[{"id":"harder-h2h-cells","title":"Cells: configuration, calls and outcomes","columns":{"$k":["key","label","unit"],"$r":[["config","Configuration","text"],["calls","Calls","count"],["strict","Strict passes","text"],["interval","95% Wilson interval","text"],["formatMisses","Format misses","count"],["wrong","Wrong answers","count"],["noAnswer","No answer","count"],["priced","Calls with a token report","count"],["completed","Completed calls (timing and token n)","count"],["medianTotal","Median completed-call total (s)","seconds"],["rangeTotal","Completed-call time range (s, not an interval)","text"],["p95Total","p95 total (s)","seconds"],["medianOut","Median output tokens","tokens"]]},"rows":{"$k":["config","calls","strict","interval","formatMisses","wrong","noAnswer","priced","completed","rangeTotal","medianTotal","p95Total","medianOut"],"$r":[["GPT-6.1 Sol (medium) · Codex CLI",16,"11/16","44% to 86%",0,2,3,13,13,"46.24 to 273.46",120.24,240.94,4994],["Claude Opus 5.5 · Claude Code",12,"5/12","19% to 68%",1,3,3,11,9,"3.82 to 279.5",80.34,208.56,8420],["Claude Sonnet 5.5 · Claude Code",16,"6/16","18% to 61%",0,6,4,12,12,"4.32 to 210.08",70.43,208.72,9287],["Claude Haiku 4.5 · Claude Code",12,"0/12","0% to 24%",0,10,2,10,10,"25.73 to 223.95",108.98,202.31,12508]]}},{"id":"harder-h2h-tasks","title":"Candidate tasks, pilot result and selection (prompts not shown)","columns":{"$k":["key","label","unit"],"$r":[["task","Task","text"],["kind","Kind","text"],["checks","Validator checks","count"],["round","Pilot round","text"],["pilot","Sonnet 5.5 pilot (strict passes)","text"],["selected","In the counted set","text"],["controls","Controls (wrong answers rejected, wrapped reference flagged)","text"]]},"rows":{"$k":["task","kind","checks","round","pilot","selected","controls"],"$r":[["Write a TTL and LRU cache class (random operation sequences)","code",13,"first 12 candidates","2/2","no","5/5; yes"],["Write a glob matcher (braces, **, character sets, hidden files)","code",68,"first 12 candidates","2/2","no","4/4; yes"],["Write a unified diff (minimal script, fixed tie-break, hunk headers)","code",20,"first 12 candidates","2/2","no","4/4; yes"],["Evaluate Python-style integer expressions (precedence, chained comparisons)","code",64,"first 12 candidates","2/2","no","5/5; yes"],["SQLite sessions report (gaps and islands, median, logout rule)","SQL",7,"first 12 candidates","2/2","no","5/5; yes"],["SQLite as-of price and currency report (missing days, rounding)","SQL",14,"first 12 candidates","2/2","no","5/5; yes"],["Solve a 6x6 Skyscrapers puzzle (14 clues, one solution)","reasoning",1,"first 12 candidates","0/2 (+2 format misses)","yes","3/3; yes"],["Pick the most profitable jobs for two machines (one optimum)","reasoning",1,"first 12 candidates","2/2","no","3/3; yes"],["Decode UTF-8 with one U+FFFD per maximal subpart","spec",35,"first 12 candidates","2/2","no","4/4; yes"],["Next run of a cron expression in UTC (day-of-month or day-of-week rule)","spec",37,"first 12 candidates","2/2","no","4/4; yes"],["Simulate a retry queue (priorities, timeouts, backoff, ties)","simulation",1,"first 12 candidates","2/2","no","4/4; yes"],["Correctly rounded sum of doubles (ties, overflow, negative zero)","numeric",32,"first 12 candidates","2/2","no","4/4; yes"],["Solve a 10x10 nonogram (one solution)","reasoning",1,"replacement candidates","1/2","yes","4/4; yes"],["Pick the most profitable jobs for three machines (30 jobs, one optimum)","reasoning",1,"replacement candidates","2/2","no","5/5; yes"],["Predict the output of a seeded shuffle (32-bit integer arithmetic)","code reading",1,"replacement candidates","0/2","yes","4/4; yes"],["Solve a 9x9 Sudoku with 22 givens (one solution)","reasoning",1,"replacement candidates","1/2","yes","3/3; yes"]]}}]