Trends

Shadow-review suggestion activity from the selected dataset and the disagreement-action pattern matrix behind the 84.3% snapshot. All computed by deterministic code — no model generated these numbers.

Suggestion Timeline (newest first)

10 events spanning 48 min — 5 pending, 5 errors (10 total)

WhenModelStatusSummary
13:12:54 openrouter:tencent/hy3:free ⏳ pending Folded A3o turn with pair+ACE vs board; missed thin value. Weak preflop calls with junk.
13:12:46 nvidia:mistralai/mixtral-8x7b-instruct-v0.1 ❌ error (no summary — suggestion errored)
13:00:36 openrouter:tencent/hy3:free ⏳ pending Folds were reasonable for weak hands; no clear leaks. Consider wider steal vs overfolders.
13:00:32 nvidia:mistralai/mixtral-8x7b-instruct-v0.1 ❌ error (no summary — suggestion errored)
13:00:12 featherless:deepseek-ai/DeepSeek-V4-Flash ❌ error (no summary — suggestion errored)
12:44:27 venice:zai-org-glm-5-1 ⏳ pending Hero overfolding blinds to steals; losing chips on -2 deltas with trash but missing +EV resteal spots vs frequent stealers.
12:44:09 featherless:deepseek-ai/DeepSeek-V4-Flash ❌ error (no summary — suggestion errored)
12:36:59 openrouter:tencent/hy3:free ⏳ pending Folded A7o pre in BB check option ok; thin river bet on A7 top pair got called by flush, minor leak.
12:36:54 nvidia:mistralai/mixtral-8x7b-instruct-v0.1 ❌ error (no summary — suggestion errored)
12:24:26 openrouter:tencent/hy3:free ⏳ pending Folded KTo preflop after raise despite +EV; Q9o fold turn weak. Adjust: stack KTo vs raises, value Q9o top pair.

Disagreement Pattern Matrix

1497 / 1775 linked reviews agreed (84.3%). Disagreements shown first.

Our actionLLM actionCount% of all
fold fold 793 44.7% ✓ agree
raise raise 335 18.9% ✓ agree
call call 160 9.0% ✓ agree
check check 136 7.7% ✓ agree
bet check 118 6.6% ⚠ disagree
bet bet 71 4.0% ✓ agree
raise call 67 3.8% ⚠ disagree
raise fold 44 2.5% ⚠ disagree
fold call 20 1.1% ⚠ disagree
call fold 16 0.9% ⚠ disagree
call raise 6 0.3% ⚠ disagree
fold raise 5 0.3% ⚠ disagree
all-in all-in 2 0.1% ✓ agree
check bet 1 0.1% ⚠ disagree
check fold 1 0.1% ⚠ disagree

Snapshot Figures

Linked Reviews

2712

reviews with a 1:1 decision_id link

Agreement Rate

84.3%

1497 / 1775

Top Disagreement

118

largest disagreeing action-pair count

Note: the fixture does not carry per-review timestamps, so an agreement-rate over-time chart would be fabricated. The timeline above shows policy-suggestion creation events (logged epoch-ms), not per-review agreement drift.