SCC Nexus · UPF26 evidence observatory
Academic submission
A transparent prospective protocol and methods programme—not an unpublished empirical result presented as a completed study.
Protocol and methods
Scientific architecture, draft manuscript, rule registry, JSON Schema, deterministic engine, reproducibility tests and precommitted pilot protocol exist.
Remaining: freeze and checksum a citable methodology release; verify a reproducibility bundle; approve author order, contributions, interests, funding, ethics applicability, code/data availability, references, independent scientific review and venue requirements.
Empirical pilot results
Not ready: the prospective 20-product pilot has not started. Zero formal products selected and zero predictions locked are recorded in the current protocol status.
Do not treat the live bibliographic harvest as pilot cases or validated findings.
Protocol question and test design
How much information is lost when a packaged food is represented only by its NOVA class instead of independently described processing, whole-food retention, formulation, nutrition and exposure?
The proposed test uses 20 adversarial cases, two independent blinded reviewers and a frozen release. Case 0001 (the motivating coated-peanut example) is excluded from prospective endpoints because it informed the design.
| Prospective decision criterion | Precommitted threshold |
|---|---|
| Complete auditable evidence trails | At least 18 / 20 |
| Exact agreement before reconciliation | At least 16 / 20 |
| Material-discordance cases | At least 4 / 20, spanning 2 types |
| Unexplained rule conflicts | At most 2 / 20 |
| Operationalisability | Explicit human scientific adjudication; a required qualitative gate |
Methods safeguards
- Evidence → transcription → observation → rule firing → classification → independent reviews → reconciliation → interpretation.
- Permitted unresolved and insufficient-evidence outcomes.
- No collapsed “confidence score”; uncertainty is shown by component.
- Nutrition and exposure descriptors cannot change a NOVA classification.
- Changes after a prospective endpoint is locked require a new version and, where needed, a fresh pilot.
What can be evaluated now?
Researchers may critique the public protocol design, versioned evidence stream, redaction policy, endpoint definitions and stop/go criteria.
Complete reproducibility and formal manuscript review still require a frozen release and an authorised evidence package. Private data are not silently exposed for convenience.
Source-of-truth: private sccupf/research, including docs/SUBMISSION_READINESS.md, protocol/UPF26_PROTOCOL_v0.1.md and methodology/releases/0.1.0.yml. Current methodology release is DRAFT, not frozen. This is a public synopsis, not a peer-reviewed manuscript.