CHANGELOG
This page records meaningful public changes, not every small internal change.
The knowledge-graph dataset release is published and archived: 902 records and 3,073 links between them, the same as the release of 25 October 2026. The descriptions of 15 period records now say what each dated passage's date means: the date its source was published, a time the passage itself states, or a time written about later. No record or link was added, removed or changed.
Answers about molecules now use only a molecule's exact name or a recorded other name, give each claim's own wording when it names the molecule differently, and show the cited claims' own limits. Dated record counts keep the numbers they showed on their date, and period descriptions in the data say what each passage's date means. Claim, source and entry pages, Learn lessons, lists, places, origins, materials, printed pages and the Spanish pilot's review page use plainer wording and clearer layouts, claims that could not be checked against their source's own words are marked on more of the pages that list them, and the data pages describe the data as downloadable rather than open, because no reuse licence has been granted yet. No smell, claim, source, molecule or knowledge-graph link was added or changed, and search results and ranking are unchanged.
SmellBook now keeps two things apart that were easy to confuse: a claim being VERIFIED and a claim having been checked against its source's own words. The few claims that could not be checked that way say so on their entries, claim pages, the claims list, reference cards, labels and the methodology and evidence pages. Answers to questions about exact molecular forms read Greek letters written out as words the same way as the letters themselves, and say plainly when no checked claim names a form in those words. Why answers quote what each cited claim says before its limits, historical passages say when they do not state the time they describe, and search cards, Connections, Learn pages and long lists use plainer wording. The public data page states its licence position, the accessibility record includes the audit of the previous release, and printed pages no longer show an empty navigation band. No smell, claim, source, molecule, dataset release or knowledge-graph link was added or changed, and search results and ranking are unchanged.
SmellBook now says more exactly what its records cover. A historical passage is described as written at the time only when it states its own time; otherwise its page says it does not. Answers to why questions show each claim's own limits, and claims that could not be checked against their source's own words say so. Public status and data pages keep the website's release apart from the dataset release, the accessibility record lists the recent audits, the Spanish pilot states its review status from its worksheet, and sources are described in linked data by what kind of work they are. Wording on trust, Learn and list pages is plainer, long lists are easier to move through, molecule answers fold their role notes, search cards say what they matched, and records that share a name say what kind of record each one is. No smell, claim, source, molecule, dataset release or knowledge-graph link was added or changed, and search results and ranking are unchanged.
SmellBook now makes important record boundaries easier to read without changing the underlying knowledge. Questions about exact molecular forms no longer fall back to general molecule records, molecule answers include compounds that claims name even when no molecule page is linked, and historical dates say whether the passage states the time or whether a publication date is only the latest time it can describe. Claims and sources distinguish evidence class, source role and study method more clearly; search cards and evidence topics explain what they are; claim limits are easier to read; and long lists are easier to move through. A fresh review also corrected an overstatement about historical transcriptions, repaired broken layouts and clarified several older labels. No smell, claim, source, molecule, dataset release or knowledge-graph link was added or changed, and search results and ranking are unchanged.
Evidence is easier to follow across SmellBook. When SmellBook has no recorded answer, it now says exactly what is missing. Scientific evidence and historical records are kept apart, and evidence summaries say both what a claim supports and how the study was done. Molecule roles explain what the source says without turning presence into cause, while claim pages use clearer sections and keep exact chemical forms visible. Search and browsing read more cleanly on phones, related links use one set of words, and records with the same name say what kind of record they are. No smell, claim, source, molecule or knowledge-graph link was added or changed, and the public dataset is unchanged.
The knowledge-graph dataset release with the renewed evidence is published and archived: 902 records and 3,073 links between them. It adds five VERIFIED claims and their five sources (the first for Coffee Shop and Detergent, and further claims for Coffee, Bergamot and Rosemary) and 13 links to molecules those claims name. One molecule link was removed from Fresh Orange Juice because its claim names one particular form of limonene, and two Rosemary molecule links now also cite the new Rosemary claim.
Answers to "why does it smell?" now appear only when a claim's own checked words say what gives the smell; claims that only report what was measured stay as evidence. Evidence summaries and answers say when the scientific evidence is chemistry or emission work rather than a study of how the smell is perceived. A claim that names one particular form of a molecule is no longer linked to a more general molecule record, and a claim page lists any names it does not link, with the reason for each. The new claims for Coffee Shop, Detergent, Coffee, Bergamot and Rosemary are described in the two evidence entries below. Smells still without suitable evidence say so plainly.
Three entries that already had VERIFIED evidence each gained a claim that answers a different question: which odorants have the greatest impact on the flavour of a roasted coffee, tested by leaving them out of a model of its potent odorants; which compounds of bergamot essential oil have a bergamot-like odour; and which compounds have the greatest impact on the odour of rosemary essential oils. Seven other entries were researched for a different kind of evidence in the same way; no checked source answered their questions, so they stay as they were.
Coffee Shop and Detergent each gained their first VERIFIED claim, from measurements of what is in the air: two compounds in air sampled on café baristas, and d-limonene in dryer-vent air when washing was done with fragranced or fragrance-free detergent. Neither claim describes how the place or product smells, and neither names a cause of its smell. Other entries without evidence were searched again in different ways; where nothing suitable was found, each search is recorded with its reason.
SmellBook pages now say each thing once, in plain words. The Universe page shows a small preview of the smell map and leads to the full Interactive Map. A smell's reference card summarises it, and its connections, such as where it comes from, where you meet it and what was written about it, are listed once below. Place pages keep smells found in a place apart from smells met there. Search counts a longer word for a shorter one only when it is a regular form of it, such as roses for rose, and when it has neither an answer nor a result it suggests other ways to look. Answers about places known only from historical sources say so, and answers about why something smells rest only on scientific sources. Record pages name evidence by its title instead of a code. Every Spanish pilot page says what the pilot is and that English stays SmellBook's reference language. Evidence coverage is explained in one report, by category, type and milestone. On phones, the links to related pages fold into one list. The data for institutions now lists smells met at a place separately from smells found there, and search data uses the titles the pages show. No smell, source, claim or link between records was added or changed, and the knowledge-graph dataset is unchanged.
Search now answers supported questions, such as what a smell is like, what it comes from, where it is met, which evidence or molecules SmellBook records for it, or what checked historical sources wrote, in one line built only from SmellBook's own records, with the label and limits of those records. When the records hold no answer, search says so plainly, which is not the same as saying the answer is unknown or untrue. Search now matches whole words: a word no longer finds pages where it only appears in the middle or at the end of a longer word, as “paris” did inside “comparison”. It still finds the same word with a common ending, such as roses for rose. Smell entries show their connections in one place, and place, origin and material pages bring their links together. Pages for places named in sources, such as London or Valencia, now start with what SmellBook records there. Labs and the evidence coverage report explain in plain words what each number counts. In the Spanish pilot, answers are shown in English, SmellBook's reference language, until their Spanish names have been reviewed by a native speaker. Institutions can download the connections SmellBook records as data. The dataset itself is unchanged.
The knowledge-graph dataset release for Smells in history is published and archived: 892 records and 3,051 links between them. It adds 15 period records at the precision of checked historical passages, four checked localities (Paris, London, Greenwich and Maine), and four historical sources with their VERIFIED claims. The new historical links say which periods and named places those checked records document; they do not describe a whole city or era, and nothing was removed or changed.
A new page lists what SmellBook's checked historical sources wrote about smells, by the time each passage describes and by the places it names. Each time keeps the precision of its source: a century, a year, a month or a day, never a wider era. Paris, London, Greenwich and Maine join Valencia as places named in those sources, each linked only where a passage names it. Four entries gained their first historical source: Fresh Bedsheets (sheets scented with lavender in seventeenth-century inns, in a history of England), Ironed Shirt (a novel of 1839), Flower Market (a travel book of 1884) and Hotel Room (a novel of 1922). A passage records what one writer described then; it is not a picture of a whole city or era, says nothing about the smell today, and is not evidence that a smell has changed or disappeared.
A second knowledge-graph dataset release of the day is published and archived: 865 records and 3,022 links between them. Plants, foods, drinks and animals can now be represented separately from the smell records that come from them: it adds 61 of them (20 plants, 39 foods and drinks, 2 animals) and 71 editorial links saying which smells come from each. Nothing was removed or changed.
A new page lists the plants, foods and drinks, and animals that smells in the Atlas come from: 61 of them, such as the rose plant, coffee and the dog, each with its own page and permanent identifier, and no record number, since they are not smells. 71 smell entries now say what they come from, for example Orange Peel comes from the orange (its peel) and Toast from bread (after it is toasted). The links are editorial, taken from each entry's own name or description; they are not scientific findings, and they do not mean every rose or dog smells the same. Search answers questions such as "smells from rose" or "what smells come from pine". The reference card at the top of an entry now says when its scientific claims come only from studies that measured compounds or emissions, with nobody smelling the material, and two claims about pineapple and yogurt are now labelled as general accounts from reviews, with correction notices.
The knowledge-graph dataset release with the stronger evidence is published and archived: 804 records and 2,951 links between them. It adds 16 VERIFIED claims with their sources and 10 molecules named by those claims; four claims were narrowed to what their sources say, and four molecules gained another name. Nothing was removed.
Fifteen smells gained a VERIFIED claim from a further study, most of them studies in which people smelled the compounds or the material, such as rose, sage, thyme, honey, apple, popcorn, caramel and cork; Hair gained its first claim, from a study of what the scalp releases. Sage, Coffee and Garlic each have a state now backed by a claim. Two existing claims were narrowed to what their sources say. The other entries without evidence were searched again with new kinds of study; nothing suitable was found, and each search is recorded with its reason.
On a phone, the map now has its own taller drawing, centred on the smell you chose with its related smells around it and every dot easy to tap; the list of smells on the map is grouped A to Z. Long lists keep their jump bar in view as you scroll and show where you are. Each entry's buttons to compare, see the evidence, cite or download it now come straight after its description, and links and buttons are larger to tap. The map, the Universe and the links between smells use everyday words. A name used by more than one record, such as Café, is marked on the names list and in search, which lists each record with its SB number. The Map, the Universe and the Directory now use the same names for the same groups, and the Directory explains how categories, smell families, types and states differ. No new entries were published in this update.
A smell's map neighbourhood is now drawn around it, with its related smells around it and listed below it, including, in the two-step view, the smells two steps away. Filtered maps name more of their smells, and those with many smells spread them across the whole map. Long lists, such as smells by SB number, all names and the larger family and evidence lists, now have a jump bar of letters or number ranges with counts, and the freshness page lists every profile once with its update date and revision. Lists stay complete and in alphabetical or number order; nothing is ranked. No new entries were published in this update.
New entries are prepared for human review in small groups, each on one page showing the proposed name, possible duplicates, whether it is only another state of an existing smell, the exact source wording and the proposed claims kept apart from the editorial text. A draft is published only after a named person other than its author has reviewed it and its publication has been approved separately; automated and AI checks never count as that review. A new entry's sources stay unverified until they are checked through source intake. No new entries were published in this update.
Each entry opens with a reference card: its SB number and permanent identifier, a short description, its smell words and families, the states and settings it is met in, its evidence, the molecules its VERIFIED claims name, its history and comparisons, with buttons to compare, see the evidence, cite or download the entry as data. Further down, every entry shows its names, words, science, context and history side by side, and a Cite panel with copy buttons, BibTeX, RIS, a print reference view and a QR code to the record.
Smells can now be browsed by family, such as wood, earth and stone or fresh and clean, with each smell in every family its own words belong to. Many entries now record the states they are met in, such as wet, aged, roasted or outdoors, taken from the entry's own records or from a VERIFIED claim and independently reviewed. A state never makes a new smell.
One directory opens every way into the Atlas: A to Z, SB number, names, smell words and families, states, settings, places, materials, evidence, molecules, history and comparisons. Any two smells can be compared field by field, with what differs and what is not known, and no similarity score. Search now finds record numbers such as SB-0042 and answers questions like cite petrichor, what evidence supports petrichor, earthy smells and smells found in kitchens.
The methodology now explains SmellBook's reference system in plain words: one smell, one identity; smell families; states; editorial and evidence-backed links; how smells are compared; how to cite a record; and that SmellBook describes smells and records evidence, but cannot guarantee that a smell can be reproduced physically. A new guide shows how to use a record in an exhibition, a class or a collection.
The knowledge-graph dataset release with the new everyday settings is published and archived: 762 records and 2,885 links. It adds 10 settings, 212 links showing where entries are met, and one new VERIFIED claim with its source. Nothing was removed.
Ten everyday settings join the places, including the market, the forest, the park, the street, the bathroom, the car and the station. 212 new links show where each entry is met, taken from the entry's own records and independently reviewed, so 172 of the 200 entries now belong to at least one setting. Each setting page is a guide: which of its smells have VERIFIED evidence, when you meet them, how they compare, and how and when every link was made. Everyday settings have no address or coordinates.
Eight editorial routes through everyday settings: a morning at home, a summer day out, market morning, town after rain, the commute, woods and park, a school day and a hotel stay. Each stop shows the entries linked to that setting. The journeys describe experiences and are not maps of real places.
The places page now separates everyday settings from geographic places. Valencia shows its coordinates as its Wikidata record gives them, checked by an automated SmellBook check, with a link to a public map. Search answers questions such as smells of a market or what does a station smell like.
Setting entries without evidence, and Valencia and the Fallas, were searched in planning, tourism and social-science research for smellscape and smellwalk studies. Train Carriage gained its first VERIFIED claim, from a case study of an odour in Australian railway cars. The other searches found nothing suitable and are recorded with their reasons.
The knowledge-graph dataset release with the new comparisons is published and archived: 750 records and 2,671 links. It adds 28 links between entries that are now compared side by side. Nothing was removed.
Search now answers questions as people type them: what does rain smell like, why does soil smell, what is the difference between soil and wet stone, smells of autumn, which smells contain linalool, what is an aldehyde. The page that answers comes first, and entry answers open at the right question. Seasons and times of day are now searchable. Search stays the same for everyone and stores nothing.
Each of the 200 entries now answers six questions: what it smells like, where you meet it, what is behind the smell, how people describe it and what it is confused with, whether it has a history, and what is not known yet. Every answer is labelled EDITORIAL or VERIFIED. Each entry has a new, independently reviewed editorial description that states no scientific, causal, historical or health facts; those come only from VERIFIED claims.
28 new comparisons put commonly confused smells side by side, such as petrichor and soil, lemon zest and bergamot, onion and garlic, fireplace and campfire. Season, time-of-day and place pages now answer what they smell like from their entries, saying where the link is editorial. Nothing is ranked.
The 26 Iconic 100 entries still without a VERIFIED claim were searched again with technical terms for each material or setting. Nothing suitable was found, and each search is recorded on its entry with its reason, such as research on a neighbouring subject only.
The knowledge-graph dataset release with the connected molecule evidence is published and archived: 750 records and 2,643 links. It adds 4 sources, 8 VERIFIED claims and 14 molecules, and two molecules gain other names. Nothing was removed.
Smells, claims, sources, molecules and Smell and Science lessons now link each other through the evidence: every link is a relationship SmellBook has recorded, never a ranking or recommendation. Molecule pages are written for readers without a chemistry background, nine entries gained claims naming the molecules their sources name, and a new guide explains the chemistry words. No entry was added and no entry identifier changed.
Nine entries gained a VERIFIED claim naming the molecules their source names: soil (geosmin), dark chocolate, vanilla bean, cocoa powder, ginger, mango, old book, library and brown sugar. Five come from sources already cited, checked word for word against their abstracts; three come from four new research sources. Each claim keeps the role and limits its source states, for example that the mango study estimated importance without a sensory test.
Molecule pages now start with what the molecule is, then show where research reports it, with the source named beside every claim and each role explained in plain words, and end with what SmellBook is not claiming. Claim and source pages list the molecules their claims name, entries tie each molecule to the claim above it, and the molecule list can be browsed by chemical family (from the ChEBI classification), by what the claims say and by kind of entry. Nothing is ranked or recommended.
A free companion guide to Smell and Science explains words such as volatile compound, odorant, concentration, aldehyde, ester, thiol, terpenoid and isomer in plain language, and lists the molecules in SmellBook's evidence that belong to each chemical family. Lessons now link the molecules their claims name. The definitions are editorial and say nothing about how any family smells.
The knowledge-graph dataset release with the evidence for harder cases is published and archived: 724 records and 2,588 links. It adds 17 sources, 17 VERIFIED claims and 5 molecules, and one molecule gains another name. Nothing was removed.
17 more published entries gained VERIFIED claims, so 137 of the 200 now have at least one. The new evidence comes from a wider set of research indexes and from non-fiction historical texts, and every claim was checked word for word against its source. No entry was added and no identifier changed.
Every entry without a VERIFIED claim now shows on its page whether it has not been researched yet or has been searched without finding a suitable source. Each search records the kind of evidence sought, how it differed from earlier searches, and one of six plain reasons why nothing qualified, such as research on a neighbouring subject only, or only novels and poems. None of this is a score or ranking.
14 new published research sources support claims for entries such as hardware store, dishwasher steam, parking garage, taxi interior, bus interior, subway, hair salon, pharmacy, sunscreen, cork, toast, fried onions, fresh corn and hay, and an iron study reports that the metallic smell after touching iron comes from the skin. An existing study of sun-dried cotton towels is now also linked to fresh towels. Emission measurements are not presented as findings about how a place smells.
The 1911 Encyclopædia Britannica on the scent of good hay, an 1863 magazine article quoting a report of drain smells drawn through a bakehouse, and an 1862 account of the first journey along London's Metropolitan Railway are now cited by page, each as evidence of its own time.
Hydrogen sulfide, ethanethiol, methanol, (Z)-3-hexenol and ethylbenzene are recorded because the new claims name them, each with its checked PubChem identifier. Compounds a claim names without saying which isomer, or names only in a historical text, are listed with the reason they are not recorded.
The knowledge-graph dataset release with the new evidence is published and archived: 685 records and 2,523 links. It adds 35 sources, 36 VERIFIED claims and 20 molecules, and two molecules gain other names. Nothing was removed.
37 published entries gained their first VERIFIED claims, so 120 of the 200 now have at least one. Each new claim rests on a source whose own abstract or cited page supports it, checked word for word against the source and then by a separate automated review. Places, materials, products and the history of smell are covered as well as food chemistry. No entry was added, no identifier changed, and the first draft batch stays frozen and unpublished.
29 new published research sources support claims for entries such as chlorine pool, freshly cleaned room, airplane cabin, new paint, new apartment, office, classroom, hot pavement, fireplace, warm electronics, printer paper, autumn leaves, pine forest, chamomile, grapefruit peel, peppermint and toothpaste. Each claim states its limits, and emission measurements are not presented as findings about how a place smells.
Six historical sources from 1782 to 1926, read on their digitised pages, support seven claims: the smell of a stable in an 1880 Hanoverian village, the sea air at Greenwich in 1845, Paris churches in 1782, eucalyptus groves in 1874, an 1898 report on why handled metals smell, and a cellar and the smell of snow remembered in a 1926 memoir. They describe their own time, not today.
Older entries are on archive pages: page 2 · page 3 · page 4 · page 5 · page 6 · page 7 · page 8 · page 9 · page 10.