{"$k":["id","title","columns","rows"],"$r":[["haiku-thinking-paired-test","Same cases, two arms: paired test on exact decisions",{"$k":["key","label","unit"],"$r":[["pair","Pair (first arm = thinking off)","text"],["cases","Cases","count"],["bothRight","Both right","count"],["onlyOff","Only thinking off right","count"],["onlyOther","Only the other arm right","count"],["bothWrong","Both wrong","count"],["p","Exact McNemar p (two-sided)","\u0001"]]},[{"pair":"Thinking off vs thinking on (the same Haiku 4.5)","cases":82,"bothRight":67,"onlyOff":4,"onlyOther":6,"bothWrong":5,"p":0.754},{"pair":"Thinking off vs Sonnet 5.5 at low effort (secondary)","cases":82,"bothRight":70,"onlyOff":1,"onlyOther":7,"bothWrong":4,"p":0.07}]],["haiku-thinking-router-cells","Routing: every arm",{"$k":["key","label","unit"],"$r":[["arm","Arm","text"],["setting","Thinking setting","text"],["calls","Calls (errors)","text"],["exact","Exact","text"],["keys","Per-question","text"],["wall","Wall p50 / p95 (s)","text"],["api","API p50 / p95 (s)","text"],["wallRange","Wall min to max (s; not an interval)","text"],["apiRange","API min to max (s; not an interval)","text"],["thinking","Thinking tokens per decision (range)","text"],["visible","Visible output tokens","tokens"],["cost","USD per 1,000 (calculation)","usd"]]},{"$k":["arm","setting","calls","exact","keys","wall","api","wallRange","apiRange","thinking","visible","cost"],"$r":[["Claude Haiku 4.5 (thinking off) · Claude Code","MAX_THINKING_TOKENS=0","82 (0)","87% (71/82; 95% interval 78% to 92%)","91% (177/194; 95% interval 86% to 94%)","4.66 / 8.18","3.79 / 7.43","2.2 to 11.57","1.37 to 10.83","0 (0 to 0)",366,3.364],["Claude Haiku 4.5 (thinking on) · Claude Code","CLI default","82 (0)","89% (73/82; 95% interval 80% to 94%)","94% (183/194; 95% interval 90% to 97%)","12.54 / 34.48","10.51 / 32.13","5.86 to 51.28","4.33 to 49.49","1101 (285 to 4443)",318,8.924],["Claude Sonnet 5.5 (low) · Claude Code","--effort low","82 (0)","94% (77/82; 95% interval 87% to 97%)","97% (189/194; 95% interval 94% to 99%)","2.6 / 4.3","1.6 / 2.58","1.99 to 5.58","1.06 to 4.76","2 (0 to 63)",105,7.324]]}],["haiku-thinking-by-decision-type","Routing: exact decisions by decision type",{"$k":["key","label","unit"],"$r":[["decision","Decision type (cases)","text"],["off","Claude Haiku 4.5 (thinking off) · Claude Code","text"],["on","Claude Haiku 4.5 (thinking on) · Claude Code","text"],["sonnet","Claude Sonnet 5.5 (low) · Claude Code","text"]]},{"$k":["decision","off","on","sonnet"],"$r":[["Failure class (18)","83% (15/18; 95% interval 61% to 94%)","94% (17/18; 95% interval 74% to 99%)","100% (18/18; 95% interval 82% to 100%)"],["Message intent (20)","100% (20/20; 95% interval 84% to 100%)","100% (20/20; 95% interval 84% to 100%)","100% (20/20; 95% interval 84% to 100%)"],["Is it a rule? (12)","100% (12/12; 95% interval 76% to 100%)","100% (12/12; 95% interval 76% to 100%)","100% (12/12; 95% interval 76% to 100%)"],["Context shape (32)","75% (24/32; 95% interval 58% to 87%)","75% (24/32; 95% interval 58% to 87%)","84% (27/32; 95% interval 68% to 93%)"]]}],["haiku-thinking-hard-cells","Hard tasks: both cells",{"$k":["key","label","unit"],"$r":[["cell","Cell","text"],["calls","Calls","count"],["strict","Strict pass","text"],["formatMisses","Format misses","count"],["wrong","Wrong answers","count"],["noAnswer","Errors or incomplete calls (failures)","count"],["total","Total time median (min to max, s)","text"],["reasoning","Reasoning tokens median","tokens"],["output","Output tokens median","tokens"],["perPass","USD per strict pass (calculation)","usd"]]},[{"cell":"Claude Haiku 4.5 (thinking off) · Claude Code","calls":24,"strict":"17% (4/24; 95% interval 7% to 36%)","formatMisses":0,"wrong":20,"noAnswer":0,"total":"2.9 (1.7 to 13)","reasoning":0,"output":291,"perPass":0.03654},{"cell":"Claude Haiku 4.5 (thinking on) · Claude Code","calls":24,"strict":"46% (11/24; 95% interval 28% to 65%)","formatMisses":5,"wrong":8,"noAnswer":0,"total":"39 (15.3 to 75.1)","reasoning":4556,"output":5064,"perPass":0.0672}]],["haiku-thinking-hard-matrix","Hard tasks: strict passes per task",{"$k":["key","label","unit"],"$r":[["task","Task","text"],["off","Claude Haiku 4.5 (thinking off) · Claude Code","text"],["on","Claude Haiku 4.5 (thinking on) · Claude Code","text"]]},{"$k":["task","off","on"],"$r":[["Fix an interval-merge function (off-by-one and edge cases)","100% (3/3; 95% interval 44% to 100%)","100% (3/3; 95% interval 44% to 100%)"],["Fix a time-zone day-length function (DST)","0% (0/3; 95% interval 0% to 56%)","33% (1/3; 95% interval 6% to 79%)"],["Write a CSV parser (quoted newlines, strict errors)","0% (0/3; 95% interval 0% to 56%)","67% (2/3; 95% interval 21% to 94%) (+1 format miss)"],["Predict JavaScript event-loop output order","0% (0/3; 95% interval 0% to 56%)","0% (0/3; 95% interval 0% to 56%)"],["Solve a multi-constraint room schedule","0% (0/3; 95% interval 0% to 56%)","0% (0/3; 95% interval 0% to 56%) (+2 format misses)"],["Write a strict SemVer 2.0.0 regex","0% (0/3; 95% interval 0% to 56%)","100% (3/3; 95% interval 44% to 100%)"],["Refactor to remove duplication, keep 20 tests green","33% (1/3; 95% interval 6% to 79%)","67% (2/3; 95% interval 21% to 94%)"],["Write a SQLite reporting query (fan-out, ties, boundaries)","0% (0/3; 95% interval 0% to 56%)","0% (0/3; 95% interval 0% to 56%) (+2 format misses)"]]}]]}