The Question Genome Project.

how the source field expands

Difference should be designed, not hoped for.

A source is not a name typed into a question row. People, traditions, organizations, communities, and works receive stable identities in a source graph. Relationships such as influence, collaboration, critique, translation, and shared practice remain separate, directed claims with evidence and confidence.

Discovery begins with what the collection lacks. The system measures coverage across domain, practice, era, region, language, source form, lived context, and genome movement; then searches documented lineages, bibliographies, practitioners, archives, oral histories, community partners, critics, and deliberate counter-canon walks for sources that add something genuinely new.

source priority = marginal genome coverage + evidence + responsible access − concentration

Fame adds nothing. A celebrated source may enter when it contributes a missing question. An unfamiliar source should outrank it when that source adds more coverage, stronger evidence, or a form of inquiry the collection cannot yet see.

Representation is a system constraint.

The source field is audited across gender, region, language, era, discipline, source form, and lived context. Gender is never guessed from a name or reduced to a forced binary: identity metadata is multi-valued, evidence-linked, self-described where possible, and explicitly unknown when it is not documented.

Coverage is measured within every domain, not only across the collection as a whole. Women should be present in work, science, systems, meaning, relationships, making, decisions, and public life—not concentrated in a corrective side collection. When a lane becomes dominated by one demographic, institution, geography, or intellectual tradition, its concentration penalty rises and discovery priority moves elsewhere.

The production database is the system of record. Spreadsheets and seed files remain useful as portable evidence ledgers and reproducible imports, but stable IDs, versioned migrations, provenance, assessments, graph edges, and review states live in the relational database.

the premise

The list is the surface.

The public experience should remain almost painfully simple: a quiet, searchable list of questions.

The complexity belongs underneath it. The durable asset is a defensible understanding of which questions are worth carrying, how they work, where they came from, and which question a person might listen to next.

Ask it once. Do not rush to explain it. Notice what appears. Listen for the question underneath the first answer.

one system, five graphs

Not every line means the same thing.

Similarity is not lineage. Popularity is not quality. A common next click is not automatically the right next question.

what lives inside a question

The genome explains how a question works.

A question is not defined by one category. It is described through several independent facets. Publicly, all of this can be called its genome. Internally, the distinctions remain precise.

question move

How does it move attention—surface to depth, blame to agency, certainty to curiosity?

human territory

What does it bring into the room—identity, belonging, fear, love, truth, time?

function

Is it for reflection, connection, decision, healing, discovery, learning, or action?

time

Does it look backward, stay in the present, project forward, or hold a lifetime?

context fit

Where is it useful, with whom, and how much readiness or care does it require?

listening target

What might someone listen for—the story, body, need, contradiction, absence, or invitation?

provenance

Where did the wording and underlying inquiry come from, and how certain are we?

Decimal scores stay backstage. Readers should encounter clear language, not false precision.

how a question earns its place

The Canon should be small enough to defend.

Before scoring, a question must pass an integrity check. It cannot be a slogan wearing a question mark, a hidden accusation, several questions stacked together, an unattributed quotation, or a duplicate that adds nothing.

Admission is confidence-gated, not approval-gated. Clear cases can move automatically. Disagreement, sensitive material, new source classes, and a stratified audit sample receive human review. Curator decisions become calibration evidence for the next version.

from 1,000 to 100,000

Collection is not endorsement.

The system keeps four layers separate so scale does not dilute trust. A source occurrence can be preserved without becoming a public question; a useful collection can grow without making the Canon meaningless.

occurrence

The exact words as they appeared in a source, with locator, language, rights, and extraction evidence.

candidate

A normalized question proposed for research. Adaptation is explicit; attribution never transfers silently.

collection

A question that passed integrity, quality, provenance, duplicate, and safety checks.

canon

A small, defended set with stable wording, reviewed lineage, a complete genome, and useful graph connections.

What “Browse the collection” means.

The browseable questions are the reviewed public collection: questions that have crossed the current admission boundary. They are not every question found on the internet, every model-generated candidate, or automatically the Canon. The Canon is the smaller subset whose wording, lineage, genome, and graph position are complete enough to defend.

Demand can nominate. The rubric can recommend. Evidence and confidence decide what becomes public.

A question moves through a reproducible protocol.

The first 1,000 will be sampled across domains, cultures, practices, eras, source types, and question moves. No person, tradition, topic, or style can dominate simply because its archive is easier to scrape.

the listening graph

A connection should tell you why to move.

“Related” is not enough. Every public connection should name the kind of movement it offers.

go underneath

Move from the surface account toward a need, fear, identity, origin, or meaning.

see it another way

Change the frame without pretending the new frame is automatically true.

try the other side

Test an inversion, tension, or possibility the first question omitted.

make it real

Move from abstraction to a moment, behavior, person, cost, or choice.

ask someone

Carry the inquiry out of private reflection and into a relationship.

go back to where this began

Listen for the first adaptation, inherited story, or earlier need.

A good trail does not rush someone toward an answer. It helps them remain with what the question is trying to reveal.

a search-led research system

Every inquiry can show the Genome what it cannot yet see.

The permanent Genome is the public evidence base. A search creates a second, temporary graph: a private Query Genome built around one person’s words. That graph can retrieve from the collection, research source-defensible questions beyond it, adapt a question with visible lineage, or generate provisional wording only when research leaves a genuine gap.

The internet is therefore not the database and not the judge. It is a discovery surface. A question may be found anywhere; it earns a public place only through provenance, integrity, quality, distinctiveness, and demonstrated graph usefulness.

private inquiry

Every search is sent securely to the research model and stored in the private research system with no account identity, IP address, or browser fingerprint added by ask.

temporary Query Genome

The system represents intent, language, stakes, missing context, possible assumptions, and underneath territories as confidence-bearing hypotheses.

live research

The current Genome is memory, not a boundary. A model searches broadly across source classes and the open web for questions, practices, transcripts, papers, talks, interviews, and unfamiliar lineages.

independent judgment

One judge is blind to sources; another is adversarial about evidence and readiness. They score candidates and complete paths separately before a final arbiter chooses the arc.

durable evidence

Every candidate preserves wording, source, locator, context, exactness, rights state, confidence, scores, edges, and its role in every path trial.

research demand

Repeated or poorly served situations raise the priority of a private learning signal. Popularity can reveal a gap; it cannot admit a question.

public Genome

Only reviewed questions and claims become public. A private search is never published as written, and generated session wording does not silently become Canon.

A generated question has a higher burden, not a lower one.

Generated wording must pass every integrity gate—continuity, movement, non-assumption, context fit, readiness, single-question clarity, agency, and plain language—then score at least 85 out of 100.

The system keeps four origins visible: canonical, sourced, adapted, and generated for this inquiry. Adaptation preserves its lineage. Sourced and generated session wording remains provisional and private until separate research and review establishes that it deserves consideration.

situation-aware retrieval

Search is a path compiler, not a result ranking.

A person rarely knows the title of the question they need. They arrive with a situation. The system takes longer on purpose: it interprets the inquiry, explores several underneath hypotheses, rejects weak questions, compares complete paths, and returns only the strongest three to five.

research

A frontier reasoning model forms several plausible underneath hypotheses, searches the reviewed Genome and the open web, verifies provenance, and assembles a broad candidate dossier.

judge the wording

A blind judge sees no source names or fame. It scores the questions and complete paths for underneath power, specificity, movement, openness, continuity, clarity, generativity, and complementarity.

challenge the evidence

A separate adversarial judge tests attribution, readiness, safety, hidden diagnosis, unsupported assumptions, repetition, and whether each question preserves agency.

arbitrate the path

A final model compares the dossier and both judgments, then selects exactly one three-to-five-question arc. It cannot invent a new candidate during arbitration.

explain

The result reports how many searches, sources, candidates, and complete paths were considered, while keeping private reasoning and the person’s inquiry private.

learn without auto-publishing

Every candidate, edge, trial path, judgment, and final selection is preserved as research evidence. Nothing found or generated in a live search silently enters the public collection.

One model does not get to grade its own first answer.

The model policy is versioned separately from the rubric and evaluated on frozen inquiries. The strongest available model is a replaceable instrument, not part of the method’s identity. A model change earns production use only when it improves held-out path quality, source integrity, safety, and calibration.

The system retrieves candidates broadly, but returns conservatively. If no complete path clears the threshold, one clarifying question is better than five plausible results.

how the system learns

Every claim should be testable.

The rubric, search engine, and graph are hypotheses. Each version is evaluated against frozen, source-balanced test sets and reported with uncertainty. Changes must improve held-out performance rather than merely make the live map look richer.

admission reliability

Agreement across independent assessors, disagreement by rubric dimension, confidence calibration, and curator-overturn rate.

search usefulness

Recall of expert-selected questions, ranking quality, situation coverage, diversity, and the rate of plausible-but-unhelpful results.

path quality

Movement coverage, redundancy, coherence, readiness fit, dead ends, and whether the final question could not be seen from the opening query.

graph integrity

Unsupported edge rate, directional errors, duplicate communities, orphan questions, source concentration, and confidence drift.

human outcomes

Voluntary saves, returns, shares, and reflections—used as behavioral evidence, never mistaken for truth or quality.

The data model follows established ideas for interoperable provenance and controlled vocabularies, drawing from W3C PROV-O, W3C SKOS, and the FAIR principles. The Question Genome remains its own domain model: those standards support traceability and reuse; they do not decide what makes a question good.

where questions come from

Every question has a history.

The lineage map connects questions to the people, books, traditions, conversations, and practices that shaped them. Every public claim of influence needs evidence. Resemblance can suggest a research lead; it cannot prove a lineage.

A source record also preserves whether wording is exact, near-verbatim, translated, adapted, independently created, or still disputed. That distinction protects both the people who asked these questions and the integrity of ask.

open by design

Anyone should be able to inspect the work.

This page is the first public release of the method. The live map data is inspectable now. Versioned taxonomy, question, edge, provenance, and change-log exports will follow as attribution and rights are made ready for public reuse.

Private searches, generated session questions, personal notes, reader responses, source excerpts without permission, and row-level behavior will never be included in the open map. Only aggregated research gaps may influence what the project studies next.