corpus n=510   avg_wlen 566.2
probe-term df: 2-12

baseline b=0.75 (shipped). One lever moves; `flen` is the shipped one in every arm.

   family    n  hit@1 b=0.75   hit@1 b=0.4    hit@1 b=0.3    hit@1 b=0.2    hit@1 b=0.15 
     main   30      0 /30         0 /30         0 /30         0 /30        30 /30   
  inverse   30     30 /30        30 /30        30 /30        30 /30        30 /30   
  placebo   30     30 /30        30 /30        30 /30        30 /30        30 /30   
     dump   30     30 /30        30 /30        30 /30        30 /30        30 /30   
  content   30      0 /30         0 /30        30 /30        30 /30        30 /30   
  verbose   30     30 /30        30 /30        30 /30        30 /30        30 /30   

Paired against the baseline, per family (b = better, w = worse, net = b - w):

   family      b  better  worse  discordant    net         p  outcome
     main    0.4       0      0           0     +0    1.0000  inconclusive
  inverse    0.4       0      0           0     +0    1.0000  inconclusive
  placebo    0.4       0      0           0     +0    1.0000  inconclusive
     dump    0.4       0      0           0     +0    1.0000  inconclusive
  content    0.4       0      0           0     +0    1.0000  inconclusive
  verbose    0.4       0      0           0     +0    1.0000  inconclusive

     main    0.3       0      0           0     +0    1.0000  inconclusive
  inverse    0.3       0      0           0     +0    1.0000  inconclusive
  placebo    0.3       0      0           0     +0    1.0000  inconclusive
     dump    0.3       0      0           0     +0    1.0000  inconclusive
  content    0.3      30      0          30    +30    0.0000  b=0.3 RANKS BETTER
  verbose    0.3       0      0           0     +0    1.0000  inconclusive

     main    0.2       0      0           0     +0    1.0000  inconclusive
  inverse    0.2       0      0           0     +0    1.0000  inconclusive
  placebo    0.2       0      0           0     +0    1.0000  inconclusive
     dump    0.2       0      0           0     +0    1.0000  inconclusive
  content    0.2      30      0          30    +30    0.0000  b=0.2 RANKS BETTER
  verbose    0.2       0      0           0     +0    1.0000  inconclusive

     main   0.15      30      0          30    +30    0.0000  b=0.15 RANKS BETTER
  inverse   0.15       0      0           0     +0    1.0000  inconclusive
  placebo   0.15       0      0           0     +0    1.0000  inconclusive
     dump   0.15       0      0           0     +0    1.0000  inconclusive
  content   0.15      30      0          30    +30    0.0000  b=0.15 RANKS BETTER
  verbose   0.15       0      0           0     +0    1.0000  inconclusive

🔴 The decision rule is the PRE-REGISTRATION's, and this command does not apply it:
   the first value, DESCENDING, netting positive on `content` AND `main` —
   each individually, neither negative — with EVERY control holding
   (inverse, placebo, dump, verbose), and the net clearing SR-RS decision 19's
   floor for the discordant count observed. An ambiguous result goes to Arpit.
   ⚠ `dump` is a CONTROL since 2026-09-16, not a benefit family: it is 30/30
      at the baseline and can never net positive.
