Public Observatory hardened
The public presentation was rebuilt around evidence status, falsifiability, accessible navigation and the one-way publication boundary, without adding unsupported scientific claims.
SCC Nexus · Research Observatory
What disappears when one food label tries to explain everything?
UPF26 tests whether a single ultra-processed-food category conceals meaningful differences between packaged foods. Evidence, rules, classification, composition and reviewer judgement remain separate so uncertainty stays visible.
No health scoreNo AI final classificationBlinded reviewVersioned rulesFail-closed ambiguity
Current status
Built to make uncertainty auditable.
UPF26 is currently a methodology and evidence infrastructure project. The public dashboard distinguishes readiness from findings and refuses to turn incomplete evidence into apparent certainty.
20formal adversarial pilot slots defined
0products formally selected
0predictions locked
v0.1foundation / methodology draft
Interpretation guard: these figures describe research readiness only. They are not findings about ultra-processed foods, health, diet quality or disease risk.
Prospective pilot
Progress without manufactured evidence.
The pilot is designed around deliberately difficult boundaries rather than a convenient sample. Case 0001, which motivated the project, is excluded from the prospective endpoint because it influenced methodology design.
Precommitted stop / go gate
The project continues only if the richer dimensions are reproducible and materially informative.
| Criterion | Threshold |
|---|---|
| Complete auditable evidence trails | ≥18/20 |
| Exact blinded reviewer agreements | ≥16/20 |
| Confirmed material-discordance cases | ≥4/20 |
| Different discordance types | ≥2 |
| Unexplained rule conflicts | ≤2 |
FalsifiableIf the richer framework adds little information beyond NOVA or cannot be applied reproducibly, the correct outcome is to stop or redesign.
Methodological architecture
Evidence first. Interpretation last.
UPF26 keeps source evidence, transcription, deterministic rule firing, classification and human review distinct. Ambiguity is an output state, not a reason to invent a confidence percentage.
1 · Evidence & transcription
Source material and transcribed facts remain traceable so later judgements can be audited back to the underlying evidence.
2 · Deterministic classification
Versioned rules can emit NOVA 1–4, unresolved 3/4, rule conflict or insufficient evidence. Determinative conflicts fail closed.
3 · Blinded independent review
Two reviewers assess the same locked evidence and methodology independently. Reconciliation creates a new record and does not overwrite either review.
01Evidence
02Transcription
03Rule firing
04Classification
05Confidence components
06Review A / B
07Reconciliation
08Interpretation
evidence completenesstranscription certaintyrule determinacyreview state remain a vector rather than being collapsed into a single confidence score.
Project record
Research and engineering milestones
One-way publication boundary established
The public repository became a publication surface only. Private working records remain in the research source; approved output is copied after validation.
v0.1 evidence-first foundation created
Protocol, append-only schema, deterministic engine, blinded-review model, adversarial pilot registry, stop/go analysis, tests and CI were established.
Affiliation
SCC Nexus
UPF26 is developed within SCC Nexus: Search · Corroborate · Communicate. Methodological challenge, corrections and research collaboration are welcome.
Corrections and collaboration
Use the same canonical route for general enquiries, research collaboration, corrections, provenance amendments, privacy and accessibility.
| Purpose | Route |
|---|---|
| General / collaboration | hello@sccnexus.co.uk |
| Correction / provenance | hello@sccnexus.co.uk |
| Privacy / accessibility | hello@sccnexus.co.uk |
For corrections, identify the page or case, the statement or data point, the proposed amendment and the strongest available source.