==============================================================================
Mutant reduction: what each technique removes, and what it costs
==============================================================================

technique           mutants  dropped    time  speedup  survivors   lost    loss
-------------------------------------------------------------------------------
baseline               1080        0   8.01s    1.00x        125      0    0.0%
dedup                  1080        0   8.38s    0.96x        125      0    0.0%
arid                   1080        0   8.89s    0.90x        125      0    0.0%
one-per-line            520      560   5.73s    1.40x         82     43   34.4%
all three               520      560   4.83s    1.66x         82     43   34.4%
extreme mutation        240      840   4.90s    1.63x          0    125  100.0%

survivors: gaps in the test suite the run reports.
lost:      mutants that survived the full run and are no longer generated at
           all. Each one is a real gap the reduced run will not report, which
           is the price of the speedup in the column beside it.

A reduction technique only pays when per-mutant cost dominates the run. Under the
schemata engine a mutant costs a field write, so halving the mutant count halves
something that is no longer the bottleneck -- and extreme mutation, whose whole-body
replacement cannot be compiled into a schemata class, takes the slower redefinition
path for every one of its mutants. Measure before reaching for any of these.

Extreme mutation is listed alongside these but is not the same kind of thing:
it replaces the mutator set rather than filtering it, so its mutants are not a
subset of the baseline's and its loss figure counts every baseline survivor.
Read its mutant count and survivor count, not its loss.

All reduction techniques are off by default. PIT does not do any of them, so
enabling one turns every dropped mutant into a difference against the
correctness oracle; see tools/parity.