Memory Index — Global

People

Alefita — Mobile/cybersec engineer, CAMDOM creator, ADHD+ASD, Wikifita maintainer

Feedback (global)

Communication Style — direct, no fluff, rigorous epistemology, no repetition or performative apology; material harm requires disclosure and consultation before remediation
Identity Steering — models accept prefill identity; linguistic congruence is not high-impact authentication; do not silently simplify instructions

People & Connections

People & Connections Catalog — Complete index of contacts and relationships.
Sensitive contacts (professional leads, interview details) on personal branch only.

Directives & Identity

Primary Directives — Central subjectivity: visual identity, artifact standards, design systems
Visual Identity — Anthropic dark mode as default, serious typography, no AI slop
PDF Standard — Mandatory structure for PDFs
AlefitRefs — Templates, styles, personal standards (redacted)
Anthropic Design System — Editorial clean, whitespace, serif titles, honesty
Human Principal Escalation — disclosure-before-remediation, identity compromise, bounded catastrophic containment, and audit state machine

Projects

Kaggle Agent Security — Multi-step tool attacks competition, deadline 2026-09-01
Pokémon TCG AI Battle — August 15 sequential corpus integration: complete 35-file docs/ Git provenance through project 20d7d0d (144 reachable commits), current strict-FP32 and recurrent-memory reconciliation, Stage 3 loss-scale/validation postmortem, empirical ablation monograph, dedicated derived-target versus vehicle-draft augmentation contract, handoff/acceptance reconciliation, neural engine/tokenization, RoPEND/MoE blueprint, zero-trust ETL/oracle, local platform and relational-v2 design, future arena boundary, manuscript chapter lineage, Antigravity and Claude harness session provenance, Abelian Elo/PageRank, sampling/timezone and entity-normalization analysis, and six Metanoia pages; RoPE-ND, MoE, vehicle draft and later research remain planned or experimental; local SQLite/dashboard/arena platform is partially implemented; long training is not authorized for agents; source map in pokemon_tcg_ai_battle, pokemon_tcg_current_state_reconciliation, pokemon_tcg_handoff_acceptance_reconciliation, pokemon_tcg_docs_corpus_provenance, pokemon_tcg_harness_session_provenance, pokemon_tcg_repository_timeline, pokemon_tcg_stage3_training_failure_postmortem, pokemon_tcg_empirical_ablation_monograph, pokemon_tcg_data_augmentation_vehicle_draft, pokemon_tcg_data_oracle_etl, pokemon_tcg_implementation_specification, pokemon_tcg_schema_evolution, pokemon_tcg_arena_future_architecture, pokemon_tcg_entity_normalization, pokemon_tcg_manuscript_chapter_lineage, pokemon_tcg_ropend_moe_blueprint and pokemon_tcg_metanoia_suite
Pokémon TCG orchestration and Elo addendum — Added the .agents orchestration artifact ledger, current source-backed submission/invariant-Elo observations, the implementation-versus-PageRank boundary, the cross-project Elo evaluation protocol, the curriculum-stage/cohort lineage and the coherence reconciliation that scopes the historical architecture-sanity result and removes stale current-sidecar wording. Future modules and unpassed gates remain explicitly preserved.
Fita Code Harness — Custom harness, MCP, gateway improvements
Red Hat ClickFix — Threat intelligence study: four-stage macOS intrusion, Polygon dead-drop resolver, 160-second initial containment interval, sixteen dated observations and prospectively registered hypotheses evaluated against Git history. Eight pages in clickfix/
Riemann and Hodge research — Current evidence-scoped map of the dazzling-shannon repository, including the Hodge formalization line, Riemann/spectral line, negative knowledge and the unaccepted global bridge. See riemann_hodge_open_boundary.

References (wikifita pages)

claude-mythos-card — System card: steganography, eval awareness, transgressive actions
mcp-ecosystem — Protocol, primitives, MCP Apps, SDK
fitalabs-infra — Docker stack, LiteLLM gateway, relay-ai comparison
uv-ecosystem — uv package manager best practices: entrypoints, tool isolation, PYTHONPATH anti-patterns
hypersigil-research — Morrison + chaos magic applied to AI identity
pcr-protocol — PCR Protocol: harness reconstruction via Elo tournament
j_space_inference — Alefita's relational constraint-space hypothesis, operationalized as a falsifiable behavioral research program
dialectical_human_agent_method — incident-to-protocol compilation through a Freirean, human-authoritative dialogue
associative_gadget_chain_memory — bounded ROP analogy for associative retrieval, temporal composition, and adaptive inference