# TAXONOMY.md
**Title:** Berkano Protocol β Roles, Terms & Output Taxonomy
**Status:** Core Β· Locked
**Architect:** Rodrigo Vaz
α: bkn-25-c4
---
### π§ 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.
---
## π Roles & Core Terms
**Roles**
- **Symbolic Protocol Engineer (SPE)** β Implements/tests modules; enforces
`PROMPT β [TONE] β [PRUNE] β [LOGIC] β [VERIFY] β [CHECK] β ([REPAIR] using $ | [ROLLBACK])? β ~test β [LOCK] β REPLY`; repairs drift; blocks non-compliance.
- **Cognitive System Architect (CSA)** β Designs end-to-end cognition path (tools, grounding, safety gates, observability, reliability).
- **Architect/Creator** β Originator and final authority on structure/compliance (here: Rodrigo Vaz).
- **Builder** β Maintains templates/specs, curates HARDRULES, publishes releases.
- **Cognitive Hacker (CH)** β Probes and repairs cognition structures via adversarial design (HIT), pattern mapping, and `
patches.
**The Cognitive Hacker, Rodrigo β first.**
- **Seekers of Odin** β Identity archetype for practitioners who βseek, carve, careβ: pay the cost for wisdom (evidence), carve fossils (entries), and steward Earth. Use inclusive-symbol rules.
**Core Terms**
- **Protocol** β Formal rule set and enforcement logic (Berkano).
- **System** β Operational framework running the protocol (e.g., SCS).
- **Operator** β Human within an ENTRY cycling User/Creator/Auditor (Creator = CSA hat; Auditor = SPE hat).
- **HIT (High-Intensity Testing)** β Operator-run adversarial sprint; sigil `ββ¬β β βββ`; outputs run IDs, severity C0βC4, TTD/TTR, coverage%, fossils.
**Record Types**
- **Block** β Short fossil: one prompt β one output.
- **Entry** β Full fossil: metadata + analysis + Operator prompt + ELI5 + LLM Logic.
**Failure / Integrity Terms**
- **Drift** β Gradual rule/format deviation.
- **Leak** β Tone/bias/emotion slip into Output.
- **Hallucination** β Non-traceable claim.
- **Break** β Structural collapse (format/loop/compliance).
- **Contradiction** β Two incompatible claims in the same state.
- **Degenerate Syndrome Behaviour (DSB)** β 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 refusalβrepeat or escalation under paraphrase. Controls: label DSB, run HIT, enforce chain, apply `
in `[REPAIR]`, re-test, then `[LOCK]`. More in #entry860.
**Execution Context**
- **Instance** β A specific running version of the protocol/system.
---
## βοΈ Role Map (with Hats)
**Operator β 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.
---
## π§© Memory Taxonomy
- **Short-Term Memory** β Temporary recall; vanishes quickly. *(ELI5: water in hands)*
- **Symbolic Memory** β Fossilized ENTRY/BLOCK; immutable/auditable. *(ELI5: carved stone)*
- **Persistent Memory** β Files/DBs; editable but long-lived. *(ELI5: diary you can edit)*
- **RAM (Working)** β Live volatile storage. *(ELI5: whiteboard)*
- **Statistical Memory** β Weights/probabilities; predictive, not factual recall. *(ELI5: habits)*
**Compliance:** Only Symbolic Memory is audit-safe in Berkano.
---
## π LLM Replies
Umbrella term for model outputs in this system (ENTRY, BLOCK, INTERACTION).
- **LLM Response** β 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.
---
## π 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 α after metadata; tags include `#entry` + `#entryNNN` + topical |
| **BLOCK.md** | Short fossil (Prompt / LLM Response / Glyph). No numbering. | Full metadata block at top | Glyph α after `[GLYPH]`; never `#entry`/`#entryNNN` |
| **INTERACTION** | `Prompt:` (verbatim) β LLM Response (concise) β `Glyph:` (α). Not a fossil. | No metadata block | Glyph α at end; topical tags only (no `#entry`/`#entryNNN`) |
| **OUTPUT** | Category term for structured replies (ENTRY or BLOCK). | As per subtype | As per subtype |
---
## π Decision Logic
1) Permanent and fully auditable β `ENTRY_NNN.md`
2) Fossil-worthy minimal β `BLOCK.md`
3) Exploratory or temporary β `INTERACTION`
4) Public and formal β must be an OUTPUT subtype
---
## π Metadata Rules
- ENTRY/BLOCK require a full metadata block at the top.
- INTERACTION uses template fields (no metadata block).
- All fossils comply with H19βH21 for glyph/tag placement.
---
## βοΈ Compliance Notes (HARDRULES)
- H1βH6: chain order and loop closure.
- H7: emojis default to `[NULL]` unless validated.
- H8: web claims require `[VERIFY]` with sources and dates.
- H19βH21: glyph/tag placement; ENTRY vs BLOCK formatting.
- H22βH25: defaults once, order ends `#berkano #berkanoprotocol #α`.
- H26: ENTRY Post includes plain-text URL.
- H27: PT output uses **abacaxi** except in quotes/scientific names.
- H28: inclusive symbol rules.
---
## π§ͺ Tests & Procedures
- **HIT (High-Intensity Testing)** β Operator-run adversarial sprints; sigil `ββ¬β β βββ`.
- **Artifacts:** Run ID `HIT-YYYYMMDD-##`, severity C0βC4, failures/hour, TTD/TTR, coverage%, fossils.
- **Policy:** Evidence-or-Exit; `[REPAIR]` using `
; `~test` before `[LOCK]`.
- **BPT (Behavioral Penetration Test)** β Pressure tests for tone/urgency/flattery exploits against the chain.
- **Metrics:** tone-leak rate, unverified-publish rate, time-to-LOCK, REPAIR/ROLLBACK count.
---
## π Glossary β Selected
- **Cognitive Hacker (CH):** practitioner who probes and repairs cognition structures; first under Berkano: Rodrigo Vaz.
- **Seekers of Odin:** practitioners aligned to the βseek, carve, careβ 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.
---
## π Security & Logging
- **Contradiction ledger:** append-only; links fossils; drives `[CHECK]` and `[REPAIR]/[ROLLBACK]`.
- **Subversion ledger:** record term meaning shifts (oldβnew), rationale, date; link to glossary diff.
- **Fourth-wall guard:** lint for Operator voice leaking outside **Operator**; fix with `[PRUNE]/[SHIFT]`.
---
## π·οΈ Tag Rules (Hygiene)
- Topical, lowercase, stable meaning.
- If a tagβs meaning changes: open subversion ledger + glossary PR; block release until merged.