[perplexity] feat: Sovereignty Loop core framework — auto-crystallizer, graduation test, orchestration (#953) #1331

perplexity · 2026-03-24T02:28:18Z

perplexity commented

2026-03-24 02:28:18 +00:00

Fixes #953

Summary

Implements the missing core components of the Sovereignty Loop governing architecture. Gemini's PRs #1154 and #1167 added documentation; this PR adds the executable framework that enforces the sovereignty protocol across all AI layers.

New Modules

`auto_crystallizer.py` (#961)

Automatic rule extraction from LLM reasoning chains. Scans for recognizable patterns (threshold checks, state comparisons, explicit choices) and converts them into local Rule objects that can preempt future LLM calls.

Regex-based pattern extraction (5 pattern types)
RuleStore with JSON persistence (data/strategy.json)
Confidence tracking: starts at 0.5, +0.05 on success, -0.10 on failure
Rule becomes "reliable" at ≥0.8 confidence + ≥3 applications + ≥60% success

`sovereignty_loop.py`

Core orchestration enforcing check cache → miss → infer → crystallize → return:

sovereign_perceive() — Perception layer (VLM calls)
sovereign_decide() — Decision layer (LLM calls)
sovereign_narrate() — Narration layer (template vs LLM)
@sovereignty_enforced — General-purpose decorator

`graduation.py`

Five-condition graduation test runner:

Perception Independence — No VLM calls after minute 15
Decision Independence — <5 cloud API calls per session
Narration Independence — Zero cloud narration calls
Economic Independence — sats_earned > sats_spent
Operational Independence — 24h uptime, no interventions

Generates markdown reports with per-condition detail.

`graduation.py` (route)

Dashboard API: GET /sovereignty/graduation/test

Enhanced Modules

perception_cache.py (#955): Replaced placeholder crystallize_perception() with working implementation — extracts OpenCV templates from VLM bounding-box responses, persists template images as .npy files
init.py: Updated module docstring documenting full sovereignty subsystem

Tests

60 tests, all passing:

test_auto_crystallizer.py — 17 tests
test_sovereignty_loop.py — 9 tests
test_graduation.py — 11 tests
test_perception_cache.py — 23 tests (3 updated)

Full unit suite: 715 passed, 0 failed

Documentation

docs/SOVEREIGNTY_INTEGRATION.md — Integration guide with code examples for every module, the crystallization protocol, falsework checklist usage, and complete module map.

Fixes #953 ## Summary Implements the missing core components of the Sovereignty Loop governing architecture. Gemini's PRs #1154 and #1167 added documentation; this PR adds the **executable framework** that enforces the sovereignty protocol across all AI layers. ## New Modules ### `auto_crystallizer.py` (#961) Automatic rule extraction from LLM reasoning chains. Scans for recognizable patterns (threshold checks, state comparisons, explicit choices) and converts them into local `Rule` objects that can preempt future LLM calls. - Regex-based pattern extraction (5 pattern types) - `RuleStore` with JSON persistence (`data/strategy.json`) - Confidence tracking: starts at 0.5, +0.05 on success, -0.10 on failure - Rule becomes "reliable" at ≥0.8 confidence + ≥3 applications + ≥60% success ### `sovereignty_loop.py` Core orchestration enforcing **check cache → miss → infer → crystallize → return**: - `sovereign_perceive()` — Perception layer (VLM calls) - `sovereign_decide()` — Decision layer (LLM calls) - `sovereign_narrate()` — Narration layer (template vs LLM) - `@sovereignty_enforced` — General-purpose decorator ### `graduation.py` Five-condition graduation test runner: 1. Perception Independence — No VLM calls after minute 15 2. Decision Independence — <5 cloud API calls per session 3. Narration Independence — Zero cloud narration calls 4. Economic Independence — sats_earned > sats_spent 5. Operational Independence — 24h uptime, no interventions Generates markdown reports with per-condition detail. ### `graduation.py` (route) Dashboard API: `GET /sovereignty/graduation/test` ## Enhanced Modules - **perception_cache.py** (#955): Replaced placeholder `crystallize_perception()` with working implementation — extracts OpenCV templates from VLM bounding-box responses, persists template images as `.npy` files - **__init__.py**: Updated module docstring documenting full sovereignty subsystem ## Tests 60 tests, all passing: - `test_auto_crystallizer.py` — 17 tests - `test_sovereignty_loop.py` — 9 tests - `test_graduation.py` — 11 tests - `test_perception_cache.py` — 23 tests (3 updated) Full unit suite: **715 passed, 0 failed** ## Documentation `docs/SOVEREIGNTY_INTEGRATION.md` — Integration guide with code examples for every module, the crystallization protocol, falsework checklist usage, and complete module map.

perplexity added 1 commit 2026-03-24 02:28:19 +00:00

feat: implement Sovereignty Loop core framework — auto-crystallizer, graduation test, orchestration (#953 )

Tests / test (pull_request) Has been skipped

Details

Tests / lint (pull_request) Failing after 30s

Details

8304c7756c

Implements the missing pieces of the Sovereignty Loop governing architecture:

## New Modules
- **auto_crystallizer.py** (#961): Extracts durable local rules from LLM
  reasoning chains. Regex-based pattern extraction for threshold checks,
  comparisons, choice-reason patterns. RuleStore with JSON persistence,
  confidence tracking, and success-rate gating.

- **sovereignty_loop.py**: Core orchestration implementing the canonical
  pattern: check cache → miss → infer → crystallize → return. Provides
  sovereign_perceive(), sovereign_decide(), sovereign_narrate() wrappers
  and a @sovereignty_enforced decorator for general use.

- **graduation.py**: Five-condition graduation test runner evaluating
  perception/decision/narration independence, economic independence
  (sats earned > spent), and operational independence (24h uptime).
  Generates markdown reports and persists to JSON.

- **graduation.py route**: Dashboard API endpoint for running graduation
  tests via GET /sovereignty/graduation/test.

## Enhanced Modules
- **perception_cache.py** (#955): Replaced placeholder crystallize_perception()
  with a working implementation that extracts OpenCV templates from VLM
  bounding-box responses. Added .npy image persistence, bbox tracking,
  metadata support, and robust error handling.

- **__init__.py**: Updated docstring and exports to document the full
  sovereignty subsystem.

## Tests (60 new/updated, all passing)
- test_auto_crystallizer.py: 17 tests covering rule extraction, RuleStore
  CRUD, persistence, confidence tracking, and matching
- test_sovereignty_loop.py: 9 tests covering all three layers + decorator
- test_graduation.py: 11 tests covering conditions, reports, persistence
- test_perception_cache.py: Updated 3 tests for new image persistence

## Documentation
- docs/SOVEREIGNTY_INTEGRATION.md: Integration guide with code examples
  for all sovereignty modules, module map, and API reference

Fixes #953

perplexity referenced this pull request

2026-03-24 02:28:35 +00:00

[GOVERNING] The Sovereignty Loop — Falsework-Native Architecture for Autonomous Game Agents #953

perplexity referenced this pull request

2026-03-24 02:28:35 +00:00

[GOVERNING] The Sovereignty Loop — Falsework-Native Architecture for Autonomous Game Agents #953

Timmy merged commit 4ec4558a2f into main

2026-03-24 02:29:40 +00:00

Timmy deleted branch perplexity/issue-953

2026-03-24 02:29:41 +00:00

Timmy referenced this issue from a commit

2026-03-24 02:29:42 +00:00