A source-indexed knowledge base and architecture workbench for studying OPS concepts, comparing model versions, constructing exploratory type codes and recording evidence without treating hypotheses as diagnoses.
Trace concepts from source → definition → rule → hypothesis.
Official creator statements are separated from community guides, community tools and independent claims. Each record has a direct link and notes.
Functions, animals, dimensions, glossary terms and explicit rules live in a structured JSON corpus. Version conflicts stay visible.
Compose a candidate code, inspect the selected inputs and see basic checks. This first version is not an exhaustive canonical type enumerator.
Save observations, support/contradict links, alternative interpretations and follow-up questions. Keep confidence distinct from calibrated probability.
Compare claims without silently merging them.
Search the corpus by term, definition, source or model.
Compose and document a candidate. This is not a typing test.
Basic structure checks only; not proof of a correct type.
Example format from a secondary community guide.
Compare code components without assuming either candidate is correct.
A navigable overview of how the corpus groups OPS concepts. Select a node to inspect its record.
Select any node in the map to see its description and source links.
Stored in this browser only. Use fictional or anonymised labels when possible.
Use behavioural descriptions, not trait labels alone.
Review observations, alternatives and open questions.
Paste a public URL, upload a file, or paste text directly.
TXT, MD, CSV, JSON, HTML, PDF and DOCX. Files are processed in this browser.
Signals and evidence—not a forced type verdict.
Potential language associated with cognitive functions and Observer/Decider and Introverted/Extraverted orientations.
Possible references to Play, Sleep, Blast and Consume behaviours, with matched source excerpts where available.
Sample size, repeated versus isolated clues, contradictory signals, and important context gaps.
Follow-up questions to test alternative explanations instead of treating keyword matches as proof.
Load or paste source material, then select “Generate insights report.”
Import CSV/JSON rows that have text, an OPS label, source metadata and a person/source grouping key.
Recommended source to investigate: AOP interview-lines dataset on Hugging Face. Review its dataset card, licensing, annotation method and reuse terms before use. The studio does not copy or redistribute its data.
CSV or JSON; up to 5,000 rows per import for browser memory safety. CSV accepts common OPAI/AOP field names.
Text: text or transcript; type: ops_type, type or type-128; group: person_id, speaker_id, name or source_id; provenance: source URL/title, license and annotation method. Missing provenance is flagged.
Retrieve similar corpus excerpts and aggregate their labels by OPS dimension.
Retrieval uses local TF-IDF vectors over unigrams and bigrams. It is a lexical similarity baseline, not a neural embedding model. Text stays in this browser tab.
Load a corpus and enter text to see retrieved evidence, candidate distributions, margins and uncertainty warnings.
Deterministic group split: all rows from a person/group stay on one side. Near-duplicate rows from the same source should share a group ID.
What is encoded, what is testable, and what remains unverified.
Check the classic type-code space against the currently encoded rules.
Open the original material. Source type is explicitly labelled.
Open questions that prevent overclaiming completeness.