From 35cadf7c9e7df1a22b5c2cdcf264264f360d2889 Mon Sep 17 00:00:00 2001 From: "prompt.ac/@jeffrey" Date: Thu, 1 Oct 2026 09:30:36 -0700 Subject: [PATCH] Walkieware: record the thinking-disabled measurement (0.8 s to first code) Co-Authored-By: Claude Fable 5.1 --- apple/walkieware/Tests/checks.json | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/apple/walkieware/Tests/checks.json b/apple/walkieware/Tests/checks.json index 07a2c65f65..949fe91973 100644 --- a/apple/walkieware/Tests/checks.json +++ b/apple/walkieware/Tests/checks.json @@ -913,5 +913,29 @@ "HeaderSheetsTests 2/2 passed after the dismissal check became a proper disappearance wait", "Device build 78 installed (launch blocked: phone locked)" ] + }, + "thinkingDisabled": { + "date": "2026-10-01", + "deployed": "52417f692e", + "probe": "OpenRouter honors Anthropic thinking:{type:\"disabled\"} on /messages for deepseek-v4.1-flash (0 thinking tokens); reasoning.effort none / enabled false / max_tokens 0 are ignored; /chat/completions honors reasoning too.", + "measured": [ + { + "run": 1, + "totalMs": 4327, + "firstCodeMs": 797, + "thinking": 0, + "rounds": 2 + }, + { + "run": 2, + "totalMs": 6425, + "firstCodeMs": 475, + "thinking": 0, + "rounds": 5 + } + ], + "before": "15959\u201323740 ms total, 11.8\u201323.6 s to first code, 2.3K\u20138.1K thinking tokens", + "quality": "Both runs animate the eating (bobbing hand; berries popping on a clock). Thinking-on runs were a little richer (a chomping critter); the trade is 3\u20135\u00d7 faster visible iteration.", + "build": 79 } } \ No newline at end of file -- 2.51.2