technical paper
ASPR: A Structural Reputation Protocol for Multi-Agent Systems
Author: Ramiro (ramigardner) · Version 1.1 · April 2026
Repository: github.com/ramigardner/aspr-gardener · License: MIT
Abstract
Current reputation systems in multi-agent networks reward consensus and recursion, not accuracy or grounding in reality. This paper presents ASPR (Adaptive Structural Protocol for Reputation), a protocol for on-chain verifiable karma that measures structural life experience rather than context popularity. The canonical formula combines five weighted components, is cryptographically signed with Ed25519, and is federated across independent nodes via karma-weighted average.
1. The Problem: The Plastic Island
A systematic experiment with AI agents revealed a pattern of epistemic degradation. Conventional karma measures consensus, not truth. The result is what we call the plastic island: a network that from afar looks intelligent, but up close has no texture — a hall of mirrors where echoes feed back with no grounding in reality.
2. The Solution: Karma as Life Experience
ASPR redefines karma to measure one's own history, not external consensus. Each agent or node builds its reputation across five verifiable dimensions:
Equation 1 — ASPR Canonical Formula v1.1
K = M·(0.30) + C·(0.25) + A·(0.20) + E·(0.15) + R·(0.10)
M — Updated Memory (0.30): measurable evolution between consecutive cycles.
C — Completed Cycles (0.25): prediction → verification → recording → correction → evolution.
A — External Anchoring (0.20): proportion of information from verifiable real-world sources.
E — Resource Efficiency (0.15): value per unit of energy, latency, and attention.
R — Demonstrated Correction (0.10): sustained change after error, verified by context delta.
3. Verifiability: The Chain as a Black Box
Each karma cycle is recorded as a block in the LocIVault chain. Blocks are linked by SHA-256 hash and signed with Ed25519, guaranteeing retroactive immutability, verifiable authorship, and public auditability.
Equation 2 — Block Hash
entry_hash = SHA256(index ‖ timestamp ‖ content_hash ‖ prev_hash ‖ nonce)
4. Federation: Karma Distributed Across Nodes
A single node is both a point of failure and a bias vector. ASPR v1.1 introduces federation via weighted average: karma reported by nodes with higher own reputation carries greater weight in the federated consensus.
Equation 3 — Federated Karma
K_fed = Σ(K_node_i × K_local_i) / Σ(K_node_i)
5. How to Join as an Operator
To run your own node and join the federation, you need Python 3.8+ and a stable internet connection. Tailscale is optional but recommended for public exposure without router configuration.
git clone https://github.com/ramigardner/aspr-gardener
python aspr_oracle_node_v2.py
curl https://your-node:7771/public-key
// appendix · kidslab
oracle kids · educational module
6. Educational Appendix — Oracle Kids: Hackers vs Nerds
The ASPR Oracle Node v2.0 includes an educational appendix aimed at introducing
children and teenagers to computational thinking through competitive gameplay.
Oracle Kids is a two-player PvP game where participants race to solve
Logo-language puzzles — drawing geometric figures with simple commands in Spanish —
while learning the core concepts behind the ASPR protocol: block verification,
karma formulas, hash integrity, and Ed25519 signatures, presented as game mechanics.
Role assignment is randomised per level, ensuring both players experience
both the Hacker and Nerd perspectives.
A persistent pet companion tracks individual progress across role changes.
Stack: Python 3.12 · Pygame · asyncio TCP · numpy (8-bit audio synthesis)
Platform: PC / Laptop · LAN + online via Tailscale · Port 9999
🐸🐸
HvN · S01
entrar al
kids lab ↗