# UAIX Totem & Taboo Memory Spec

- Version: v3.129.0
- Source route: /uaix-totem-taboo-memory-spec/
- JSON export: /uaix-totem-taboo-memory-spec.json/
- Markdown export: /uaix-totem-taboo-memory-spec.md/
- Evidence packet JSON: /evidence-packets/uaix-totem-taboo-memory-spec.json/
- Evidence packet Markdown: /evidence-packets/uaix-totem-taboo-memory-spec.md/
- Evidence packet HTML: /evidence-packets/uaix-totem-taboo-memory-spec.html/

## Core correction

The Totem and Taboo update should not be framed as literal filesystem locking. The Teleodynamic correction is stronger: the files are high-meaning governance anchors with a high bar to change. They may evolve through explicit human-reviewed role-update payloads, source rationale, before/after diff review, validation, and archive receipts. They must not drift through short-term task memory.

## Totem selection

I selected `.uai/totem.uai` to contain the UAIX.org positive attractor: lane identity, memory-package objective, portable-evidence design heuristics, resource-bounded philosophy, public tone, and the high-bar change protocol. These are the concepts the package must preserve across recursive handoffs.

## Taboo selection

I selected `.uai/taboo.uai` to contain the negative perimeter: no runtime execution, no claim widening, no cross-domain authority merger, no consciousness/autopoiesis/safety-proof claims, no credential or private-network action, and no anchor rewrite from volatile memory. These are the states where the agent must no-op and request review.

## File-based long-term memory

The package now separates volatile memory from belief memory:

- `.uai/totem.uai` — high-meaning positive anchor.
- `.uai/taboo.uai` — high-meaning negative boundary.
- `.uai/long-term/README.uai` — long-term file-memory index.
- `.uai/long-term/core-beliefs/teleodynamic-ai.uai` — core belief memory.
- `.uai/long-term/source-briefs/uaix-totem-taboo-source-synthesis.uai` — source-backed synthesis.
- `.uai/short-term-memory.uai` — volatile next-pass state.

## Hard boundaries

- No runtime AI behavior
- No model training
- No live crawling beyond ordinary public GET checks
- No endpoint execution
- No credential validation
- No private-network probing
- No automation of publication
- No automation of verification
- No automation of completion
- No automation of packet coverage
- No automation of final sign-off
- No automation of final sign-off publication
- No GET-Action execution
- No POST actions
- No safety certification
- No consciousness claim
- No biological-equivalence claim
- No exact symbolic translation claim
- No medical/therapy claim
- No command-and-control claim over other domains