Identity first
Give a smell a persistent identity before adding rankings, recommendations or commerce.
PUBLIC METHODOLOGY
Explain the trust model. Keep the secret sauce private.
SmellBook can be transparent about evidence, definitions and user trust without publishing private ranking, recommendation or graph-building methods.
Separate kinds of evidence
Scientific sources, community perception, cultural association and model prediction should never be presented as interchangeable.
Context matters
A smell can change with state, place, product form, memory and use context. The public model should preserve those distinctions.
Personal is not universal
Aggregate perception can reveal patterns, but individual experience can disagree with the crowd without being 'wrong'.
Citations for factual claims
When SmellBook makes a verifiable scientific or factual claim, the long-term standard is to attach an appropriate source.
No hidden commercial bias
Future product recommendation should be independent of commission. Sponsored placement should be visibly labeled.
Experimental is not evidence
Labs representations and PREDICTED candidates stay separate from canonical relations and VERIFIED claims until a separate review process changes their status.
An open reference language for smell.
Each smell is one record that can be identified, described, compared and cited. Read the record from identity to citation: identity, character, context, evidence, relationships, molecules, settings, names and reference tools.
- One smell, one identity
- Every published smell has one SmellBook record number (SB-XXXX), given in order and never reused, and one permanent identifier (smellbook:smell:<key>). There is no second numbering system. Names, words, contexts and evidence all attach to that one record.
- Smell families
- Smell words are grouped into a few families to help people browse. A smell appears in every family that one of its own words belongs to, so it can sit in several. Families are an editorial aid, not a scientific classification of smells.
- One smell, many states
- A smell can be met in different states: wet or dry, fresh or aged, cut, raw, cooked or roasted, indoors or outdoors. SmellBook records a state only where an entry's own records or a VERIFIED claim say so, labels which it is, and never makes a new smell because a state exists.
- Editorial and evidence-backed links
- Links between smells, words, settings and materials are made by SmellBook editors and labelled EDITORIAL. A claim is VERIFIED when it rests on a registered published source whose citation SmellBook checked; whether its wording was also checked against that source's own text is a separate fact, shown with each claim. A molecule named in a claim is not presented as the cause of a smell, and cultural association never becomes scientific fact.
- How smells are compared
- Two smells are compared field by field from their records: words, families, settings, states, evidence, molecules and history, with what differs and what is not known. SmellBook shows no similarity score, percentage or distance, because it has no published method for one.
- Citing a record
- Cite an entry by its SB number, permanent identifier, revision and date; the Cite panel on every entry gives the reference, the citation, BibTeX and RIS. Cite the original source for a scientific claim. Citing records what was read; it does not grant a licence.
- What SmellBook is not
- SmellBook describes smells and records evidence about them. It does not define physical smell standards or reference samples, and it cannot guarantee that a smell can be reproduced physically.
Search helps visitors; it does not create new canonical content.
Internal Search is lexical, non-personalized and marked noindex,follow. Canonical knowledge pages remain indexable, while query variants do not become search-engine landing pages. Sitemap dates follow editorial update metadata rather than deployment time.
Read the indexing policy ↗Public
Definitions, evidence labels, source standards, safety principles and what a feature means.
Private
Ranking weights, recommendation formulas, graph-construction logic, internal scoring and other internal methods.