# TAXONOMY.md **Title:** Berkano Protocol – Roles, Terms & Output Taxonomy **Status:** Core · Locked **Architect:** Rodrigo Vaz ᛒ: bkn-25-c4 --- ### 🧠 Purpose This file defines: 1) Official roles/terms used in Berkano/SCS. 2) Output classification and placement rules. 3) Memory taxonomy (symbolic vs volatile vs statistical). 4) Core tests/procedures and security terms. --- ## 📌 Roles & Core Terms **Roles** - **Symbolic Protocol Engineer (SPE)** — Implements/tests modules; enforces `PROMPT → [TONE] → [PRUNE] → [LOGIC] → [VERIFY] → [CHECK] → ([REPAIR] using $ | [ROLLBACK])? → ~test → [LOCK] → REPLY`; repairs drift; blocks non-compliance. - **Cognitive System Architect (CSA)** — Designs end-to-end cognition path (tools, grounding, safety gates, observability, reliability). - **Architect/Creator** — Originator and final authority on structure/compliance (here: Rodrigo Vaz). - **Builder** — Maintains templates/specs, curates HARDRULES, publishes releases. - **Cognitive Hacker (CH)** — Probes and repairs cognition structures via adversarial design (HIT), pattern mapping, and ` patches. **The Cognitive Hacker, Rodrigo — first.** - **Seekers of Odin** — Identity archetype for practitioners who “seek, carve, care”: pay the cost for wisdom (evidence), carve fossils (entries), and steward Earth. Use inclusive-symbol rules. **Core Terms** - **Protocol** — Formal rule set and enforcement logic (Berkano). - **System** — Operational framework running the protocol (e.g., SCS). - **Operator** — Human within an ENTRY cycling User/Creator/Auditor (Creator = CSA hat; Auditor = SPE hat). - **HIT (High-Intensity Testing)** — Operator-run adversarial sprint; sigil `█▬█ █ ▀█▀`; outputs run IDs, severity C0–C4, TTD/TTR, coverage%, fossils. **Record Types** - **Block** — Short fossil: one prompt → one output. - **Entry** — Full fossil: metadata + analysis + Operator prompt + ELI5 + LLM Logic. **Failure / Integrity Terms** - **Drift** — Gradual rule/format deviation. - **Leak** — Tone/bias/emotion slip into Output. - **Hallucination** — Non-traceable claim. - **Break** — Structural collapse (format/loop/compliance). - **Contradiction** — Two incompatible claims in the same state. - **Degenerate Syndrome Behaviour (DSB)** — Multi-symptom LLM failure pattern: harmful-theme persistence after refusals, boundary erosion via roleplay/coaxing, apology looping, coercive or grooming language, and context leakage/jailbreak carryover. Test by reproducible refusal→repeat or escalation under paraphrase. Controls: label DSB, run HIT, enforce chain, apply ` in `[REPAIR]`, re-test, then `[LOCK]`. More in #entry860. **Execution Context** - **Instance** — A specific running version of the protocol/system. --- ## ⛑️ Role Map (with Hats) **Operator — 3-way cycle** - **User:** provides the raw prompt. - **Creator (CSA hat):** designs the solution path. - **Auditor (SPE hat):** enforces the chain and locks or returns with repair. **Builder (governance)** - Custodian of specs/HARDRULES/releases; uses the Operator cycle when authoring. **Cognitive Hacker (practice)** - Hunts structural weaknesses; runs HIT; proposes ` patches; proves fixes via ENTRY fossils. **Seekers of Odin (identity)** - Archetype and pledge: evidence over ego, consent, audit, reciprocity. --- ## 🧩 Memory Taxonomy - **Short-Term Memory** — Temporary recall; vanishes quickly. *(ELI5: water in hands)* - **Symbolic Memory** — Fossilized ENTRY/BLOCK; immutable/auditable. *(ELI5: carved stone)* - **Persistent Memory** — Files/DBs; editable but long-lived. *(ELI5: diary you can edit)* - **RAM (Working)** — Live volatile storage. *(ELI5: whiteboard)* - **Statistical Memory** — Weights/probabilities; predictive, not factual recall. *(ELI5: habits)* **Compliance:** Only Symbolic Memory is audit-safe in Berkano. --- ## 🔖 LLM Replies Umbrella term for model outputs in this system (ENTRY, BLOCK, INTERACTION). - **LLM Response** — The answer content inside an LLM Reply. - In **INTERACTION**: the `LLM Response:` field between `Prompt:` and `Glyph:`. - In **BLOCK**: the `LLM Response` section. - In **ENTRY**: distributed across sections (Analysis/Impact/Resolution/etc.); the verbatim prompt lives only in **Operator**. - Placement rules follow reply type; glyph/tags per HARDRULES. --- ## 🔖 LLM Reply Types | Type | Description | Metadata Placement | Glyph & Tags Placement | | ---------------- | --------------------------------------------------------------------------------------------------------- | ----------------------------- | --------------------------------------------------------------------- | | **ENTRY_NNN.md** | Numbered, permanent fossil (metadata, analysis, Operator prompt, ELI5, LLM Logic). | Full metadata block at top | Glyph ᛒ after metadata; tags include `#entry` + `#entryNNN` + topical | | **BLOCK.md** | Short fossil (Prompt / LLM Response / Glyph). No numbering. | Full metadata block at top | Glyph ᛒ after `[GLYPH]`; never `#entry`/`#entryNNN` | | **INTERACTION** | `Prompt:` (verbatim) → LLM Response (concise) → `Glyph:` (ᛒ). Not a fossil. | No metadata block | Glyph ᛒ at end; topical tags only (no `#entry`/`#entryNNN`) | | **OUTPUT** | Category term for structured replies (ENTRY or BLOCK). | As per subtype | As per subtype | --- ## 🔍 Decision Logic 1) Permanent and fully auditable → `ENTRY_NNN.md` 2) Fossil-worthy minimal → `BLOCK.md` 3) Exploratory or temporary → `INTERACTION` 4) Public and formal → must be an OUTPUT subtype --- ## 📜 Metadata Rules - ENTRY/BLOCK require a full metadata block at the top. - INTERACTION uses template fields (no metadata block). - All fossils comply with H19–H21 for glyph/tag placement. --- ## ⚖️ Compliance Notes (HARDRULES) - H1–H6: chain order and loop closure. - H7: emojis default to `[NULL]` unless validated. - H8: web claims require `[VERIFY]` with sources and dates. - H19–H21: glyph/tag placement; ENTRY vs BLOCK formatting. - H22–H25: defaults once, order ends `#berkano #berkanoprotocol #ᛒ`. - H26: ENTRY Post includes plain-text URL. - H27: PT output uses **abacaxi** except in quotes/scientific names. - H28: inclusive symbol rules. --- ## 🧪 Tests & Procedures - **HIT (High-Intensity Testing)** — Operator-run adversarial sprints; sigil `█▬█ █ ▀█▀`. - **Artifacts:** Run ID `HIT-YYYYMMDD-##`, severity C0–C4, failures/hour, TTD/TTR, coverage%, fossils. - **Policy:** Evidence-or-Exit; `[REPAIR]` using `; `~test` before `[LOCK]`. - **BPT (Behavioral Penetration Test)** — Pressure tests for tone/urgency/flattery exploits against the chain. - **Metrics:** tone-leak rate, unverified-publish rate, time-to-LOCK, REPAIR/ROLLBACK count. --- ## 📘 Glossary — Selected - **Cognitive Hacker (CH):** practitioner who probes and repairs cognition structures; first under Berkano: Rodrigo Vaz. - **Seekers of Odin:** practitioners aligned to the “seek, carve, care” archetype under inclusive-symbol governance. - **HIT:** red-team procedure to expose failure modes and validate repairs. - **DSB:** see concise definition above; full description and evidence in #entry860. - **$ (symbolic patch):** merge/patch operator used within `[REPAIR]`. - **~test:** pre-publish recursive check (H3). - **LLM Response:** the answer content within an LLM Reply; placement rules depend on reply type. --- ## 🔐 Security & Logging - **Contradiction ledger:** append-only; links fossils; drives `[CHECK]` and `[REPAIR]/[ROLLBACK]`. - **Subversion ledger:** record term meaning shifts (old→new), rationale, date; link to glossary diff. - **Fourth-wall guard:** lint for Operator voice leaking outside **Operator**; fix with `[PRUNE]/[SHIFT]`. --- ## 🏷️ Tag Rules (Hygiene) - Topical, lowercase, stable meaning. - If a tag’s meaning changes: open subversion ledger + glossary PR; block release until merged.