ZPE Diagram

What This Is

Bounded structural-with-style diagram encoding. Graph geometry, visual style, and draw order survive within the declared validation scope.

The product claim is the structural-geometry base / bounded-style fiber separation: exact geometry reconstruction, exact color and stroke-width preservation, and explicit draw-order encoding — on every in-scope case in the public proof packet. 1.000 structural, style, and draw-order fidelity (6/6 cases); 1.000 rejection rate (3/3 out-of-scope probes). Useful now, improving continuously.

The current public evidence surface is limited to:

Codec Mechanics

Key Metrics

Source: validation/results/bounded_style_validation.json, regenerable via V_02.

Repo Identity

Readiness

Honest Blocker

We do not claim fill support; We do not claim dashed support on the encode path; We do not claim taper, pressure variation, or broader authorial-style recovery.

What We Prove

• Exact geometry reconstruction on every in-scope case in the public proof packet (structural_exact_worst = 1.000, 6/6 cases). Proof anchor: validation/results/bounded_style_validation.json.
• Exact color and stroke-width preservation on every in-scope case (style_exact_worst = 1.000, 6/6 cases). Proof anchor: validation/results/bounded_style_validation.json.
• Exact draw-order preservation on every in-scope case (stroke_order_exact_worst = 1.000, 6/6 cases). Proof anchor: validation/results/bounded_style_validation.json.
• Every out-of-scope input (fill, palette escape, dashed stroke) is rejected at encode time (reject_probe_rejection_rate = 1.000, 3/3 probes). Proof anchor: tests/test_style_authority.py.
• Structural and style axes are non-aliasing: same geometry/different style → style codes differ; different geometry/same style → geometry codes differ; same elements/different order → order codes differ. All separation metrics = 1.0.

What We Don't Claim

• No fill encoding. Fill inputs are explicitly rejected.
• No palette escape. Only the frozen 8-color palette is in scope; out-of-palette colors are rejected.
• No dashed stroke encoding. Dashed inputs are rejected at encode time.
• No arbitrary SVG coverage. The in-scope surface is line-based SVG with the frozen palette and solid strokes.
• No Compass-8 product claim. Any 8-direction technique is internal implementation, not product claim (Compass-8 posture: NO, per V2 §7.2).
• No compression claim. ZPE-Diagram is a structural-fidelity codec, not a compression codec.
• No claims outside the 6 in-scope synthetic SVG cases currently in the public proof packet.

Verification Status

Proof Anchors

Repo Shape

Quick Start

python3 -m venv .venv
. .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"
python proofs/artifacts/reproduce_validation.py
python -m pytest tests/test_style_authority.py

Upcoming Workstreams

This section captures the active lane priorities — what the next agent or contributor picks up, and what investors should expect. Cadence is continuous, not milestoned.

• Real-world SVG corpus expansion — Active Engineering. Expand corpus from 6 synthetic SVGs to: Mermaid output (~30 cases), Graphviz output (~30 cases), one open UI icon set (~50 cases). Re-run existing fidelity test harness. Compass-8 NO posture preserved.

Scope

See SCOPE.md for the plain-language product boundary.

Citation

Use CITATION.cff for software citation metadata.