You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Editorial method: declared perspectives, checkable facts and accountable curation #204
Delivery child of #193: an early About / Editorial method page explaining declared perspective, checkable evidence, selection rules, limitations and accountability. Pair it with #203, the curation-history/weights/profile issue. This is researched draft copy and future implementation specification, not marketing and not a claim that planned features have shipped.
Use an informational page linked from the existing About/navigation and the curated-section disclosure. Coordinate exact URL/history/profile destinations with #194; this issue does not establish a new public namespace. Preserve #201's one-default MVP and later domain/personal roadmap. No new curator, public accounts, chooser, training or additional model runtime in this scope.
Delivery priority: publish this page early, after copy/source and navigation review. It can ship before the model runtime, saved compositions or public history. Describe those features as planned and omit their links until working destinations exist. This is planning priority, not authorization to implement the page in this issue-editing task.
Research record and attribution boundaries (checked 26 September 2026)
Bierce: original The Devil's Dictionary, author’s preface and entries Historian, History and Satire. The preface declares a taste for wit; the Satire entry distinguishes wit from humor. Our broader case for humor and art below is our editorial philosophy, not a quotation or empirical claim attributed to Bierce. These are satire, not a reliable literal assertion that historical inquiry is false. Use chapter/entry locators, not an invented print page. The reading below is our interpretation of his satirical method; historical text and generated commentary remain separate.
Zinn: read Chapter 1, Columbus, the Indians, and Human Progress, in a reproduction of the original chapter. The methodological passage explicitly includes young women in Lowell textile mills among its chosen perspectives. Corroborate with his 1998 interview, “History as a Political Act,” preserved by the Howard Zinn project, where he explains centering excluded people's experiences. The chapter mirror does not establish a print edition/page or permission to republish it; the draft paraphrases briefly and makes neither claim. These citations establish Zinn's declared method, not automatic endorsement of every historical claim in his book.
Historical practice: the American Historical Association's professional standards expressly discuss perspective alongside evidence and critical scrutiny. Do not assert that every other historian pretends to have no viewpoint or that nothing objective can be known.
Comparison guardrail:Wikipedia's page-history documentation already describes revision records, contributors and diffs. We can learn from that interface; do not describe Wikipedia as only anonymous editors without accountability. The essay needs no unsupported claims about Instagram's intent or simplified ranking mechanisms; omit them.
All reference-based wording in the draft below is paraphrase except Gray's four-word quotation. These are conceptual influences, not claims of endorsement by the authors or a unified school of thought. Keep the sources attached to the sentences they support.
Draft editorial copy — “How we choose, and how you can question it”
This dictionary takes its central literary inspiration from Ambrose Bierce’s The Devil’s Dictionary: a dictionary can be an instrument of satire as well as explanation. Its farcical reversals invite us to notice what respectable language conceals. We want room for that kind of wit, alongside art, poetry and carefully sourced definitions. Bierce’s original text and preface.
An encyclopedia makes choices before a reader reaches its first sentence. Which sources belong here? Which definition comes first? Which picture makes an unfamiliar idea easier to grasp, and which example makes it look more settled than it is? Our aim is to make those choices visible and answerable. There is no editorial “view from nowhere”: selecting sources and arranging a narrative already involves perspective. A declared perspective does not remove bias; it gives readers something specific to examine. Scholarship can still test factual claims and arguments against evidence.
We distinguish three kinds of work. Factual claims need evidence: whether a passage appears in a particular work, who is credited, what a source actually asserts. Interpretations ask what that material means or illustrates; some interpretations are much better supported than others. Selection decides which of the eligible materials receives attention. Moving a quotation to the top of a page is an editorial act, not new evidence that its interpretation is correct.
This distinction matters especially in history. A record can support a date, an action or a documented statement while leaving room for disagreement about causes, significance and whose experience should organize the account. Historians already examine such questions. The American Historical Association describes perspective as part of historical work while requiring evidence and scrutiny of competing interpretations. Acknowledging a viewpoint does not make fabrication acceptable or every explanation equally good. Professional standards.
Humor and art can illuminate hypocrisy, motives, contradictions and lived experience more effectively—and sometimes more persuasively—than prose that presents itself as detached and universal. A joke can make an evasion visible; a poem or image can give a reader a way to recognize an experience that an abstract description leaves out. This is our editorial philosophy about expressive insight, not a universal finding that art is truer than scholarship. Persuasiveness is not verification: a brilliant joke can also be unfair, and an arresting image can mislead. Claims about what happened still need evidence.
Bierce’s satirical treatment of historians and history gives us one example of how a definition can challenge an institution’s self-image. We read it as an invitation to inspect language and power, not as proof that historical inquiry is worthless. Our default editorial rule puts an applicable, verified Devil’s Dictionary definition first, identifies its source and satirical character, and retains other definitions. Bierce-first expresses our aesthetic and editorial preference; it does not grant Bierce factual authority. Bierce’s original text, entries Historian and History.
Throughout this site, readers must be able to distinguish satire, an exact sourced historical quotation, a factual statement and generated editorial commentary. These labels answer different questions: a verified quotation can itself be satire, and verifying its wording does not establish its literal truth. New AI commentary is attributed to an inspired-by profile, never presented as something the historical writer said. The same distinctions apply to the opening section, profile pages and version history.
Howard Zinn makes another kind of choice explicit. In the opening chapter of A People's History of the United States, he describes organizing accounts around people pushed aside by narratives centered on rulers—including Indigenous people, enslaved people and the young women working in Lowell's textile mills. We draw a methodological lesson from that declaration: readers should be able to identify whose experiences guide an account and question its omissions. This does not require accepting every conclusion Zinn reaches. Chapter 1; Zinn discussing his approach.
John Gray offers a related challenge to sweeping collective stories. In a 2023 interview he says, “Humanity isn't an agent.” His point there concerns treating the human species as though it were one actor: people and groups have different, often conflicting purposes. For us, that prompts a question whenever a source speaks for everyone: whose purposes are being gathered into that claim? This application is ours; it is not Gray's endorsement of the project or a denial that evidence can establish facts. Interview transcript.
The proposed curation pilot uses five explicitly inspired-by AI profiles: Bierce, Voltaire, Vonnegut, Hitchens and Ursula K. Le Guin. We chose these names to test different editorial questions: where respectable language hides self-interest; where authority resists scrutiny; who bears the human consequences; whether an argument withstands challenge; and what arrangements or experiences a familiar story leaves out. These are our proposed rubrics to ground and test against selected writings, not complete descriptions of the authors' beliefs. A profile is neither a resurrected writer nor permission to put new words in that writer's mouth.
Those lenses have limits. Cynicism can miss cooperation; argumentative force can obscure uncertainty; an appealing alternative can outrun its evidence. Each dossier should disclose its sources, counterexamples, exclusions and documented historical prejudices rather than presenting a polished, infallible character. Five profiles using one model can repeat the same errors. Their agreement cannot stand in for independent sources, and no selection panel represents every reader or every experience.
Under the pilot policy, each profile has one vote. Four of the fixed five must support a recommendation; silence and abstention do not make the threshold easier. Evidence and rights requirements come first and cannot be outvoted. Eligible recommendations are ordered by a published procedure using support counts, declared evidence-relation categories, variety and a stable tie-break. These are editorial preferences, not percentages of truth. Routine publication is authorized by a finalized panel decision that meets consensus and passes the evidence and policy checks; it does not require human approval of every selection. An operator can make an explicit override, correction or withdrawal, and that action and reason remain distinguishable from the panel’s decision.
Accountability requires a record. We are building a history showing which configuration and profiles took part, what each proposed or supported, which objections remained, what authorized publication, and any optional human acceptance, change or withdrawal. The recorded explanations will be concise statements supplied by a model or person; they cannot promise access to every internal reason a model had. Source restrictions and privacy may require withholding details, including failed sensitive nominations. A lawful notice can explain an omission without repeating the restricted material.
Readers should be able to follow the evidence, compare versions and challenge an attribution, interpretation or editorial choice. A correction should leave a dated record of what changed and why, subject to the same privacy and rights limits. The method page will link the correction channel actually available at launch; it will not promise an open editing system that has not been built.
Where this stands: the existing application organizes sourced definitions and discovered material. The curated opening has a development-only manual-fixture implementation in draft PR #202; this does not establish public human approval or a running persona panel. The first composition/configuration persistence foundation is merged, which does not establish deployed curation. Persona recommendations, public dossier pages and the public history described here remain planned work. The first enabled panel will use the same five profiles across the pilot. Domain-specialist defaults and personal saved views are later proposals, not available settings. This status paragraph must be updated against the deployed release before publication, with working links only to capabilities that actually exist.
Implementation and content phases
Editorial review: review the draft above, sources and distinction between the authors' claims and our application. Record author/reviewer display identity, content version, change reason and checked-on date. Final copy must remain candid prose, not slogans, comparative attacks or sales claims. No invented quotations, biography facts or page numbers.
Early About / method page (independent of AI/history delivery): use existing informational-page/component conventions, accessible reading width/headings/source links and mobile/keyboard support. A maintained capability/status block comes from verified release facts; manual fixtures and generated fixture notes must not be described as human-approved curation. Add “Editorial method” to existing navigation/disclosure under Routing before launch: On overviews, standards-informed namespaces, and stable subject URLs #194's URL ownership, without new route taxonomy.
Corrections: link the implemented challenge path where it applies (for example the existing claim workflow for authorized contributors), state access limits honestly, and provide an owner-approved public correction contact if one is needed. Use an existing suitable contact if approved; a new correction tool is not a prerequisite for publishing the About page. If no public contact is available yet, state that limitation and omit the link. Choosing/implementing a fuller contact path remains a launch decision; do not invent an email address, promise anonymous edits or imply every reader has reviewer access.
Acceptance
About page can ship before AI/history, with accurate current/planned status and no dead future-feature links. Bierce’s satire is central; the humor/art principle is expressly editorial philosophy, not a universal empirical truth claim.
Satire, exact sourced quotation, factual statement and generated commentary are distinguishable across About, profiles and history; quotation accuracy does not imply literal truth.
Published copy distinguishes facts, interpretation, selection, recommendation, consensus-based publication and optional human intervention; it neither treats votes as truth nor denies checkable historical facts.
Source check confirms Zinn's Chapter 1 includes Lowell workers, Gray's quotation/context matches the dated interview, and Bierce's satire is labelled as such. No invented editions/page locators; clearly disclose the chapter reproduction and link primary corroboration.
Each offered profile has reviewed sources/rubric/limits before public use; five shared-model outputs are not marketed as independent evidence or the authors' own judgments.
Every “you can” capability claim is checked against the release; draft/manual-only/no-panel status is truthful. History/profile/correction links resolve with appropriate access, and future work is marked planned.
Editorial content has an attributable version/change record; restricted source bytes and private reviewer metadata do not enter the page. Focused navigation/accessibility tests and mix precommit when implemented.
No app files, model runtime, migrations or deployment changed during this research/specification update. PR #202's Phase 1 scope is preserved; this is a linked follow-up, not a request to retrofit a running AI system into its fixtures.
Post-MVP personality directory and dossier disclosure
The detailed public personality feature is owned by #203's Post-MVP public personality dossiers section, clarified with the owner on 2026-09-27. Link the method page to an actual active-personality directory and individual profile/version pages when shipped; each profile links to its historical subject's ordinary entity entry where available, while remaining explicitly a separate inspired-by AI profile.
The method explanation must distinguish (1) inherited pretrained model knowledge, whose full contents and causal influence we cannot enumerate, (2) the sourced dossier and editorial instructions we author, (3) the exact evidence/context supplied for a run, and (4) the recorded outputs and publication authority. Disclose actual model/version and fine-tuning status, bibliography versus actually supplied material, dossier changes and limits. A reviewed public instruction template is permitted; unrestricted private prompt dumps and raw internal reasoning are not. Do not present generated self-reports as proof of what the model knows or how every output arose.
This directory/profile/history UI is a post-MVP feature, not an expansion of the in-flight runtime task or a prerequisite to the early method page. Until implemented, describe it as planned and omit dead links.
Outcome and scope
Delivery child of #193: an early About / Editorial method page explaining declared perspective, checkable evidence, selection rules, limitations and accountability. Pair it with #203, the curation-history/weights/profile issue. This is researched draft copy and future implementation specification, not marketing and not a claim that planned features have shipped.
Use an informational page linked from the existing About/navigation and the curated-section disclosure. Coordinate exact URL/history/profile destinations with #194; this issue does not establish a new public namespace. Preserve #201's one-default MVP and later domain/personal roadmap. No new curator, public accounts, chooser, training or additional model runtime in this scope.
Delivery priority: publish this page early, after copy/source and navigation review. It can ship before the model runtime, saved compositions or public history. Describe those features as planned and omit their links until working destinations exist. This is planning priority, not authorization to implement the page in this issue-editing task.
Research record and attribution boundaries (checked 26 September 2026)
All reference-based wording in the draft below is paraphrase except Gray's four-word quotation. These are conceptual influences, not claims of endorsement by the authors or a unified school of thought. Keep the sources attached to the sentences they support.
Draft editorial copy — “How we choose, and how you can question it”
This dictionary takes its central literary inspiration from Ambrose Bierce’s The Devil’s Dictionary: a dictionary can be an instrument of satire as well as explanation. Its farcical reversals invite us to notice what respectable language conceals. We want room for that kind of wit, alongside art, poetry and carefully sourced definitions. Bierce’s original text and preface.
An encyclopedia makes choices before a reader reaches its first sentence. Which sources belong here? Which definition comes first? Which picture makes an unfamiliar idea easier to grasp, and which example makes it look more settled than it is? Our aim is to make those choices visible and answerable. There is no editorial “view from nowhere”: selecting sources and arranging a narrative already involves perspective. A declared perspective does not remove bias; it gives readers something specific to examine. Scholarship can still test factual claims and arguments against evidence.
We distinguish three kinds of work. Factual claims need evidence: whether a passage appears in a particular work, who is credited, what a source actually asserts. Interpretations ask what that material means or illustrates; some interpretations are much better supported than others. Selection decides which of the eligible materials receives attention. Moving a quotation to the top of a page is an editorial act, not new evidence that its interpretation is correct.
This distinction matters especially in history. A record can support a date, an action or a documented statement while leaving room for disagreement about causes, significance and whose experience should organize the account. Historians already examine such questions. The American Historical Association describes perspective as part of historical work while requiring evidence and scrutiny of competing interpretations. Acknowledging a viewpoint does not make fabrication acceptable or every explanation equally good. Professional standards.
Humor and art can illuminate hypocrisy, motives, contradictions and lived experience more effectively—and sometimes more persuasively—than prose that presents itself as detached and universal. A joke can make an evasion visible; a poem or image can give a reader a way to recognize an experience that an abstract description leaves out. This is our editorial philosophy about expressive insight, not a universal finding that art is truer than scholarship. Persuasiveness is not verification: a brilliant joke can also be unfair, and an arresting image can mislead. Claims about what happened still need evidence.
Bierce’s satirical treatment of historians and history gives us one example of how a definition can challenge an institution’s self-image. We read it as an invitation to inspect language and power, not as proof that historical inquiry is worthless. Our default editorial rule puts an applicable, verified Devil’s Dictionary definition first, identifies its source and satirical character, and retains other definitions. Bierce-first expresses our aesthetic and editorial preference; it does not grant Bierce factual authority. Bierce’s original text, entries Historian and History.
Throughout this site, readers must be able to distinguish satire, an exact sourced historical quotation, a factual statement and generated editorial commentary. These labels answer different questions: a verified quotation can itself be satire, and verifying its wording does not establish its literal truth. New AI commentary is attributed to an inspired-by profile, never presented as something the historical writer said. The same distinctions apply to the opening section, profile pages and version history.
Howard Zinn makes another kind of choice explicit. In the opening chapter of A People's History of the United States, he describes organizing accounts around people pushed aside by narratives centered on rulers—including Indigenous people, enslaved people and the young women working in Lowell's textile mills. We draw a methodological lesson from that declaration: readers should be able to identify whose experiences guide an account and question its omissions. This does not require accepting every conclusion Zinn reaches. Chapter 1; Zinn discussing his approach.
John Gray offers a related challenge to sweeping collective stories. In a 2023 interview he says, “Humanity isn't an agent.” His point there concerns treating the human species as though it were one actor: people and groups have different, often conflicting purposes. For us, that prompts a question whenever a source speaks for everyone: whose purposes are being gathered into that claim? This application is ours; it is not Gray's endorsement of the project or a denial that evidence can establish facts. Interview transcript.
The proposed curation pilot uses five explicitly inspired-by AI profiles: Bierce, Voltaire, Vonnegut, Hitchens and Ursula K. Le Guin. We chose these names to test different editorial questions: where respectable language hides self-interest; where authority resists scrutiny; who bears the human consequences; whether an argument withstands challenge; and what arrangements or experiences a familiar story leaves out. These are our proposed rubrics to ground and test against selected writings, not complete descriptions of the authors' beliefs. A profile is neither a resurrected writer nor permission to put new words in that writer's mouth.
Those lenses have limits. Cynicism can miss cooperation; argumentative force can obscure uncertainty; an appealing alternative can outrun its evidence. Each dossier should disclose its sources, counterexamples, exclusions and documented historical prejudices rather than presenting a polished, infallible character. Five profiles using one model can repeat the same errors. Their agreement cannot stand in for independent sources, and no selection panel represents every reader or every experience.
Under the pilot policy, each profile has one vote. Four of the fixed five must support a recommendation; silence and abstention do not make the threshold easier. Evidence and rights requirements come first and cannot be outvoted. Eligible recommendations are ordered by a published procedure using support counts, declared evidence-relation categories, variety and a stable tie-break. These are editorial preferences, not percentages of truth. Routine publication is authorized by a finalized panel decision that meets consensus and passes the evidence and policy checks; it does not require human approval of every selection. An operator can make an explicit override, correction or withdrawal, and that action and reason remain distinguishable from the panel’s decision.
Accountability requires a record. We are building a history showing which configuration and profiles took part, what each proposed or supported, which objections remained, what authorized publication, and any optional human acceptance, change or withdrawal. The recorded explanations will be concise statements supplied by a model or person; they cannot promise access to every internal reason a model had. Source restrictions and privacy may require withholding details, including failed sensitive nominations. A lawful notice can explain an omission without repeating the restricted material.
Readers should be able to follow the evidence, compare versions and challenge an attribution, interpretation or editorial choice. A correction should leave a dated record of what changed and why, subject to the same privacy and rights limits. The method page will link the correction channel actually available at launch; it will not promise an open editing system that has not been built.
Where this stands: the existing application organizes sourced definitions and discovered material. The curated opening has a development-only manual-fixture implementation in draft PR #202; this does not establish public human approval or a running persona panel. The first composition/configuration persistence foundation is merged, which does not establish deployed curation. Persona recommendations, public dossier pages and the public history described here remain planned work. The first enabled panel will use the same five profiles across the pilot. Domain-specialist defaults and personal saved views are later proposals, not available settings. This status paragraph must be updated against the deployed release before publication, with working links only to capabilities that actually exist.
Implementation and content phases
Acceptance
mix precommitwhen implemented.No app files, model runtime, migrations or deployment changed during this research/specification update. PR #202's Phase 1 scope is preserved; this is a linked follow-up, not a request to retrofit a running AI system into its fixtures.
Post-MVP personality directory and dossier disclosure
The detailed public personality feature is owned by #203's Post-MVP public personality dossiers section, clarified with the owner on 2026-09-27. Link the method page to an actual active-personality directory and individual profile/version pages when shipped; each profile links to its historical subject's ordinary entity entry where available, while remaining explicitly a separate inspired-by AI profile.
The method explanation must distinguish (1) inherited pretrained model knowledge, whose full contents and causal influence we cannot enumerate, (2) the sourced dossier and editorial instructions we author, (3) the exact evidence/context supplied for a run, and (4) the recorded outputs and publication authority. Disclose actual model/version and fine-tuning status, bibliography versus actually supplied material, dossier changes and limits. A reviewed public instruction template is permitted; unrestricted private prompt dumps and raw internal reasoning are not. Do not present generated self-reports as proof of what the model knows or how every output arose.
This directory/profile/history UI is a post-MVP feature, not an expansion of the in-flight runtime task or a prerequisite to the early method page. Until implemented, describe it as planned and omit dead links.