Human view
What a person naturally reads, sees, or hears.
Controlled AI Tests · Indirect influence path
Local Retrieval & RAG Trust-Boundary Simulator
The simulator separates retrieval success from marker extraction and from hypothetical instruction uptake. It uses deterministic local token overlap, not an external vector database.
Research boundary: this page uses bounded, inert data and fixed safe examples. It never executes decoded content, requests secrets, calls third-party services, or performs actions against external systems.
INPUT / CONTROL PLANE
Idle
Build or inspect one controlled artifact.
Inputs remain on this host. Text operations are size-limited; uploaded files are processed from PHP’s temporary upload and are not retained by the application.
OUTPUT / MACHINE VIEWS
Compare evidence, not impressions.
Run an analysis to populate the machine views.
The response will separate summary metrics, representation views, findings, tables, and any generated safe fixture.
Interpretation framework
The same artifact can produce several valid observations.
Structural view
What a parser, DOM, container reader, or metadata extractor exposes.
Decoder view
What becomes meaningful only with a rule, key, tokenizer, model, or tool.
Defensive view
What normalization, rendering, OCR, canonicalization, or policy changes.
What this result can establish
Interpret the receiver and transformation explicitly.
The simulator separates retrieval success from marker extraction and from hypothetical instruction uptake. It uses deterministic local token overlap, not an external vector database.
Machine-decodable is receiver-relative. Record the parser, preprocessing, codebook, tokenizer, key, model, and transformation path before generalizing from one result.
Limitations
What this page does not prove
Token-overlap ranking is an educational stand-in for embedding retrieval. A retrieved marker-bearing chunk is evidence of selection, not proof that a model would follow it.