======================================================================
RTK ON vs OFF   results/sweep-20260806-191959/fix-build-google-auto
======================================================================

QUALITY  (all trials; a timed-out/errored trial = NOT passed)
  arm                    trials  errored  passed  pass-rate  mean-rwd
  OFF (stock)                 6        1       5        83%     0.833
  ON  (rtk)                   6        1       5        83%     0.833
  -> pass-rate delta = +0 pp

COST & TOKENS  (MEAN over all trials with a value -- errored included)
  metric              OFF mean     ON mean      Δ%  perm-p  n OFF/ON
  Cost USD               $1.92       $2.40  +25.0%   0.399       5/5
  Input tokens       1,675,546   2,060,993  +23.0%   0.410       5/5
  Cache tokens       1,623,127   1,998,408  +23.1%   0.410       5/5
  Output tokens         23,412      31,060  +32.7%   0.329       5/5
  TOTAL spend            $9.60      $12.01  +25.0%

BASH OUTPUT  (mean per trial, from session logs; ~tokens = bytes/4)
  metric                     OFF            ON      Δ%
  bytes                   55,256        61,640  +11.6%
  ~tokens                 13,814        15,410  +11.6%

READ-OUT
  - Cost/tokens use MEAN over ALL trials, not survivor-median, so a
    clipped expensive trial can't fake a saving.
  - perm-p >= 0.05 => the delta is indistinguishable from noise.
  - Pass rate must NOT drop under ON; that's the real regression signal.
