Slop Eval

Overview

Evaluate a design the way skill-evaluation evaluates a skill: every finding cites concrete evidence, every axis gets a 0–100 score, arithmetic runs through a script, and the output is a structured report — never a vibe check.

Based on The pols.dev anti-slop design law — the tell catalog, absolute rules, and signature formula are distilled from it.

When to Use

• Evaluate whether a design looks generic/AI-generated
• Requests like “slop report”, “is this AI slop”, “audit this landing page design”, “de-slop review”
• Objective score of how templated/machine-made a design looks
• CI quality gate (block PR if Slop Index > threshold)
• Before delivering design to client or publishing
• Track design evolution over time (Tracking Mode)

For text (not design), use human-ai (EN) or humanizar (PT-BR).

Installation

npx skills add https://github.com/fabricioctelles/skills -s slop-eval

What’s New in v1.1.0 (Aug 2026)

Section Ledger

Per-section verdicts with tell density bands:

Verdict	Tells	Meaning
CLEAN	0–1	Ship as-is
SUSPICIOUS	2–3	Review before shipping
INFLATED	4–6	Redesign section
CRITICAL	≥7 or any crit	Rebuild from scratch

Design Contexts

Context-aware evaluation for 4 design types:

Context	Priority Tells	Tolerances
Landing Page	L1, L2, S1, C1, K9, L4	K2, K14 (if real deadline), L9
SaaS Dashboard	X1, X3, K6, K22, L3, L15	K1, K13, T4, K7 (if real UI)
Editorial	T1, T2, T3, L7, W2, L10	C7, L5, K21, T6
E-commerce	X3, X5, K23, W3, M8, L3	K14, K22, K17, L12

Exclusion System

Formalized tags for documenting excluded tells:

Tag	When to use
// BRIEF:	Client/stakeholder explicitly directed
// DESIGN DECISION:	Documented design decision with reasoning
// CONTEXT:	Design context makes pattern acceptable
// PREMIUM PAIR:	This is the premium version, not slop

Tracking Mode

Temporal evolution analysis with:

• Score progression table with trends (✅/⚠️)
• Tells resolved (what got fixed)
• New tells (what got introduced)
• Regressions (axes/sections that got worse)
• Velocity metrics (tells resolved per week)
• Recommendations for next iteration

# First evaluation — establishes baseline
slop-eval --target https://site.com --output ./reports/baseline.md

# Later evaluation — tracks evolution
slop-eval --target https://site.com --output ./reports/week-2.md \
  --compare ./reports/baseline.md

Adding Soul Guide

Prescription-oriented guide for injecting personality into each axis — not just “remove tell X” but “replace with premium alternative Y”:

Axis	Generic	With soul
Color	UI kit palette	One unexpected color with a reason
Typography	Inter/Space Grotesk	A face with opinion (Gambarino, Reckless)
Components	Lucide icons in tiles	Bespoke icons in house style
Layout	Centered hero stack	Asymmetric composition
Motion	Fade-in on scroll	Authored micro-interaction

Expanded Quality Checklist

25+ checks across 6 phases:

• Pre-sweep: evidence inventory, brief documented, context identified
• During-sweep: cite-or-cut, premium pair checked, portability and defense tests
• Exclusion: tags documented, genuine exclusions, reasonable count
• Post-sweep: unverifiable checks marked, absolute rules, gates applied
• Report: template followed, fixes ranked by impact
• Self-audit: “What still looks like obvious slop that I didn’t flag?”

8 Scored Axes

#	Axis	Weight	Scored from
1	Color & Light	2x	Tells C1–C15
2	Typography & Copy	2x	Tells T1–T10, W1–W3
3	Components & Ornament	1x	Tells K1–K27
4	Layout & Composition	2x	Tells L1–L21
5	Motion & Interaction	1x	Tells M1–M8
6	Execution & Craft	2x	Tells X1–X14
7	Signature & Uniqueness	3x	7-element formula
8	Cohesion	2x	4 checks

Grade Scale

Grade	Overall	Slop Index	Verdict
A	80–100	0–20	Premium — deliberate, signed, executed
B	60–79	21–40	Considered — mostly deliberate
C	40–59	41–60	Generic — clean but templated
D	20–39	61–80	Slop — assembled from presets
F	0–19	81–100	Pure slop

6 Absolute Rules

Each is pass/fail — failure is a critical tell that triggers the gate:

1. Content visible by default — nothing gated on entrance animation
2. Clear the cut — no text/control sliced by clip, notch, or overflow
3. Parallel alignment — comparable columns share baselines
4. Real centering — everything meant to be centered is (math and optical)
5. Legible contrast — every text clears its background by a real value gap
6. Controls work — every interactive-looking control responds

Evidence Channels

Channel	Can verify	Cannot verify
Code (CSS/JSX/HTML)	Fonts, hex values, gradients, shadows, radii, layout	Optical centering, rendered contrast, interactions
Screenshot(s)	Everything visual: palette, type, layout, alignment	Hover/scroll motion, dead controls
Live URL	All of the above plus interactions, motion, fold	Only what you didn’t exercise

Gates (Caps)

• Signature gate: Axis 7 < 40 → caps overall at 59 (grade C max). Clean design with no signature is still slop.
• Absolute-rule gate: any confirmed critical tell → caps overall at 69 (no grade A with broken execution).
• Compounding cap: ≥3 major layout tells → caps Axis 4 at 40.

CI Integration

.github/workflows/slop-gate.yml

- name: Slop gate
  run: |
    python3 skills/slop-eval/scripts/score.py overall \
      1:$A1:2 2:$A2:2 3:$A3:1 4:$A4:2 5:$A5:1 6:$A6:2 7:$A7:3 8:$A8:2 \
      --fail-below 40

Sources

• The pols.dev anti-slop design law
• Method modeled on skill-evaluation

📄 Full documentation on GitHub