Grading in ReviewHub¶
ReviewHub is where human taste enters the system. The engine can generate, render, and measure — but it cannot know what good looks like; that knowledge comes only from grades. Every grade you give steers what the engine makes next, so it's worth knowing exactly what each grade does.
The grades¶
| Grade | Meaning | What the engine does |
|---|---|---|
| ⭐ Outlier | commercial winner — would sell | clones it (Lab priority 1), captures its DNA as vocabulary |
| 🔥 Catalyst | unique spark worth evolving | evolves it (fresh garments around the idea) |
| 💚 On Brand | meets the bar | validated-bounds seed; keeps the lane open |
| 🚫 Off Brand | violates the brand | extracts kill phrases → future prompts fence them out |
| 💀 Ugly | render failure / unusable | negative learning only |
| Delete | remove from rotation | excluded everywhere |
Image level vs garment level¶
- Quick buttons / O-C-B-X-U keys grade the whole image.
- Clicking a garment (in the detail view) grades that garment on that model — the highest-signal input: learning lands on the garment file and the Lab knows which outfit slot won.
- The Outfit 1/2/3 button rows submit image-level grades (legacy behavior of the prod ReviewHub build).
What to expect after grading¶
Within seconds: the grade appears in Strategist → Overview; winners join the Lab's seed queue (Lab tab); off-brand language shows up in Learning → register as new bans. The next batch already reflects it.