Remove generation profile tiering; use explicit runs count

docs/performance.md

@@ -23,8 +23,8 @@ BENCH_RUNS=12 BENCH_WARMUP=3 npm run benchmark:cli
 # Increase inner-loop work for micro-benchmarks
 BENCH_INNER_ITERS=5000 npm run benchmark:hot
 
-# Benchmark generation profile matrix
-BENCH_GENERATION_PROFILES=quick,standard,thorough npm run benchmark:all
+# Benchmark with varying test counts
+BENCH_RUNS=10,50,200 npm run benchmark:all
 ```
 
 ## Capture CPU Profile for Qualify
@@ -41,10 +41,10 @@ This writes Chrome-compatible CPU profiles to `.profiles/qualify.cpuprofile` and
 - CLI benchmark uses spawned `node dist/cli/index.js` commands so startup costs are included.
 - Hot path benchmark runs in-process for lower-noise function-level comparisons.
 - Use fixed `--seed` for qualify benchmarks to keep runs deterministic.
-- Generation now adapts to depth: `quick` favors bounded payload generation speed, `thorough` keeps broader generation.
+- Schema generation uses fixed defaults (string≤128, array≤10) regardless of run count.
 
-You can override generation per run:
+You can override runs per preset:
 
 ```bash
-apophis qualify --profile oauth-nightly --generation-profile quick --seed 42
+apophis qualify --profile oauth-nightly --seed 42
 ```