Document the no-Docker on-ramp ladder (audit → URL-vs-URL compare → twin servers)

[justin808]

Problem

Every quickstart leads with the hardest path — twin Docker servers (production Dockerfile, embedded services, DB seeded at build time; honestly "~2 hours"). But the tool already supports much cheaper first wins, none of them documented as an adoption path:

1. shaka-perf audit --url <u> — zero config, minutes to a full report (perf + a11y + agent-readiness).
2. URL-vs-URL compare — compare --controlURL <a> --experimentURL <b> works against any two servers (packages/shaka-perf/src/compare/cli/program.ts): two preview deployments (Vercel/Netlify/Heroku review apps), staging vs production, or two local checkouts on two ports. Paired-simultaneous sampling still cancels shared client-side noise; the caveat (server-side isolation is now the user's job) belongs in the doc.
3. Single-server validation — control URL == experiment URL to validate test structure (the discover-abtests skill already uses this mode).
4. Twin Docker servers — full A/B rigor.

Teams on serverless/preview-deploy stacks (a huge share of frontend) may never need step 4, and today the docs give them nothing.

Direction

• A "Choose your setup" page: the four rungs, time-to-value, what each can/can't conclude (statistical validity caveats per rung), and when to climb.
• Lead the README/site quickstarts with rung 1–2; link rung 4 as "the full harness."
• Add a preview-deploy example (e.g. GitHub Actions computing the two preview URLs and passing them via --controlURL/--experimentURL).

Refs

Complements #36 (adoption playbook = the guided version of this ladder).

[justin808]

Correction to rung 1: audit --url is not zero-config — loadTests throws when no .abtest.ts/.abtest.js files exist (src/config-loader/load-tests.ts), since audit visits the paths the test files declare. So rung 1 is "no Docker, no A/B pair, one stub test file" (an empty-bodied abTest('Homepage', { startingPath: '/' , options: { visreg: {} } }, async () => {}) suffices). A genuinely URL-only audit mode (synthesize a default homepage test when none exist) would be a nice enhancement to fold into this issue — and note the flip side: audit hard-failing on zero tests is exactly what #36 item 2 asked for, so any URL-only convenience path should stay clearly separated from the evidence-grade flow.