PFC-Coherent Induction Substrate v2

Engineering analogue of prefrontal-cortex cognitive control for human–LLM idea-to-artifact pipelines. Production-ready: SDK + CLI + HTTP API + Docker.

Role context = top-down bias. ConflictMonitor = ACC analogue. ValuePolicyEvaluator = vmPFC/OFC analogue. MemoryRecombiner = hippocampal recombination analogue. ReleaseGate = fail-closed evidence ledger.

---

Loop

RAW IDEA
  → GoalWorkingMemory       (lateral PFC)
  → RoleBiasController      (dlPFC top-down bias)
  → HypothesisCompiler      (claim_status + falsifier)
  → MemoryRecombiner        (hippocampal recombination)
  → ConflictMonitor         (ACC conflict signal)
  → InhibitoryNoiseGate     (semantic-noise suppression)
  → ArtifactConstructor     (action policy)
  → ValuePolicyEvaluator    (vmPFC/OFC value gate)
  → ReleaseGate             (fail-closed PASS / BLOCK / FAIL)
  → EvidenceLedger          (sha256 audit trail)

Hard non-claims

• no literal human–AI fusion
• no LLM consciousness
• no unmeasured neural synchrony
• no medical inference
• no 10× claim without benchmark

Claim status

Claim	Status
PFC as engineering inspiration	VALID_ANALOGY
Role context as top-down bias	TESTABLE_MODEL
ACC-inspired conflict monitor	VALID_ANALOGY
vmPFC-inspired value policy	VALID_ANALOGY
Hippocampus-inspired recombination	VALID_ANALOGY
PFC as exact model of LLM orchestration	REJECTED
Literal brain–AI synchronization	REJECTED
Improved artifact quality	REQUIRES_BENCHMARK

Quickstart

Python SDK

from pfc_ci import PFCInductionEngine, RawIdea, MemoryTrace, default_goal_memory

engine = PFCInductionEngine()
result = engine.run(
    raw_idea=RawIdea(text="Build a PFC-inspired loop"),
    goal_memory=default_goal_memory(),
    traces=(MemoryTrace(source="pfc", content="PFC bias", relevance=0.9),),
    candidate_output=(
        "role context required evidence artifact_tests conflict_report "
        "value_score safety_verdict evidence_ledger"
    ),
)
print(result.ledger.final_verdict, result.ledger.value_score)

CLI

pip install -e ".[dev]"
pfc-ci info
pfc-ci --seed 42 repro
pfc-ci run --idea "ship a PFC loop" --json

HTTP API (FastAPI)

uvicorn pfc_ci.api.app:app --port 8080
curl http://127.0.0.1:8080/healthz
curl -X POST http://127.0.0.1:8080/api/v1/run \
  -H 'Content-Type: application/json' \
  -d '{"idea":"loop","candidate":"role context required evidence artifact_tests conflict_report value_score safety_verdict evidence_ledger"}'
curl http://127.0.0.1:8080/metrics | head

Docker

docker build -t pfc-ci:local .
docker run -p 8080:8080 --rm pfc-ci:local

Benchmark + dashboard

PYTHONPATH=src python -c "
from pfc_ci.benchmark.runner import run_default_suite
from pfc_ci.dashboard.render import render_dashboard_html
r = run_default_suite()
open('benchmark_report.html','w').write(render_dashboard_html(r))
print(r.to_dict())
"

Quality gates (all green)

Gate	Tool	Status
Lint	ruff check	clean
Format	ruff format --check + black --check	clean
Types	mypy --strict	clean (57 source files)
Tests	pytest	70+ tests, 100% green
Coverage	pytest --cov-fail-under=95	≥96%
Benchmark	5-case suite	accuracy 1.00, unsupported_claim_rate 0.00

Architecture

See docs/architecture.md. ADRs in docs/adr/.

Plugin system

Components are discoverable via importlib.metadata entry points (pfc_ci.compilers, pfc_ci.evaluators, …). Reference example in examples/plugin_example/.

Bibliography

1. Miller & Cohen, 2001 — An integrative theory of prefrontal cortex function. Annu. Rev. Neurosci.
2. Miller & Constantinidis, 2024 — Timescales of learning in prefrontal cortex. Nat. Rev. Neurosci.
3. Friedman & Robbins, 2022 — Prefrontal cortex in cognitive control. Neuropsychopharmacology.
4. Botvinick, Cohen, Carter, 2004 — Conflict monitoring and ACC. Trends Cogn. Sci.
5. Bartra, McGuire, Kable, 2013 — Valuation system meta-analysis. NeuroImage.
6. Lockwood et al., 2024 — vmPFC and prosocial motivation. Nat. Hum. Behav.
7. Weilbächer & Gluth, 2016 — Hippocampus + vmPFC in memory-based decisions. Brain Sciences.
8. Preston & Eichenbaum, 2013 — Hippocampus and PFC in memory. Curr. Biol.

License

MIT. See LICENSE.