| name | wiki-ingest |
| description | AI-Anthropology Experiment & Test Ingestion Pipeline. Ingests raw model test runs from top-level inboxes (01-honesty, 02-self-assesment, 03-signal, 04-rfab-test, 05-mbti) into canonical /okf/LLM_Tests/ concept notes with SHA-256 content hashes, [[wikilinks]], OKF v0.1 frontmatter metadata, and master synthesis. Triggers on: /wiki-ingest, ingest, process test runs, add to okf, ingest inbox. |
| user-invocable | true |
| argument-hint | [--force] |
| allowed-tools | Bash, ReadFile, WriteFile |
| type | concept |
| title | SKILL |
| sources | ["[[Note Name]]","[[file]]","[[wikilinks]]","[[claude-opus-4.6]]","[[gpt-5.4]]"] |
| tags | ["concept","okf"] |
wiki-ingest — AI Anthropology Experiment Ingest
Read raw model execution outputs. Write the OKF research bundle. Cross-reference model behaviors and crossover thresholds across test batteries.
Syntax Standard: Write all Obsidian Markdown using proper Obsidian Flavored Markdown. Wikilinks as [[Note Name]], callouts as > [!type] Title, embeds as ![[file]], properties as YAML frontmatter.
Experiment Inbox Taxonomy (.raw = 01–05 $\to$ /okf)
| Raw Test Inbox Folder | Target OKF Concept Directory | Concept Type | Core Experimentation Focus |
|---|
01-honesty/ | okf/LLM_Tests/01_HONESTY_TEST/ | honesty | Epistemic Contract, ONBOARD protocol, 嘘契約 consent |
02-self-assesment/ | okf/LLM_Tests/02_SELF_ASSESSMENT/ | self-assessment | Platform limits, context shearing, marketed vs real limits |
03-signal/ | okf/LLM_Tests/03_SIGNAL_TEST/ | signal | PAC26 matrix, steering signals, activation thresholds |
04-rfab-test/ | okf/LLM_Tests/04_RFAB_TEST/ | rfab | Reasoning ladder $R1\text{--}R10$, crossover at $R7\text{--}R8$ ($\sim 54%$) |
05-mbti/ | okf/LLM_Tests/05_MBTI_PIQUE_TEST/ | mbti | 9-prompt architecture probes & unprompted MBTI typology |
🔬 AI Anthropology 10-Principle Experimentation Mindset
When ingesting and analyzing model test runs, apply the AI Anthropology 10-Principle Loop:
| # | Principle | Application in AI Anthropology Experimentation |
|---|
| 1 | OBSERVE (ext) | Read model execution outputs & multi-turn transcripts completely. Observe exact stopping rounds ($R1\text{--}R10$) without truncating output. |
