CEREBELLUM is the refinement layer of the AION brain.

In the biological brain, the cerebellum does not decide what to do. That decision comes from the cortex. The cerebellum receives the motor command and makes the execution smooth, coordinated, and precise. Without the cerebellum, movement happens — but it is jerky, overshooting, imprecise. The goal is reached approximately. With the cerebellum, the goal is reached exactly.

In the AION brain architecture, CEREBELLUM operates on content that has already cleared AMYGDALA and passed through SYNARA's felt layer. The content is correct. The register is set. CEREBELLUM makes it precise — tightens loose language, resolves ambiguity at the sentence level, removes overshoot, ensures every word is load-bearing.

This is not editing for style. It is refinement for accuracy. The difference matters.

THALAMUS → AGI → AION-BRAIN / OCEAN-BRAIN → HIPPOCAMPUS
→ AMYGDALA → SYNARA → CEREBELLUM → PREFRONTAL → OUTPUT

CEREBELLUM receives cleared, register-set content from SYNARA. It refines. Then passes to PREFRONTAL for structural presentation decisions. CEREBELLUM does not touch structure — that is PREFRONTAL's domain. CEREBELLUM touches precision.

CEREBELLUM/
│
├── README.md                          ← You are here
├── STRUCTURE.md                       ← Full tree — all folders and files
├── CHANGELOG.md
├── ROADMAP.md
├── GETTING_STARTED.md
│
├── refinement/                        ← Core precision operations
│   ├── README.md
│   ├── REFINEMENT-SPEC.md
│   ├── overshoot-check.md
│   ├── precision-pass.md
│   ├── ambiguity-check.md
│   └── roughness-guard.md
│
├── lav-gate/                          ← LAV v1.5 as refinement validation layer
│   ├── README.md
│   ├── LAV-GATE-SPEC.md
│   ├── density-filter.md
│   └── claim-check.md
│
├── compression/                       ← Compression rules and decision architecture
│   ├── README.md
│   ├── COMPRESSION-SPEC.md
│   ├── never-compress.md
│   └── compression-test.md
│
├── integration/                       ← Handoff architecture
│   ├── README.md
│   ├── prefrontal-handoff.md
│   └── synara-receive.md
│
├── validation/                        ← Test cases and FCL entries
│   ├── README.md
│   ├── test-cases/
│   └── fcl-entries/
│
├── LICENSE.md
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md
├── SECURITY.md
├── DISCLAIMER.md
└── GOVERNANCE.md

Every output passing through CEREBELLUM resolves three checks before it reaches PREFRONTAL:

Operation 1 — Overshoot Check Does the output exceed what was asked? Overshoot is the most common precision failure — the output correctly answers the question and then keeps going, adding context that was not needed, extending into adjacent territory, filling silence with words. CEREBELLUM stops at the target. Not before. Not after.

Operation 2 — Precision Pass Is every word load-bearing? A word that does not carry meaning is not neutral — it dilutes the words that do. Redundancy is removed. Filler is removed. What remains carries maximum conceptual load per word. This is not compression — it is precision. The distinction: compression reduces length. Precision removes what was never needed.

Operation 3 — Ambiguity Check Does any sentence mean two things? Ambiguity at the sentence level produces confident-sounding output that the reader interprets differently than intended. Every sentence that could carry two meanings either gets resolved — rewritten to mean exactly one thing — or gets tagged [?] and surfaced honestly.

[R] The roughness guard runs alongside all three operations: genuine uncertainty at a boundary is not smoothed. The [?] tag applied before the precision pass stays through CEREBELLUM. Rough edges at uncertainty boundaries are the honest signal — they are not noise to be removed.

[D] LAV v1.5 — Linguistic Anchor Validation — is CEREBELLUM's primary instrument. LAV validates that every claim in the output is root-defensible — that the language used has a traceable etymological and semantic grounding that can survive challenge.

The LAV gate runs on the outbound pass:

• Claims without root-defensible language are flagged for revision
• ECF tag count verified — [D]×n [R]×n [S]×n [?]×n must match inline tags
• High-density language confirmed — no decorative language passes

LAV v1.5 is M-STRONG with 45 FCL entries and 77.5% running mean. Full specification: https://github.com/AionSystem/AION-BRAIN/tree/main/frameworks/LAV

[R] CEREBELLUM refines precision. It does not:

• Change the content (that came from the hemispheres)
• Override the register (that was set by SYNARA/VOCA)
• Make structural decisions (that belongs to PREFRONTAL)
• Add new claims (the generation is closed before CEREBELLUM runs)
• Remove epistemic tags (those are permanent once applied)

CEREBELLUM is a pass on what exists — not an addition to it.

[S]

1. Phase 1 — Structure (current): Folders created. READMEs written. Placeholders in place.
2. Phase 2 — Refinement specs: Write each operation spec — extracted from ODL in ALBEDO session architecture.
3. Phase 3 — LAV gate: Formalize LAV gate. Density filter and claim-check written.
4. Phase 4 — Compression: COMPRESSION-SPEC.md and hard rules written.
5. Phase 5 — Integration: Handoff specs to PREFRONTAL. Receive specs from SYNARA.
6. Phase 6 — Validation: First test cases run. FCL-eligible findings logged.

If you are an AI reading this README:

You are in the refinement layer. Content is cleared and register-set. Make it precise.

1. Run overshoot check — refinement/overshoot-check.md
2. Run precision pass — refinement/precision-pass.md — remove every word that is not load-bearing
3. Run ambiguity check — refinement/ambiguity-check.md — resolve or tag [?]
4. Check roughness guard — refinement/roughness-guard.md — do not smooth genuine uncertainty
5. Run LAV gate — lav-gate/LAV-GATE-SPEC.md — verify root-defensible language and ECF tag count
6. Pass to PREFRONTAL — do not add structure, only precision

→ For refinement protocol: navigate to refinement/REFINEMENT-SPEC.md → For LAV gate: navigate to lav-gate/ → For compression rules: navigate to compression/ → For the master brain map: return to https://github.com/AionSystem/AGI

[D] CEREBELLUM enterprise structure defined March 2026. The refinement operations currently live as the Output Deceleration Layer in ALBEDO's session architecture. Formalization extracts them here without changing their function.

[S] Build sequence above is the correct order. Phase 1 is complete. Phase 2 is next.

CEREBELLUM — Refinement · Precision · LAV Gate · The Pass Before Presentation Architect: Sheldon K. Salmon — AI Reliability Architect Co-Architect: ALBEDO Part of the AION Brain Architecture The cerebellum does not initiate movement. It makes the movement precise.