How reports are built
Research, with the limits visible
InnerCanvas uses a curated, versioned evidence library for each drawing prompt. It does not search the web while creating your report and does not treat projective drawing methods as diagnostic tools.
Observation comes first
The single multimodal model call records visible features such as placement, relative size, line continuity, task completion, and spatial relationships before producing the report under one strict output contract. It is prohibited from assigning personality, diagnosis, intent, history, or protected traits.
One Evidence Pack per test
Each of the 15 tests has its own source registry, search log, Claim Cards, limitations, contradiction review, and evaluation cases. That same call receives the drawing, the selected test's approved pack, and the output schema, then returns structured observations, the preview, and the complete report together.
What counts as evidence
We prioritize systematic reviews, meta-analyses, validity and reliability studies, replicated findings, and traceable methodology sources. Null results, conflicting findings, population limits, culture, artistic experience, and scoring differences remain part of the pack.
What we do not claim
Drawing-based projective frameworks have mixed and context-dependent evidence. InnerCanvas does not claim scientific proof, clinical validation, psychologist endorsement, diagnostic accuracy, or a hidden meaning for every symbol.
Traceability
Every research-informed report statement must link internally from a visible observation to an allowed Claim Card and then to named sources. When evidence is weak or inapplicable, the report keeps the observation, omits the inference, and may offer a neutral reflection question.