bench: clean v0.24.0 board — 6/6, recent 3/3 sources at 1.00 support (#94)

The promised unconfounded re-run: fresh npm ci + build in an isolated
worktree at the release commit, only the shipped change set. The news
adapter alone accounts for recent's recovery (1/3 sources at v0.23.0
-> 3/3 at 1.00 support).

Author: askalf

bench/results/2026-06-12-v0.24.0-clean.md

@@ -0,0 +1,24 @@
+# deepdive bench — 2026-06-12 (v0.24.0 release tag, clean run)
+
+model: `claude-sonnet-4-6 (default)` · base-url: `http://localhost:3456 (default)` · deepdive v0.24.0 (`e73c377`)
+
+Context: the pre-release 6/6 board for the news adapter carried a disclosed
+confound — a concurrent planner-prompt edit shared the build. This run is the
+clean reading promised there: fresh `npm ci` + build in an isolated worktree
+at the release commit, containing ONLY the shipped 0.24.0 change set.
+
+| question | run completed | enough sources | citation support | answer length | topical keywords | under cost ceiling | verdict | time |
+|---|---|---|---|---|---|---|---|---|
+| factual-lookup | ✅ exit 0 | ✅ 12/4 | ✅ 1/1 (1.00) | ✅ 138 words | ✅ all present | ✅ $0.133 / $2.00 | **PASS** | 29s |
+| technical-deep | ✅ exit 0 | ✅ 12/5 | ✅ 8/8 (1.00) | ✅ 399 words | ✅ all present | ✅ $0.150 / $2.00 | **PASS** | 55s |
+| academic | ✅ exit 0 | ✅ 11/5 | ✅ 12/17 (0.71) | ✅ 682 words | ✅ all present | ✅ $0.078 / $2.00 | **PASS** | 97s |
+| comparison | ✅ exit 0 | ✅ 12/4 | ✅ 27/31 (0.87) | ✅ 746 words | ✅ all present | ✅ $0.162 / $2.00 | **PASS** | 75s |
+| recent | ✅ exit 0 | ✅ 3/3 | ✅ 4/4 (1.00) | ✅ 257 words | ✅ all present | ✅ $0.053 / $2.00 | **PASS** | 54s |
+| niche-ops | ✅ exit 0 | ✅ 10/3 | ✅ 2/4 (0.50) | ✅ 155 words | ✅ all present | ✅ $0.089 / $2.00 | **PASS** | 33s |
+
+**6/6 passed.** Structural gates only — read the answers before trusting a comparison.
+
+The perfect board reproduces without the confound: the news-adapter change
+set alone takes `recent` from 1/3 sources (v0.23.0 validation) to 3/3 at
+1.00 citation support. Scoreboard trajectory: 1/6 (v0.21.0, throttled DDG)
+→ 4/6 (v0.22.0) → 5/6 (v0.23.0) → 6/6 (v0.24.0).