# Video Quality — rubric, honest grade, and the roadmap to better
*The thing that was missing: a defined bar for good vs bad, an honest read of where we are, and the path to improve. 2026-06-08.*

## The uncomfortable truth first
**We have no real performance data, because we don't post yet.** Quality is ultimately judged by the platform — watch-time, completion, **sends**, saves. Until we post, every "this is good" is a **proxy**. So the single highest-leverage move for *knowing* if our videos are good is **wiring posting → getting real metrics.** Everything below is how we judge well *in the meantime* and improve deliberately.

**The proxies we have right now** (and should weight, in order):
1. **Ian's likes/saves** — the strongest real signal we have. (You liked: the 3 punchy speech reels — showup/keepgoing/hard — and ~11 of the stock+text reels b01-b06/b14/c01/c03-05/c07-10. You saved 16 break-out TikToks as reference.)
2. **Break-out patterns from the 5,554-scrape** — what actually performs for *others* in our niche (duo doing the activity, short relational caption + share-command, lo-fi, belonging emotion, tiny-account monster reach).
3. **My self-score on the rubric** — consistent, not vibes.

## The rubric (every video scored 1–5; hook is weighted ~2×)
| # | Dimension | The question |
|---|---|---|
| 1 | **Hook (0–2s)** ⭐ | Does the first frame + first line *stop the scroll*? |
| 2 | **Niche resonance** | Does it speak to the activity-partner / loneliness→connection wedge in *their* language + pain points? |
| 3 | **Retention shape** | hook → turn → payoff; no dead air; the open loop pays off |
| 4 | **Share trigger** | Would you *send it to someone*? ("this is so me / [name] needs this") |
| 5 | **Authenticity** | Feels real/lo-fi, not like an ad |
| 6 | **Caption craft** | Punchy, readable, muted-proof; written for the eye (≠ the spoken track) |
| 7 | **Audio** | Natural voice / good sound / trending-able (not robotic) |
| 8 | **Visual** | Clean/cinematic, good pacing, on-brand accent |
| 9 | **Postability** | Can we actually post it? original/cleared vs borrowed = the legal gate |
| 10 | **On-brand** | Recognizable voice + the product wedge, every time |

## Honest grade — where we actually are
**Best (A-):** the **VO reels** (`vo-c01`, `vo-b14`, `vo-b05`) — your liked footage + a real ElevenLabs voice + the "partner who shows up" wedge, *and brand-safe*. And the **speech reels** (`speech-showup/keepgoing/hard`) — real human voice + cinematic b-roll + emotional payoff, *you liked 3 of them* — A- on craft, **but C on postability** (borrowed speaker audio → legal gate).

**Solid baseline (B+):** the **stock+text reels** (`b01-b06`, `c01-c10`) — strong hooks, clear niche message, brand-safe — but **silent** (no audio) and text-heavy.

**Weak / study-only (C or below):** the **L2 compilation** (you called it "really bad" — correct), the **L1/L3 restyles** (de-text smears, borrowed footage), the early **kinetic-text** reels (text-only, no audio), and the **robotic-voice** ones (now replaced).

**The pattern of what's GOOD (and matches your taste):** a clear *emotional/relatable message* + a *real voice* + *cinematic visuals* + a *punchy hook*, tied to the **"showing up / find your people"** wedge.
**What's WEAK:** silent text-only · borrowed-footage restyles · generic compilations · robotic audio · anything that reads like an ad.

## The 3 best right now (for the landing)
1. **`vo-c01`** — *"the magic was never the workout… it's someone who actually shows up, every single week."* Your liked footage + real voice + dead-on the wedge + **postable**. Our best all-rounder.
2. **`speech-keepgoing`** — *"quitting guarantees you fail… one foot in front of the other keeps you in the game."* The semantic-clip engine's flagship — real voice, cinematic, emotional. You liked it.
3. **`speech-showup`** — *"keep showing up… even if you don't see results right away."* The core theme, punchy, you liked it.
*(Range: one postable VO-over-our-footage + two engine speech reels you liked.)*

## The roadmap to consistently better
**P0 — end the guessing:** wire **posting** (Postiz/Metricool) → real metrics. *Nothing else tells us the truth.*
**P1 — make our best format postable:** the **legal-clean speech lane** — sources that *want* spread (creators/sermons/CC) + **licensed b-roll** (Pexels/Epidemic) → our A- speech reels become things we can actually post.
**P2 — craft compounding wins (in order):** real voice on everything (done) → **hook A/B** (the first line is the highest-leverage variable) → **music bed** → **original/AI-regen footage** (needs a video key) → **carousels** (the highest-engagement format we have *zero* of).
**P3 — close the loop:** every batch → self-score on the rubric + your thumbs + (once posting) real metrics → **make more of what wins, kill what doesn't.**

## How we'll know we're improving
- Short term: **your like-rate** on new batches climbs; rubric self-scores rise; fewer study-only pieces.
- Real term (post-posting): **send-rate and completion** on the platforms — the only grades that count.