Methods

Deep Time and Sapiens: interactive timelines where every date is a gateway to the scientific literature

Guillaume Jouret, Genes.media. 5 October 2026

Deep Time and Sapiens are two free, bilingual web timelines: the 4.54 billion years of the Earth and of life, and the 300,000 years of Homo sapiens.

1. Introduction

A date in the history of the Earth or of humankind either results from a measurement or is an inference. It is obtained by a method, published by identified authors and open to revision. Conventional timelines (wall charts, textbook friezes, museum panels) give one date per event. They state neither its source, nor its uncertainty, nor the existence of competing estimates.

In the scientific literature, published dates and interpretations are regularly revised, disputed or sometimes rejected. First, new measurements can lead to the revision of a date. For example, at the start of the twentieth century, estimates of the age of the Earth ranged from a few million to billions of years. A first valid age, 4.55 billion years, was calculated in 1953. The value now in use, 4.54 billion years, was determined in 1980; it corresponds to the age of the material from which the Earth formed 1. The origin of Homo sapiens is a second example of how dates evolve. Until 2017, it was placed around 200,000 years ago, the age of the oldest fossils then known. In 2017, fossils from Jebel Irhoud (Morocco) were attributed to H. sapiens and dated to 315,000 ± 34,000 years 2: this result moved the date back by more than 100,000 years. These fossils document early stages of the H. sapiens lineage 3: the date assigned to the origin of the species therefore depends on the criterion that defines the species. Second, reference charts are updated. The International Chronostratigraphic Chart is published in successive versions 4. Between the version of January 2013 and the version of June 2026, the age of the base of the Cretaceous changed from about 145.0 to 143.1 million years, and that of the base of the Cambrian from 541.0 to 538.8 million years. In addition, some dates have no consensus: the ages proposed for the onset of plate tectonics range from the Hadean to the Neoproterozoic, almost all of Earth history 5. Published interpretations are also contested; fossil structures 3.7 billion years old were described in 2016 as stromatolites (layered deposits built by microbes) and as the oldest evidence of life 6. However, in 2018, another study attributed them to deformation of the rock, without biological origin 7. Finally, some proposals are rejected: in 2024, stratigraphers voted against the creation of an Anthropocene epoch 8.

This is important for science education, because students are expected to understand how scientific knowledge is produced and revised. This objective, the understanding of the nature of science, is one of the goals of science education 9. In a Delphi consultation, 23 experts (scientists, science educators, teachers, and historians, philosophers and sociologists of science) agreed on nine themes about the nature of science that school curricula should include 10. One aspect of the nature of science is that scientific knowledge is tentative, that is, open to revision. A review of 52 studies found that it is among the aspects that learners acquire with most difficulty 11. Practising science is not sufficient to acquire it. In a study of 62 sixth grade students, one group discussed four aspects of the nature of science explicitly after its inquiry activities, including the tentative character of scientific knowledge; the understanding of these aspects improved in this group, whereas that of the group that performed the same activities without discussion did not 12. A timeline that gives one date per event, without its source or its uncertainty, provides no material for such discussions: it does not show whether a date has been revised or is disputed.

Geological durations are difficult to estimate. In a study of 126 university students, the dates estimated for seven major events ranged over several orders of magnitude, with a tendency to underestimate the age of the events 13. Students generally place events in the correct order but misjudge the durations between them 14. The same result was obtained in first year university students from 26 European countries 15. More generally, people have difficulty reasoning about magnitudes outside human perception 16. In two experiments, the progressive and hierarchical alignment of scales along a timeline improved the understanding of geological durations 17. Visual representations, whether static, animated or interactive, are central to the teaching of deep time (time at the geological scale, from millions to billions of years) 18.

Curiosity and active engagement improve learning. Curiosity arises from the perception of a gap in knowledge 19, and information that arouses curiosity is better remembered ^{20, 21}. The ICAP framework (Interactive, Constructive, Active, Passive) classifies the engagement of a learner into these four modes, defined by observable behaviour. It predicts that learning increases when engagement changes from passive to active, then constructive, then interactive 22. In a meta-analysis of 225 studies of undergraduate science, engineering and mathematics courses, active learning increased examination scores by 0.47 standard deviations compared with lecturing 23.

Deep Time and Sapiens were designed from these observations. Deep Time covers the 4.54 billion years of the Earth and of life. Sapiens covers the 300,000 years of Homo sapiens with the same timeline engine. The approach is not dogmatic: each date and each event is presented in accordance with the current state of knowledge and attributed to the publications that support it, with an explicit statement of its uncertainty and the alternative estimates or interpretations that have been published. The timelines are dynamic in two respects. Their use is dynamic: the user zooms, moves along the time axis and discovers the articles. Their content is dynamic as well: the corpus is revised when a publication or a new version of a reference chart modifies the state of knowledge.

Objectives

Scale. Represent 4.54 billion years (Deep Time) and 300,000 years (Sapiens) on continuous timelines in which durations remain proportional at every zoom level, from the whole time range to a view of about ten years.
Attribution. Attribute each date and each event to identified scientific publications, accessible from the article.
Uncertainty. Assign each date to an uncertainty category (precise, approximate, interval or disputed) and explain each disagreement.
Active exploration. Make exploration the mode of use, in order to arouse curiosity.
Revision. Maintain the corpus as a dated state of knowledge: record each correction and review each revised text.
Access. Provide the timelines free of charge, in English and in French, on computers and mobile devices, for learners, teachers and the general public.

2. Methods

2.1 Time model

Time was represented by a single variable: the number of years before the present. The present was set to the year 2026. This reference year will be updated each year. Calendar years were used for display only, when the view spanned less than 40,000 years. Durations were expressed in ka (10³ years), Ma (10⁶ years) and Ga (10⁹ years). Conversions between time and screen position, date formatting and zoom operations were covered by unit tests at nine reference values: 4.54 Ga, 3.5 Ga, 541 Ma, 66 Ma, 7 Ma, 300 ka, 12 ka, the year 1800 and the present.

2.2 Timeline engine

Both editions (Deep Time and Sapiens) used the same timeline engine, with different constants. The engine was built on three principles.

First, on the display, distances are proportional to durations, at every scale. Second, a single numerical representation of time was used in all computations. Third, computation is independent of the drawing of the timeline: conversions between time and position, zoom, bounds and graduations were computed by pure functions, which could be tested without a browser and did not depend on the drawing technology.

Time values were stored as double precision numbers (64 bits) of years before the present. At 4.54 Ga, the resolution of this format was about one millionth of a year. At the minimum span, one pixel represented about 0.005 year, three orders of magnitude above this resolution.

Zooming was selected to give access to several scales. Logarithmic time axes were rejected, because a change of scale within a view can impair the comparison of durations 24. Each position of the display is defined by its centre and its span. Time was mapped linearly to horizontal position. A zoom step multiplies the span by a factor and keeps the time under the pointer at its position. The span ranged from 10 years to 5.40 Ga in Deep Time and from 12 years to 417 ka in Sapiens. The maximum span was the time range of the edition (4.54 Ga and 350 ka) plus a margin of 8% of the view on each side.

2.3 Corpus and data model

Each edition (Deep Time and Sapiens) contains 200 articles, available in French and in English (Table 1). This number results from three constraints. The first was legibility: on a screen 1,280 pixels wide, a view of 1 Ga displayed 30 of its 150 articles as individual markers and grouped the others, so that a larger corpus would mainly have added grouped articles. The second was coverage: the corpus was extended to 200 events to cover the time intervals that would otherwise have had none. The third was review: each article was reviewed individually and the whole corpus was reread against its sources, which limited the number of articles that one reviewer could maintain. Sapiens was extended to 200 articles to reach the same size. The corpus was built as a reasoned selection of events (purposive sampling) and not as an exhaustive inventory. An event was eligible when it met three criteria:

Temporal scope and datability. The event fell within the time range of the edition (from 4.54 Ga to the present in Deep Time, from 350 ka to the present in Sapiens) and could be assigned a date or an interval in one of the four dating categories defined in Section 2.5. The time range of Sapiens started at 350 ka, 35,000 years before the age of the oldest fossils attributed to the species (315 ka), in order to include the lineage that preceded it.
Documentation. At least two scientific publications supported its date and its description. Each article cited two to four publications (mean 3.1 in Deep Time and 3.2 in Sapiens), selected and verified as described in Section 2.4.
Importance. The author graded each event on an ordinal scale of five levels defined in writing: 1, major milestone, without which the history covered would show a gap; 2, important milestone; 3, significant event; 4, secondary event; 5, specialist detail. This grade was an editorial judgement and no quantitative indicator, such as a citation count, was used: the number of citations received by a publication varies widely between disciplines and between years of publication 25, and the sources of the corpus covered several disciplines and were published between 1846 and 2026. Levels 1 to 5 contained 13, 52, 98, 29 and 8 articles in Deep Time. In Sapiens, levels 1 to 3 were used (35, 140 and 25 articles) and corresponded to the scale of the species, of a continent and of a region: candidate events far outnumbered the 200 selected, and none below the level of a major regional landmark was retained.

An event was excluded when fewer than two verified publications supported it or when an existing article already covered it.

Articles were stored as structured data (JSON format), separately from the program code, in one file per language. Each article had a permanent identifier, which was used in its web address. Each article was assigned to one to three thematic layers, for example life, climate or genetics in Deep Time and technology, society or peopling in Sapiens (Table 1). Automated checks made it possible to verify the uniqueness of the identifier, the order of the bounds of a period, the number of sources and the consistency of the uncertainty record.

Table 1. Characteristics of the two corpora on 5 October 2026

Characteristic	Deep Time	Sapiens
Time range covered	4.54 Ga	350 ka
Articles	200	200
Thematic layers	9	10
Median length of an article (words)	279	279
Scientific references cited	615	646
Distinct sources	566	628

The 400 articles cite more than 1,150 distinct sources. The median year of publication was 2017, and 34% of the sources were published in 2020 or later.

2.4 Literature search and source verification

Publications were searched article by article, before writing, in two search engines, Consensus and PubMed. For each article, one query in English, occasionally two, was submitted to Consensus, an academic search engine that combines semantic and keyword retrieval. At the time of the searches, its index contained the abstracts and metadata of more than 200 million scholarly publications from all publishers, the full text of open access publications and, under agreements with publishers, the full text of publications from subscription journals.

A query combined the name of the event, site or object with its defining terms and, in 38% of cases, an age or a date. Example: "onset Northern Hemisphere glaciation Quaternary ice ages intensification". No filter on publication year, study type or journal was applied; preprints were excluded in two queries. Each query returned a ranked list of publications with their abstracts.

A total of 459 queries were submitted to Consensus for the two editions between 14 August and 4 October 2026, for the articles and for the contextual content (time scales, environmental bands and illustrations). PubMed was searched directly for the original publications of historical discoveries (29 queries), for which Consensus returned few relevant results.

For disputed dates, publications that supported each position were retained.

Each reference was then verified in three steps:

Identifier. The DOI or PubMed identifier was resolved against PubMed and Crossref by a program, which compared the author and the year with the citation.
Attribution. Each reference was matched to the sentence it supported.
Content. Each sentence was compared with the cited publication, retrieved from Consensus, PubMed, Europe PMC, Crossref or OpenAlex.

Sources were research papers and reviews published in peer reviewed journals. Institutional documents, for example from space agencies and the United Nations, were used for some events of recent history.

2.5 Dating uncertainty

Each date was assigned to one of four categories (Table 2). When applicable, the older and younger bounds were also taken into account. Dates were displayed with the precision that the sources supported, for example "~3.7 billion years ago" or "between 4.4 and 3.8 billion years ago". An explanatory note, displayed in the article, was mandatory for disputed dates. Approximate, interval and disputed dates accounted for 314 of the 400 articles (78.5%).

Table 2. Dating categories

Category	Definition	Deep Time	Sapiens	Total
Precise	Date established without significant margin	34	52	86
Approximate	Best estimate	47	104	151
Interval	Period between two bounds	65	19	84
Disputed	Published estimates disagree	54	25	79

2.6 Writing, review and validation

Articles were written from the publications they cited. Technical terms were defined at first use. The French and English versions of each article were written as separate texts and not as literal translations. Between articles, contradictory statements were resolved by reference to the scientific sources. All 400 articles were reviewed and approved individually.

2.7 Updating and quality control

Content and code were kept under version control (Git); each change was dated. Each article recorded the date of its last review. Three events triggered a revision: a publication that modified a date or an interpretation, a new version of a reference chart, and the periodic rereading of the corpus.

After the first release (August 2026), the whole corpus was reread against the abstracts of its sources in September 2026. The geological time scale was checked against the June 2026 version of the International Chronostratigraphic Chart on 20 September 2026, and four boundaries were updated. The chart version was stored with the data. Errors can be reported through the contact form of the host website.