomes a required constant triple. Both go before the type triple, in all four branches (main/ephemeral, with/without a stored status).
• Each is equivalent to the form it replaces: the old FILTER already rejected rows without the value, and VALUES keeps both stored assignee forms. The default priority 2 (applied to seeds without one), priority ranges and --unassigned keep the old form.
• Only count uses the leading position; filter_fields still returns the same constraints for every other caller.

Evidence

On a restored copy of the live board, through a loopback read-only quipu-server 0.16.6:

• sd count --assignee franklin --priority 1 --by status: 20/20 runs, p50 0.19s, p95 0.30s (before: 10.7s uncapped; killed at 5s live).
• 36 scoped counts (6 owners × {no priority, P1, P2} × {total, --by status}) against sd 0.10.0: 35 byte-identical; the other is a count that sd 0.10.0 could not finish inside the server's 10s limit (now 49, matching its own by-status breakdown). Every count is faster; the heaviest went from 10.1s to 3.5s.
• New driven-cohort cases in the scoped-vs-whole-snapshot server test (string and foreign assignees, the default priority, an assignee with no seeds, --assignee with --unassigned, priority ranges) pass with this change and on unchanged main.
• cargo fmt --check, cargo clippy --all-targets -- -D warnings and a release build pass.

Demo: no recording — the observable change is query latency with identical output; the before/after timings and the differential above are the evidence.

Refs aegis-efa.

🤖 Generated with Claude Code

https://claude.ai/code/session_014TTLDAEqq5Whc7qUZsJMUM

Bead: aegis-efa

Overlap preflight: dc9b3e3e3ad91177897f733dc6ecadf673b4d4237d3c1a4ff0da51985a2f4df2

• PR perf(remote): prefer compatible board projection counts #187 @ a8aae0e: independent — Independent and complementary: perf(remote): prefer compatible board projection counts #187 (draft) answers supported count scopes from the board projection and explicitly leaves combined assignee/priority filters on SPARQL because the projection API cannot express them; this candidate speeds exactly that SPARQL path (filter_parts/filter_drive, effective_*_driven). perf(remote): prefer compatible board projection counts #187 moves count_summary's body into sparql_count_summary, which this candidate edits, so whichever lands second rebases the two count_summary hunks; no perf(remote): prefer compatible board projection counts #187 projection logic is absorbed.