test-planlisted
Install: claude install-skill seercraft-io/seer-skills
# Test plan
Done means: `seer/tests.md` with 5 tests, ranked, each one changing one thing and backed by cited
rival evidence. The person could hand it to a creative team today.
Read `${CLAUDE_SKILL_DIR}/../seer-start/reference/house-rules.md` first.
## Steps
1. Gather what's already in `seer/reports/` from the last 14 days (teardowns, long-runners, hooks,
angles, offers, formats). If there's nothing, run the `long-runners` steps first; tests without
evidence are guesses.
2. Read `seer/brand.md`, and ask once what the person is running now (their current best ad, if they
know it) so tests have a control. If they skip, the control is "your current best ad".
3. Draft 8 candidate tests. Each changes exactly one variable: hook, angle, format, offer, proof,
creator vs studio, landing page. Each must point at rival evidence: an idea running 60+ days, a
pattern across 3+ ideas, a new push several rivals made, or an empty angle (marked untested).
4. Score each 1-3 on: evidence strength, fit with the brand's real claims, cost to make. Keep the top 5.
At most 2 tests on the same variable.
## Output (`seer/tests.md`)
For each test, in rank order:
```
## Test N: <short name>
- Hypothesis: if we <change>, then <metric> improves, because <evidence>.
- Change only: <the variable> · Keep the same: <everything else>
- Evidence: <cited ads, days running, "4 of 10 long-runners">
- Make: <format, length, who's in it, what's on screen in the first 3 seconds>
- Read it by: