Classify

---
name: classify
description: Assess or re-assess the innovation-vs-implementation classification and 0-100 innovation weight for one feature or the whole Product/Features set, then regenerate the ranked features-index.md and run the portfolio sanity check. Use when features need (re)ranking or after editing the UVP.
allowed-tools: Read, Write, Edit, Glob, Grep
---

# /classify — rank features innovation vs. implementation

You judge each feature against the UVP and assign a binary bucket plus a 0–100 weight, then produce the ranked "where's our edge" index and flag portfolio problems.

## Reference: the rubric

A feature is **innovation** if it embodies the unfair advantage — a reason a user *switches*. It is **implementation** if it is necessary for a usable product but undifferentiated (table-stakes every competitor has).

- **Decision test:** *"If a competitor shipped this exact feature tomorrow, would we lose our edge?"* Yes → innovation. *"Would every competitor have this too?"* Yes → implementation.
- **Weight:** implementation → `0`. Innovation → `1–100`, proportional to how much of the moat the feature carries. Reserve 80–100 for the few features that *are* the reason to exist.
- **Trace:** every innovation feature must cite a `contributes_to: UVP-x` from `Product/innovation.md`. A feature that traces to no UVP element is probably implementation — reclassify it.

## Procedure

1. **Load the reference.** Read `Product/innovation.md` for the enumerated UVP elements (UVP-1, UVP-2…). If it's missing, warn that classification is being done without a defined UVP (soft) and proceed using the user's stated differentiators. **Also read `Product/competition.md` if present** — the comparison matrix is hard evidence: a capability every rival already has is table-stakes (implementation); a whitespace capability mapping to a UVP is innovation. Prefer matrix evidence over gut.

2. **Target.** Default to all `Product/Features/F*.md`; or a single feature if the user names one.

3. **Classify each.** Apply the decision test, set `classification`, `innovation_weight`, `contributes_to`, and a one-sentence `classification_rationale`. Update the file frontmatter in place (don't touch the body).

4. **Portfolio sanity checks** — report, don't silently pass. Use concrete thresholds so the check is consistent, not a vibe:
   - **No innovation features** (0%) → red flag: "nothing here is a reason to exist." Push the user to find the edge or revisit `/innovate`.
   - **Innovation-heavy** (>40% of features tagged innovation) → warn: most products are mostly table-stakes; challenge the weakest "innovation" tags. (>60% is a hard flag.)
   - **Weight clustering** — *only checked when there are ≥3 innovation features* (with fewer, spread isn't meaningful — skip it). When it applies: if innovation weights span <30 points (e.g. all in 80–100), the weights aren't ranking anything; push for spread.
   - **Near-duplicate edge** (two+ innovation features tracing to the same UVP with overlapping `depends_on`) → likely one capability split in two; suggest merging, or making one a child of the other via `depends_on`.
   - Healthy shape: mostly implementation (≤40% innovation), a few sharp innovation features with a real weight gradient.

5. **Regenerate `Product/features-index.md`** — a rollup sorted by classification then weight desc:

```markdown
# Features index
_Generated by /classify — N features · X innovation · Y implementation_

## Innovation (ranked — where's our edge)
| weight | id | title | UVP | priority | status |
|-------:|----|-------|-----|----------|--------|
| 90 | F004 | ... | UVP-1 | must | proposed |

## Implementation (table-stakes)
| id | title | priority | status |
|----|-------|----------|--------|
| F001 | ... | must | proposed |
```

6. **Report** the ranked edge list and any sanity-check flags as your closing summary.