Skip to content

NETWORK_PRECISION_GAIN.md

Latest commit

 

History

History
124 lines (90 loc) · 3.14 KB

NETWORK_PRECISION_GAIN.md

File metadata and controls

124 lines (90 loc) · 3.14 KB

Network Precision Gain

Status: deterministic proxy metric.

Network Precision Gain answers one question:

How much precision did the cooperative network add over a single answer?

It separates two claims:

  1. Measured route reward gain — already produced by the Nash-style route stability probe.
  2. Full stack precision gain — a proxy score that adds support from trace, evidence gates, living memory, reflection, depth fit, and human boundary.

Run

python scripts/run_network_precision_gain_demo.py
python scripts/run_network_precision_gain_demo.py --json

Regression test:

PYTHONPATH=.:python:python/modules python -m pytest python/tests/test_network_precision_gain_demo.py

Contributor protocol:

Current Local Result

The deterministic demo currently reports:

single baseline score:      0.1603
cooperative route score:    0.7186
full stack score:           0.8628

measured route reward gain: +0.6656
network precision gain:     +0.7025
stack added gain:           +0.1442
ratio vs baseline:          5.3824x

Interpretation:

The cooperative route already beats the single route on measured reward.
The surrounding stack adds precision by making the route traceable, gated,
adaptive, reflective, depth-aware, and human-bounded.

Components

Component Meaning
route_reward The measured LS route reward from role cooperation.
evidence_gate Support from Pythia-style evidence/action gating.
trace_integrity Support from TTM-style immutable transition trace.
adaptive_memory Support from LiminalDB-style repeatable living route memory.
reflective_clarity Support from RINSE-style interpretation over experience.
human_boundary Human goal, consent, review, and acceptance boundary.
depth_fit Whether the route ran at the right Depth Economy level.

Default weights:

route_reward       0.40
evidence_gate      0.16
trace_integrity    0.12
adaptive_memory    0.12
reflective_clarity 0.10
human_boundary     0.06
depth_fit          0.04

Why This Matters

Without this metric, the project can only say:

the cooperative route looked better

With this metric, LS can say:

the route reward improved by X
the full network precision proxy improved by Y
the stack layer added Z over raw cooperation

That gives contributors a concrete target:

  • improve evidence gating;
  • improve immutable trace linking;
  • improve route memory repeatability;
  • improve reflection clarity;
  • improve human acceptance boundaries;
  • improve depth fit.

Boundary

This metric is not a final benchmark. It is a first engineering proxy for tracking whether the stack is making repeated cooperation more precise.

Do not present it as:

  • proof of AI consciousness;
  • proof of global model superiority;
  • proof of production safety;
  • proof of formal Nash equilibrium.

The narrow claim is:

LS can measure how much more precise a cooperative route plus evidence stack is
than a single answer on the current deterministic local probe.