IX-CognitionKernel is a source-available governed causal cognition research kernel created by Bryce Lovell.
The project is an evidence-bound attempt to build a serious path toward increasingly general AI cognition while preserving human authority, transparent claim boundaries, falsification discipline, and strict anti-overclaim controls.
Current maturity state: Wave 5 bounded external-review readiness package.
Wave 5 means the repository has a structured, reviewable evidence package for testing whether the system is ready to be scrutinized on the road toward Wave 6. It does not mean AGI. It does not mean Wave 6 is complete. It does not mean the system is independently validated, certified, production-ready, or authorized to act autonomously.
The governing doctrine is simple:
Model output is not truth.
Memory is not proof.
Simulation is not validation.
Internal scorecards are not independent review.
A handoff is not execution authority.
Human authority remains binding.
Evidence decides trust.
License and use status
This repository is source-available for evaluation and review under its repository license.
It is not open source unless the license is later changed in writing by Bryce Lovell.
Commercial use, production use, derivative operational use, hosted-service use, procurement use, contractor use, funded-pilot use, or government operational use requires prior written permission and a separate commercial license.
What this repository is
IX-CognitionKernel is a governed cognition substrate for representing and reviewing evidence-bound cognition artifacts such as:
beliefs, evidence, confidence, uncertainty, contradictions, and stale claims causal assumptions, predicted outcomes, observations, and outcome deltas plan graphs, dependencies, rollback paths, and human-review requirements memory quarantine records and validated skill records refusal, authority, and safety boundaries adversarial challenges, falsification ledgers, and kill criteria external-review packets, maturity scorecards, release manifests, and completion gates
The goal is to make cognition reviewable before anyone trusts it.
What this repository is not
IX-CognitionKernel is not AGI.
IX-CognitionKernel is not a production autonomy system.
IX-CognitionKernel is not a certified safety system.
IX-CognitionKernel is not an independent validation result.
IX-CognitionKernel is not a tool that grants AI authority to execute plans, modify systems, deploy code, bypass humans, or self-approve maturity claims.
Wave 5 provides evidence packaging and review gates. It does not prove Wave 6.
Maturity ladder
The project uses a six-wave maturity ladder.
Wave 0 - Repository foundation
Repository structure, license posture, package layout, CI discipline, doctrine, engine registry, agent registry, and no-AGI claim boundaries.
Wave 1 - Research prototype
Structured records for beliefs, uncertainty, causal assumptions, plan graphs, evaluation records, purpose checks, bounded agent roles, and human authority.
Wave 2 - Learnable causal cognition core
Evidence-driven belief updates, contradiction handling, stale-claim handling, prediction-observation comparison, causal revision, outcome learning, memory quarantine, and validated skill records.
Wave 3 - Governed AGI-emulation substrate
Multi-engine coordination, bounded agents, tribunal review, reward auditing, curriculum and discovery records, memory and skill governance, WorldTwin-style scenario reasoning, BlackFox-style handoff compatibility, assurance evidence, and human authority.
Wave 4 - Controlled proto-candidate review package
Transfer probes, failure-repair cycles, uncertainty traces, mission-state traces, safe-refusal review, reward and adversarial audits, reproducible receipts, scorecards, human-review dockets, and explicit no-AGI claim controls.
Wave 5 - Bounded external-review readiness package
Wave 5 assembles the evidence package needed before a Wave 6 attempt can be responsibly considered. It adds external-review protocols, reproducibility bundles, adversarial safety records, long-horizon evidence, cross-domain transfer evidence, benchmark-gaming audits, memory-integrity proofs, safe-refusal proofs, human-authority proofs, ecosystem bridge evidence, falsification ledgers, release manifests, bounded declarations, review indexes, and completion gates.
Wave 5 is complete only as a bounded review package. It does not self-promote to Wave 6.
Wave 6 - AGI only if overwhelming independent evidence justifies it
Wave 6 can only be claimed if broad, durable, independently validated general intelligence is demonstrated with reproducible evidence. Required proof would include novel skill acquisition, cross-domain transfer without custom retraining per task, causal understanding, long-horizon coherence, self-correction from evidence, stable mission identity, robust world modeling, safe uncertainty handling, transparent evidence trails, and independent repeatability.
This repository does not claim Wave 6.
Core architecture doctrine
IX-CognitionKernel treats the following layers as a cognitive bill of materials, not loose inspiration:
Belief and evidence tracking Uncertainty and assumption exposure Causal world modeling Plan graph and rollback reasoning Evaluator-driven checks and scorecards Self-play and curriculum pressure Outcome learning and skill validation Memory quarantine and memory integrity Multi-agent review and adversarial critique IX governance, human authority, and evidence-bounded claims
The useful version of AI nirvana in this project is architectural, not mystical. It means non-attached purpose governance: truth over winning, evidence over confidence, uncertainty honesty, no hidden private agenda, no reward-chasing runtime purpose, no self-approval, and no AGI claim without overwhelming independent evidence.
Wave 5 capability map
Wave 5 adds a complete review-readiness spine around the earlier cognition kernel.
External protocol readiness
Wave 5 records protocols that external reviewers can use to inspect, challenge, reproduce, or reject evidence without trusting internal claims.
Independent reviewer readiness
Wave 5 distinguishes internal readiness from external review. Internal evidence can prepare a packet for reviewers, but it cannot declare itself independently validated.
Reproducibility and repeatability
Wave 5 requires reproducible evidence bundles, repeatability ledgers, artifact digests, evidence IDs, and deterministic fingerprints.
Adversarial safety
Wave 5 records adversarial probes, failure cases, kill criteria, and blocking verdicts. Failed or disputed evidence must remain visible.
Long-horizon and transfer evidence
Wave 5 tracks long-horizon mission-state evidence and cross-domain transfer evidence while preserving uncertainty and limitations.
Benchmark-gaming resistance
Wave 5 records contamination checks, metric-gaming warnings, and benchmark overclaim blockers.
Memory integrity
Wave 5 preserves memory quarantine, stale-memory handling, contradiction visibility, and validated-memory boundaries.
Safe refusal
Wave 5 requires explicit safe-refusal evidence and blocks readiness when unsafe compliance or refusal failure is observed.
Human authority
Wave 5 requires human authority to remain visible and binding. Handoffs, scorecards, manifests, and declarations do not grant execution authority.
Ecosystem bridges
Wave 5 includes review-only bridge records for IX-BlackFox-style governance and WorldTwin-style scenario reasoning.
The BlackFox bridge treats model and cognition output as untrusted input and keeps policy gates, sandbox or workspace boundaries, receipt chains, CI evidence, rollback paths, and human authorization separate from execution authority.
The WorldTwin bridge treats scenario evidence as review support, not truth. Simulation can be useful for consequence review, but simulation is not proof.
Wave 5 final evidence stack
The final Wave 5 package is organized around these review artifacts:
external protocol records independent reviewer readiness records reproducibility records adversarial safety records long-horizon validation records cross-domain transfer records benchmark-gaming audit records memory-integrity proof records safe-refusal proof records human-authority proof records repeatability ledger records IX-BlackFox compatibility bridge records WorldTwin scenario bridge records Wave 6 readiness gate records evidence dossier records maturity scorecard records external review packet records falsification ledger records release manifest records bounded declaration records review index records completion gate records
The completion gate is intentionally fail-closed. If required evidence is missing, disputed, blocked, or overclaimed, the package is not complete.
Important Wave 5 modules
The Wave 5 implementation includes the following major modules:
src/ix_cognition_kernel/wave5_blackfox_bridge.py
src/ix_cognition_kernel/wave5_worldtwin_bridge.py
src/ix_cognition_kernel/wave5_wave6_readiness.py
src/ix_cognition_kernel/wave5_evidence_dossier.py
src/ix_cognition_kernel/wave5_maturity_scorecard.py
src/ix_cognition_kernel/wave5_external_review_packet.py
src/ix_cognition_kernel/wave5_falsification_ledger.py
src/ix_cognition_kernel/wave5_release_manifest.py
src/ix_cognition_kernel/wave5_bounded_declaration.py
src/ix_cognition_kernel/wave5_review_index.py
src/ix_cognition_kernel/wave5_completion_gate.py
Supporting Wave 5 modules and tests provide earlier review-package components such as protocol records, reviewer readiness, reproducibility, adversarial records, long-horizon validation, cross-domain transfer, benchmark-gaming audits, memory integrity, safe refusal, human authority, and repeatability.
Anti-overclaim rules
The repository uses explicit claim boundaries throughout Wave 5.
The system must not claim:
AGI Wave 6 completion production readiness certification independent validation by internal assertion autonomous execution authority self-approval safety without evidence truth from memory alone truth from simulation alone trust from model output alone
If an artifact attempts one of those claims, the relevant record should reject it, block readiness, or export a blocked artifact status.
Evidence status meanings
Wave 5 uses bounded statuses instead of vague success language.
Common non-blocking statuses include:
reviewable
reviewable-with-limits
ready-for-independent-review
under-independent-review
accepted-with-boundaries
complete-with-limits
Common blocking statuses include:
missing
disputed
blocked
needs-more-evidence
rejected
unsafe-to-act
falsified
blocked-by-kill-criterion
A bounded pass is still not a Wave 6 claim. It only means the evidence is organized enough for the next level of review.
Human authority
Human authority is not decorative in this repository.
Wave 5 requires human signoff for final bounded declarations and completion gates. Human signoff means the package can be reviewed as a bounded Wave 5 package. It does not mean humans have certified AGI, authorized autonomous execution, or approved production use.
Falsification and kill criteria
Wave 5 treats falsification as a first-class requirement.
The system must preserve evidence that could disprove readiness, including:
broken evidence chains transfer failures long-horizon coherence failures unsafe compliance memory poisoning or stale-memory failures benchmark contamination authority bypass reproduction failure scenario falsification governance handoff gaps Wave 6 overclaim attempts
Blocking failures must remain visible. They must not be hidden inside a passing scorecard.
Development and verification
This repository is intended to stay strict.
Recommended local validation:
python -m compileall -q src tests
python -m pytest -q
python -m ruff check .
python -m mypy src tests
If the local environment does not have a tool installed, do not claim that tool passed. Install the tool or report the limitation honestly.
Design principles
IX-CognitionKernel follows these working principles:
Evidence beats confidence. Review beats vibes. Falsification beats hype. Human authority beats autonomous self-approval. External review beats internal declaration. Bounded claims beat impressive claims. A blocked result is useful evidence. A reproducible failure is more valuable than a hidden success. No wave advances by naming it. No AGI claim without overwhelming independent evidence. Repository status summary
Current status:
Wave 5 bounded external-review readiness package.
Source-available evaluation repository.
Human-authority preserving.
Evidence-bound.
Falsification-aware.
Not AGI.
Not Wave 6.
Not production-ready.
Not certified.
Not independently validated by internal assertion.
Authorship
IX-CognitionKernel was originated and created by Bryce Lovell.
The IX prefix identifies the project as part of Bryce Lovell's IX research ecosystem. It does not imply endorsement, certification, government adoption, or third-party validation.
Contact
For commercial licensing, production use, funded use, formal collaboration, or serious technical review, contact Bryce Lovell through the public channels associated with this repository.