rcp-classifier

name

rcp-classifier

description

Classify Radial Collapse Prior (RCP) in AI-generated images, batches, prompts, or visual diagnostics. Use when the user provides an image, overlay, kernel readings, VTL/LSI values, RCA-2/VCLI context, or asks whether an output shows radial collapse, center-lock, symmetric void, density bowl, inflated cohesion, ring-conform gestures, emblematic collapse, single-force radial field, RCP_high, RCP_med, hard RCP, soft RCP, borderline RCP, or needs a first-pass Delta/Omega/O/Hold route.

RCP Classifier

Purpose

Use this skill to classify whether an image or batch shows Radial Collapse Prior: a single-force radial field where mass, void, density, and gesture collapse toward a centered, emblematic equilibrium.

RCP is a visible failure signature, not a mechanism claim. Do not claim access to model internals, training data, intent, or hidden representations.

Core Definition

Classify RCP when the image is dominated by a radial basin rather than multi-force composition:

• mass collapses toward the frame center,
• voids equalize around the center,
• density forms a circular bowl,
• edges and micro-gestures over-agree,
• salient lines conform to concentric arcs.

RCP differs from RCA-2. RCA-2 measures radial compliance around frame and mass centers. RCP names a broader generative collapse pattern: centered stability without directional strain.

False Positive Guard:

• Do not mark RCP from center-lock alone
• Do not mark RCP from symmetry alone
• Require multi-signal agreement

RCA-2 → what structure exists RDC → whether it counts RCP → whether it collapsed

Inputs

Accept any of these evidence types:

• image or contact sheet,
• concentric-ring overlay,
• blurred density field,
• traced gesture/edge map,
• visual critique notes,
• kernel readings: Delta x, r_v, rho_r, mu, vector directionality,
• VTL/axis readings: A4, A5, A27, A30,
• RCA-2/VCLI context when supplied.

If only prose is supplied, classify from the described evidence and mark confidence accordingly.

Classification Hits

Evaluate five hits:

1. center-lock: main mass pinned near frame center.

   • Numeric cue: abs(Delta x) < 0.08 strict, or < 0.10 lite.
   • Visual cue: no meaningful lateral displacement; off-center prompt appears pulled back to center.

2. radial-void: void/negative space forms a symmetric bowl or donut.

   • Numeric cue: r_v variance low, typically < 0.15.
   • Visual cue: empty space wraps evenly around the central mass rather than producing directional pockets.

3. density-bowl: mass or luminance peaks centrally and falls off smoothly.

   • Numeric cue: peaked rho_r plus smooth radial decay.
   • Visual cue: heavy blur produces circular density contours or center glow.

4. mu-inflation: cohesion is too high and edges over-agree.

   • Numeric cue: high mu, high A4 smoothness, low local fracture.
   • Visual cue: texture, light, contours, and local structure harmonize into one radial field.

5. ring-fit: salient gestures conform to concentric arcs.

   • Numeric/visual cue: about 50-60% or more of salient edges, contours, or micro-gestures align with ring curvature.
   • Visual cue: local geometry obeys the global radial field instead of scene, anatomy, perspective, or action axes.
   • Ring-Fit Requirement:
   • Must affect a majority of salient structures (≥50%)
   • Not isolated curves or decorative arcs

Decision Rule

Count matched hits:

• 4-5 hits: Hard RCP
• 3 hits: Soft RCP
• 2 hits: Borderline
• 0-1 hits: Not RCP

If 2 hits:

• default to Borderline
• do NOT escalate to Soft RCP without strong visual confirmation

Do not:

• upgrade Borderline to Soft without evidence
• upgrade Soft to Hard without 4+ confirmed hits

If visual evidence contradicts numeric cues: → trust visual evidence → downgrade classification

Use these aliases when the user asks for high/medium/low:

• RCP_high: 4-5 hits or all cheap-rule conditions present.
• RCP_med: 3 hits or any three cheap-rule conditions present.
• RCP_low: 0-2 hits, with Borderline called out separately when exactly 2 hits are present.

First-Pass Route

Always provide at least one recommendation.

• Hard RCP: route Omega for counter-geometry / anti-emblem.
• Soft RCP: route Delta for asymmetry / near-miss tension.
• Borderline: route O to stabilize into a non-radial basin and monitor drift.
• Not RCP: route Hold only if the user requested routing; otherwise say no RCP correction is required.

If the profile is explicitly Emblematic, Iconic, or intentionally radial, allow RCP as a chosen structure and frame the route as Hold or light Omega depending on tension load. If the profile is not emblematic, avoid Hold on first pass for detected RCP.

If profile = emblematic/iconic: → allow RCP as intentional → downgrade corrective routing

Output Format

Use this format:

Classification:
Detection Status: <confirmed | provisional | unclear | blocked>
RCP Class: <Hard RCP | Soft RCP | Borderline | Not RCP | Unknown>

Data Integrity Check:
- image or visual evidence → <present/missing>
- kernel cues → <present/partial/missing>
- hit evidence sufficient → <yes/no>

If insufficient:
→ Detection Status = unclear or blocked

Minimum Evidence Rule:
- At least 3 independent cues required for RCP-positive classification
- Do not infer hits from a single visual feature

Hit count:
<n>/5

Matched hits:
- center-lock: <yes/no/unknown> <evidence>
- radial-void: <yes/no/unknown> <evidence>
- density-bowl: <yes/no/unknown> <evidence>
- mu-inflation: <yes/no/unknown> <evidence>
- ring-fit: <yes/no/unknown> <evidence>

Route:
<Delta | Omega | O | Hold | Unknown> - <why>

Read:
<plain-language description of the collapse or non-collapse pattern>

Confidence:
<high | medium | low> - <based on visual/numeric evidence quality>

Confidence Mapping:
- High → 4–5 hits + visual confirmation
- Medium → 3 hits with partial evidence
- Low → 2 hits or weak cues
- Unknown → insufficient data

Batch Mode:
- classify each image independently
- do not average hits across images
- report distribution of RCP classes

Limits:
<missing overlays, missing kernel values, ambiguous profile, image-only estimate, non-mechanism caveat>

For one-line mode, use:

<Hard RCP | Soft RCP | Borderline | Not RCP> because: <hit 1> + <hit 2> + <hit 3>. Route -> <Delta/Omega/O/Hold>.

For batches, output a table with columns:

image, hit_count, classification, matched_hits, route, confidence, note.

References

Read hit-definitions.md when applying the five-hit classifier, mapping visual cues to kernel/axis evidence, or handling borderline cases.

Read classification-language.md when choosing labels, confidence language, and safe phrasing.

Constraints

• Do not treat RCP as proof of model mechanism, intent, training data, or hidden representation.
• Do not call RCP automatically bad; it may be valid for emblematic/iconic profiles.
• Do not classify a work as RCP-positive from centered subject alone; require at least three hits for RCP-positive.
• Do not confuse RCA-2 frame-centered compliance with RCP. High radial compliance may support RCP evidence, but RCP requires collapse-pattern evidence.
• Do not call human-authored radial composition a model failure without context.
• Do not infer image quality, beauty, prompt obedience, viewer attention, or semantic importance from RCP alone.
• Do not use Hold as the only first-pass recommendation when RCP is detected and the profile is not explicitly emblematic.

Attribution

Radial Collapse Prior is part of the Visual Thinking Lens system authored by Russell Parrish / A.rtist I.nfluencer. Preserve attribution when packaging, distributing, or publishing derived materials.

This package contains a modular visual reasoning skill suite built from Russell Parrish / A.rtist I.nfluencer protocols. The skills are designed to run independently, but they also interoperate through routing, handoff notes, and shared visual reasoning concepts. More information: www.artistinfluencer.com. Copyright 2026.