Methodology | My Daily Knowledge

The pipeline at a glance

Every lesson in a learning path becomes one canonical article for its level, depth and article type. That article is produced in this order:

Source discoveryA research model searches only the websites of publishers in our source registry that cover the lesson's field and subject.
Source verificationWe fetch every page ourselves and confirm it exists, matches its title and contains the passages recorded from it.
Source assessmentEach verified document is assessed, and the set must clear the sourcing bar for the field.
Grounded writingThe article is written only from the verified facts, with every paragraph citing its sources.
Automated checksCitations, figures, dates, quotations, named studies, length and duplication are checked by code.
Independent fact checkA separate model compares the draft with the sources and reports anything unsupported.
One correction roundIf problems are found, the article is rewritten once to fix them and checked again.
Approve or withholdOnly an article that passes every mandatory check is approved for delivery.

Source discovery

Research does not range over the whole web. For each field, discovery is limited to the domains of enabled publishers in our source registry: encyclopedias, statistics agencies, international organisations, central banks, research institutes, journal publishers, universities, government bodies, professional bodies, health authorities, educational publishers and museums.

The research step looks for four to seven independent sources from at least two different publishers. It prefers, in order, systematic reviews and clinical guidelines (for medicine), official statistics and institutional publications, peer-reviewed research, academic textbooks and reference encyclopedias, then reputable educational explainers. For each source it records the exact URL it opened, short passages copied word for word, and the specific facts those passages support. A source the model did not actually open is rejected.

When reliable material cannot be found, the research step is told to say so and return fewer sources rather than fill the gap.

Source verification

We never take the model's word that a source exists or says what it claims. Before a document can support an article, our own system retrieves it and checks:

Reachability: the page loads, following redirects, within a time limit.
Registry: the final address is on an approved publisher's domain.
Title match: the page's real title matches the title recorded for it.
Excerpts found: the passages recorded from the page actually appear in its text.

Only facts whose supporting passage was found on the page are kept. Everything else from that source is discarded before writing begins.

Source assessment

A prestigious publisher does not make every document reliable, so each document is assessed on its own. The assessment considers six factors:

Authority: The kind of institution that published it.
Evidence type: Whether it is, for example, a systematic review, official statistics, peer-reviewed research, a textbook or commentary.
Verification: How fully the checks above confirmed it.
Relevance: How closely it addresses the lesson.
Transparency: Whether its author and publication date are stated.
Recency: How current it is, which matters more in fast-moving fields than in history or logic.

Opinion pieces and expert commentary can never support a factual claim on their own. The assessment is an internal aid, not a measure of truth, and is never shown to readers.

Grounded writing

The writing model receives the lesson, the reader's level, the article type, the length range and the verified source records, nothing else to draw facts from. Its instructions are strict:

Use only facts in the verified source records. Every paragraph of explanation cites the sources that support it.
Do not use any figure, date, name, study, quotation or book title that is not in the sources.
Mark interpretation and contested points as such, so they are labelled for the reader.
Quote only text that appears word for word in a verified passage. Never invent a citation.
Write an original explanation, never reproducing the sources beyond short attributed quotations.

References are not written by the model at all. They are built from the verified source records each citation points to.

Automated checks

Before any further review, code checks the draft. These checks are mandatory; failing any one stops the article.

Citations exist: every citation points to a verified source, and nearly every paragraph cites one.
Factual support: a factual paragraph cannot rely only on opinion or commentary.
Figures and dates: every number and year in the article must appear in the sources.
Quotations: every quotation must be found word for word in a verified passage.
Named studies: any study or author named in the text must appear in the sources.
Length: the word count must fit the depth the reader chose.
Duplication: the article is compared with earlier lessons in the same path and rejected if it repeats them too closely.
Structure: learning objectives, introduction, main explanation, key takeaways and a well-formed quiz must all be present.

Independent fact check

A draft that passes the automated checks is then reviewed by a separate fact-checking model that sees only the draft and the verified sources. It reports claims the sources do not support, statements that contradict a source, misused technical terms, whether the level fits, and any advice addressed to the reader about their own health, legal situation or money. Any unsupported claim, contradiction or personal advice is a failure.

When an article fails

If the sources are not good enough, the article is never written. If a draft fails a check, it is rewritten once with the list of problems and checked again from the start. If it still fails, it is withheld with its reasons and the lesson is marked as needing more research.

A withheld article is never delivered. Nothing is invented to fill a schedule: we would rather delay or skip a lesson than send one we cannot stand behind.

Medicine and law

Medicine and law are held to a stricter standard. Articles in these fields need at least three usable verified sources from at least two publishers. A medical article must include at least one systematic review, clinical guideline or official health publication; a legal article must draw on primary legal sources, official publications, textbooks or reference works.

Every legal article names the jurisdiction it describes, and fails its checks if it does not. Articles in both fields carry an education-only notice: they explain how things work and are never advice about your own health or legal situation.

Translations

Each article is researched and verified once, in English. Readers in other languages receive a translation that is checked against the original before use:

The structure must be identical: the same sections, paragraphs, citations and quiz answers.
Every figure in the translation must appear in the original. No number, date, unit or formula may change.
Citations, author names, publication titles and links are never translated; references are rebuilt from the source records.

A translation that fails these checks is not used, and the delivery waits for a retry rather than sending a wrong lesson.

Copyright and fair use

Every article is an original explanation. Sources are cited and linked, not republished.
Short passages are kept only as evidence that a source says what we cite it for, and only short attributed quotations appear in articles.
Our verification crawler identifies itself and respects each site's robots.txt rules.
We never bypass a paywall. Paywalled sources are checked for reachability and title only, and can appear as further reading but cannot be used as evidence we have not read.

Neutrality and balance

Political subjects, parties, governments, ideologies and religions are presented neutrally and with respect for religious and cultural differences. The writing instructions and the fact check apply the same rules:

Describing a party, movement or belief is never an endorsement. Its claims are attributed to it, its own language is not adopted as fact, and nothing is disparaged.
Where historians or other scholars disagree, the article says so and presents the main legitimate views fairly, marked as contested. Complex events are not oversimplified.
Documents written by a government, parliament, political party, religious institution or armed force about itself are primary sources. They are used to show what that body says or decided, always attributed, and never as the neutral account of a disputed event. Research must find independent scholarship or reference works on the same points.

The independent fact check reports any draft that endorses or disparages a party, government, ideology or religion, presents a body's own account as the neutral one, presents one side of a genuine scholarly disagreement as settled, or treats a modern state or border as existing when it did not. Any such problem fails the article.

Places and periods

Many subjects can be taught for one place, such as the political history of a country or the economy of a region. Places come from a geography of countries, territories, UN regions, groups of countries and historical states, and periods from a list of eras with approximate dates. Periodisation differs between regions, and the boundaries of some groups vary between sources; lessons say so where it matters.

When a lesson is about a place in a past period, research looks for sources on the states, empires or territories that existed then. The article keeps to the place and period it was given, says where borders, states or names differed from today's, says when a modern country did not exist in the period discussed and names the entities that did, and does not project modern political categories onto earlier societies without explaining the difference.

Some subjects fit only some places, for example the colonial history of a country or the history of its monarchy. These are checked for each place before a lesson is planned and never assumed.

Levels and depths

You set a knowledge level for each field. The level shapes the vocabulary, how much is explained from first principles and how technical the material becomes.

Level	Who it is for	How articles are written

You also choose how long each article should be. Reading time is never taken from the label: it is calculated from each article's real word count at 220 words per minute, and an article whose length falls well outside its depth fails the checks.

Depth	Words	Reading time

Learning modes

Each email schedule has one learning mode. It decides what each lesson covers and how lessons follow one another:

A one-lesson overview is written as a self-contained lesson on the whole topic: background, major developments, important figures or ideas, causes and consequences, significance and key takeaways. It says near the start that it is an overview, not exhaustive coverage. A lesson in a series covers only its own part, builds on the earlier lessons without repeating them, and points ahead to the next one.

Series lessons follow a planned outline: the topic's mapped subtopics where they exist, otherwise an outline planned for the topic. An outline is checked for lessons that repeat each other and must fit the size you chose. When the number of lessons you asked for is too small for the subject, the first lesson says it is an overview; when it is more than the subject needs, fewer lessons are proposed and you are told why.

Choosing the next topic

In Universal Knowledge the next lesson is chosen by a fixed set of rules, not at random, so the same plan always produces the same order:

The field that is furthest below its share comes next. Every field has an equal share unless you gave some fields a larger one or excluded them.
A field is not chosen again straight after it was used, and the humanities, social sciences and sciences are kept in balance.
Within a field, core topics, the important things everyone should know, come first, and subjects you have had least of are preferred.
Lessons about places are spread across continents.
A topic you have already covered is never chosen again.

Your breadth setting decides how often a lesson deepens a topic you received recently instead of introducing a new one:

In Mixed Knowledge, each lesson is either Universal Knowledge or one of your focus fields, chosen by counting what each has received so far against the proportions you set, so the mix stays on target over time.

Repetition prevention

Every lesson has a canonical topic key made of its field, subject, topic or place lens, place, period and perspective, and every lesson delivered to you is recorded with it. Before a lesson is planned, it is compared with everything you have received or have planned:

It is a repeat if it is the same topic for the same place.
It is also a repeat if its title is too similar to a lesson you already had in the same field and place.
A topic is taught again only on purpose: one level higher, from a different perspective, for a different period, or as a refresher you asked for.

Separately, each finished article is compared with the earlier lessons of its path, and rejected if it repeats them too closely.

When topics finish

Each schedule has a rule for what happens when what you chose has been covered. You can change it at any time. When the rule adds lessons, they are planned before the last pending one is sent, so there is no gap.

Whichever rule applies, we send one email saying what comes next, with a link to change it. Your plan's limits always apply.

Learning progression

Each topic you choose becomes a learning path. Where a topic is mapped into subtopics, the lessons follow their prerequisites, so foundations come before what depends on them. Material well below your level is skipped unless you chose it, and a prerequisite you are not presumed to know is added first. Where a topic has no mapped subtopics, it is taught in stages starting from your level.

Lessons are never pitched below the level you set for that field. Each article knows what earlier lessons in the path covered and what comes next, and ends by suggesting the next lesson.

How lessons from different paths are combined depends on the learning structure you choose:

Build My Knowledge Plan Read the sample article

How every article is researched, written and checked