The Paper chapters
Today, Timeline, Next meal, Insights, and Library share one composer and command palette. Evidence and operational tools have separate homes.
A reader's guide · September 2026
ParkinSUM's Paper workspace connects local records with deterministic explanations and a synthetic research workbench. This guide shows where to start, how the layers fit together, and where the evidence stops.
Educational production-architecture prototype. Synthetic public demos only. Not medical advice or a medical device; no clinical validation is claimed.
I. Architecture
Today, Timeline, Next meal, Insights, and Library share one composer and command palette. Evidence and operational tools have separate homes.
Owner-entered records, catalog metadata, explicit units, and missingness remain distinct. Public demonstrations use fresh synthetic state.
Supported context passes through applicability and provenance gates. Rules own classifications, scores, safety gates, and evidence.
Mechanistic sensitivity views stay trace-only. They do not select meal times, reorder recommendations, or establish patient-specific predictions.
Explicit consent permits only rule-screened, non-BLOCK whitelist reranking or copy polish. Rule-owned results cannot be overridden.
Production-derived traces, static contracts, synthetic rule cases, and replay artifacts have separate evidence labels and limitations.
II. Demo route
Run the app with its local backend and a fresh synthetic profile. Keep personal records, real identifiers, credentials, and exports out of public demonstrations.
Show the entry composer, source-aware catalog, and a synthetic record. An empty fresh state is valid; unsupported context should remain explicit.
Supply a user-defined time window and inspect candidate rationale, missingness, and limitations. A withheld interpretation can be the correct bounded result.
Open it from the evidence group or command palette. Compare fixed non-personal fixtures and inspect explanation trees. Static registry contracts are not live trace observations.
Inspect authored expectations and actual assertions in a disposable workspace. Passing a case does not install a rule or establish clinical correctness.
Read the capability matrix and dated timeline. Distinguish a documented check from a recorded successful run and a proposal from an implemented feature.
III. Evidence limits
A browser capture shows visible rendering at a recorded source state. A passing synthetic case establishes its scoped assertion. Neither establishes clinical validity, complete standards conformance, production readiness, or physical-device and accessibility acceptance.
The model is not clinically calibrated. ParkinSUM is not for diagnosis, treatment, medication timing, dose or diet guidance, patient care, or emergencies. Owner-entered records may contain sensitive information.
Keep reading
Records, algorithms, standards experiments, sources, and releases.
Implemented features, synthetic evidence, reports, and future work.
Commands, expected results, prerequisites, and failure scope.
Improve provenance, reproducibility, accessibility, and explanation clarity.