Lessons 59.1–59.5 taught the durable workflow. This appendix is the shared setup you run before any tool-specific lab (59.B–59.H): the same brief, the same reference, the same scorecard, on two or more platforms. Same test, honest comparison.

The shared brief (use verbatim, 8–12 seconds)

Do not invent a new brief per tool. That destroys comparability. Use this one:

• Subject: your real next subject — one product, one character, one place. Name it exactly.
• Action: one physical action ("steam rises off loaf; hand dusts flour at frame edge").
• Environment: your 59.1 location in one line.
• Camera: one move ("locked close-up, shallow depth" recommended for round one).
• Duration: 8–12 seconds, single shot.
• Constraints: must-preserve facts from your CONTINUITY.md (label text, colors, wardrobe) plus "no text rendered in-frame, no extra hands."
• Reference: the same approved still file on every platform that accepts image input. Record where a platform does not support references — that is itself a score.

Save the brief text and reference file hash/name in every test card so a reviewer can reproduce the run.

The scorecard: 11 dimensions

Score each run 1–5 with one evidence line (timestamp, screenshot, or file name — never memory):

1. Visual quality — clean, stable, artifact-free at phone size. 2. Brief adherence — did it do what the brief asked, nothing else? 3. Motion / camera control — the requested move, at the requested pace, without drift. 4. Character / product continuity — must-preserve facts intact across frames. 5. Reference control — how faithfully an input image/style/character steered the output. 6. Audio / dialogue (when the job needs it) — sync, clarity, usability; mark N/A honestly when the shot is silent. 7. Editing / extension options — keyframes, extend, inpaint, scene assembly, timeline export. 8. Generation speed — wall-clock minutes per usable second, including retries. 9. Cost — credits/cash per usable second at current pricing; note plan and date. 10. Export quality — resolution, ratio options (9:16/1:1/16:9), watermark/license terms. 11. Availability — works in your region, on your account tier, today.

Then answer the only question that matters: **"Which environment is best for *this shot*?"** A fashion mood clip, a reference-locked product shot, a multi-shot scene, and a talking avatar may each deserve a different winner.

VIDEO-TOOL-TEST.md template

# VIDEO-TOOL-TEST.md — [date, region, account tier]

## Shared brief + reference
- Brief text (verbatim): ...
- Reference file: `refs/...` (or "no reference supported")

## Run A — [tool, model/surface, settings]
- Time / cost: ... / Screenshots: ...
- Scores 1–5 (11 dims with one evidence line each):
- Usable output file:

## Run B — [same headings]

## Decision
- Primary for next project + deciding constraint:
- Fallback + when I switch:
- Recheck date:

The watched field (refresh, never freeze)

Maintain a list, not a leaderboard: Higgsfield, ByteDance Seedance, Kling, Google Flow/Veo, Runway, Sora, Hailuo/MiniMax, Pika, Luma, open HunyuanVideo-style workflows, and current avatar/editor environments (HeyGen, Synthesia, Captions, CapCut and equivalents). Tool names are refreshable current examples — recheck interfaces, model names, pricing, and regional availability at lab time via official pages. Record the date of every check. No invented results: if you did not run it, write "not tested."

Rules that keep the bench honest

• Same brief, same reference, same reviewer, same phone check.
• Judge winners by the beat's job (59.1), not by demo-reel beauty.
• Generate short, select hard, log everything — settings, seeds, seeds/versions, cost, time.
• Re-run the bench when your job changes (new product, new region, new pricing) rather than defending an old winner.

Verify fast

Cover the tool names in your VIDEO-TOOL-TEST.md. Does the decision paragraph still convince — citing your hardest constraint with a file and a score? If the argument needs the brand name to work, it is hype, not evidence.