What Makes a Revolution Website Trustworthy? A Credibility Architecture for BuiltToResist

Full report

On this page

Research reviewed through September 19, 2026.

Executive finding

A radical political research site should not try to look neutral. It should try to be auditable.

For skeptical readers, the strongest credibility posture is not “trust us because we are unbiased.” It is: “Here is our political standpoint; here is exactly what we know; here is how we know it; here is what is disputed; here is what we are inferring; here is what could make us wrong; and here is the original evidence so you can check us.” That approach is consistent with established journalism norms emphasizing original sources, clear source identification, labeled advocacy, transparent methods, visible corrections, and disclosure of conflicts. The Trust Project likewise treats organizational standards, author expertise, content labels, references, methodology, and correction practices as core indicators readers can inspect.

That conclusion also fits the limited empirical research available on transparency interfaces. A 2021 peer-reviewed qualitative study interviewed 15 journalists and 16 news consumers—not enough to establish population-wide effects, so the result should be treated as directional rather than definitive. Participants nevertheless repeatedly wanted access to source material, information about verification and anonymous sourcing, correction histories, reporting processes, and conflicts of interest. The study's authors described the underlying shift as moving readers from a “trust me” model toward one in which claims can be examined.

For BuiltToResist, that means ideological candor should be treated as an asset rather than something to conceal. The site can explicitly advocate political principles while imposing unusually strict evidentiary rules on its descriptive claims. SPJ explicitly recommends labeling advocacy and commentary, while the Trust Project asks publishers to distinguish material designed to persuade from impartial reporting.

The resulting institutional promise should be:

BuiltToResist has a point of view, but facts do not receive partisan privileges. Evidence is cited regardless of whether it helps our argument. Uncertainty is disclosed rather than hidden. Competing explanations are represented in proportion to their evidentiary strength. Historical evidence is not presented as prediction. Scenarios are not forecasts. Substantive corrections remain visible. Readers can inspect the evidence behind consequential claims.

That is a stronger credibility proposition for an openly ideological research project than generic assurances of “accuracy” or “independence.” Reuters' public standards illustrate why: its sourcing rules call for named sources where possible, cross-checking, examination of source motives, talking to different sides, explicit acknowledgment of what is unknown, and—in investigative work—efforts to disprove as well as prove a working theory.

The crucial architectural idea is therefore epistemic separation. BuiltToResist should make it visually difficult to confuse five different things:

what happened; what a source says happened; what BuiltToResist infers from the evidence; what BuiltToResist believes ought to happen; and what BuiltToResist thinks might happen next.

Those are different kinds of statements, and trustworthy political research should render them differently.

What skeptical readers need to see

Skeptical readers have a special problem with a radical site: they reasonably expect motivated reasoning. BuiltToResist cannot eliminate that prior suspicion through branding. It can, however, make motivated reasoning expensive by forcing its own arguments through public evidence constraints.

The Trust Project's framework is useful here because its eight indicators are overwhelmingly inspectable rather than ceremonial: mission and funding, author expertise, content labels, references sufficient to check claims, reporting methods, knowledge of affected communities, breadth of perspectives, and a visible mechanism for feedback and corrections. For controversial or investigative stories, it specifically asks whether sources are provided for individual claims and whether enough information is given for readers to check them.

A second useful model comes from analytical institutions rather than journalism. The Congressional Budget Office says its transparency program is intended to make analyses understandable, show how estimates might change under different circumstances, and increase credibility by exposing underlying data, research, and expert feedback. It publishes methodological explanations and, for some analyses, code that permits replication. CBO separately discusses uncertainty and limitations so readers can understand why estimated outcomes might differ.

The design implication is that BuiltToResist should build trust at four layers simultaneously:

The strongest rule is progressive disclosure: the ordinary reader should see a simple trust summary, while a hostile or expert reader should be able to drill all the way down to the original document. Transparency research itself identifies a tension between giving readers enough nuance and overwhelming them; a layered interface resolves much of that tension.

A recommended article header would therefore look conceptually like this:

Research analysis

By Maya Example · Reviewed by Jordan Example

Published May 8, 2026 · Substantively reviewed August 21, 2026

Evidence current through August 15, 2026

Status: Current

Evidence confidence: High — six independent sources, including three primary records; no unresolved contradiction affecting the central conclusion

Perspective: Written from BuiltToResist's stated democratic-socialist institutional framework

Controlling sources: 5

Corrections: 1 clarification, no factual corrections

[Evidence] [Methods] [Competing explanations] [Revision history]

The point is not decorative transparency. Each item must resolve to evidence.

Named authorship should be the default because expertise and accountability matter to readers; the Trust Project explicitly asks whether authors' credentials and subject knowledge are readily available. For a politically sensitive site, however, a categorical real-name requirement would be unnecessarily rigid. A contributor facing a credible security or retaliation concern could use a stable pseudonym, provided BuiltToResist discloses that the person's identity and qualifications have been verified by a named accountable editor. The exception should read something like: “The author's identity is withheld for documented safety reasons; identity and relevant expertise were verified by the research editor.” Anonymity should be a disclosed exception, not an invisible norm. That follows the broader journalistic principle that anonymity requires a reason readers can evaluate.

Funding deserves similar prominence. For an ideological project, readers should be able to determine who owns the publication, its legal form, major funding channels, whether donors can direct editorial coverage, what conflicts authors must disclose, and how sponsorship is separated from research. The Trust Project explicitly includes mission, priorities, and funding among its institutional trust questions, while SPJ calls for disclosure of unavoidable conflicts and separation of editorial material from sponsored content.

The BuiltToResist credibility architecture

The answers to the site's core policy questions can be made unusually concrete.

These rules are stricter than merely attaching a bibliography. SPJ recommends original sources when possible, clear source identification, access to relevant source material, and continuing updates and corrections. Reuters adds cross-checking, multiple sources where practical, evaluation of source motive and position, and explicit disclosure of knowledge gaps.

BuiltToResist should formalize a source hierarchy, but not mistake hierarchy for automatic truth. A sensible default is:

Tier A — controlling primary authority. Original law, court opinion, regulation, government dataset, meeting record, contract, contemporaneous document, direct interview, raw dataset, research paper, official statistical release, or authenticated archival record.

Tier B — high-quality independent analysis. Serious scholarship, established investigative reporting, systematic reviews, reputable institutional analyses, professional technical work.

Tier C — interested-party evidence. Political organizations, companies, unions, advocacy groups, campaigns, government press offices, activist groups, think tanks, litigants. Such sources can be indispensable evidence of what an actor says, did, filed, spent, or published, but their interpretation should not automatically be treated as independent corroboration.

Tier D — discovery leads. Social posts, aggregators, unsourced summaries, search snippets, crowd-edited pages, newsletters, and commentary. They may identify a lead, but consequential claims should be traced upstream before publication.

A primary source should likewise never be labeled “true because primary.” The Library of Congress defines primary sources as the raw materials of history—original documents and objects associated with the period under study. That makes them especially valuable for provenance, not infallibility. A police report, revolutionary manifesto, corporate filing, intelligence assessment, union leaflet, party speech, memoir, and court complaint are all primary evidence for different propositions, but each may contain error, strategic framing, or self-interest. The citation should therefore tell the reader what proposition the source actually establishes.

BuiltToResist should also create a claim taxonomy and render it visibly:

This taxonomy prevents the most damaging category error on political sites: presenting a morally forceful conclusion as though it were an empirical finding, or presenting an empirical pattern as though it logically dictated a moral conclusion.

Contested claims, ideology, historical analogy, and the future

Contested claims should be visibly contested at the point of use. Reuters' standards call for speaking to different sides of disputes, giving enough context to evaluate sources, making unknowns explicit, and trying to disprove as well as prove an investigative hypothesis. That does not require artificial 50/50 “balance.” BuiltToResist should instead distinguish credible competing explanations from mere disagreement.

A good contested-claim box would read:

What is contested? Whether the decline was primarily caused by policy A or market factor B.

Evidence favoring A: …

Evidence favoring B: …

Evidence inconsistent with both: …

Our assessment: A currently explains more of the evidence, but the timing evidence does not distinguish the mechanisms cleanly.

Confidence: Medium.

What would change our assessment: release of the agency's underlying transaction data.

This accomplishes something a pile of citations cannot: it exposes the reasoning junction where reasonable researchers might diverge.

The site should adopt a rule that central investigations undergo an adversarial evidence pass before publication: What evidence would undermine the thesis? What is the strongest alternative explanation? Which source is most vulnerable to bias? Is an apparently independent corroborating source actually repeating the first source? Reuters expressly recommends attempting to disprove one's own initiative reporting, and intelligence analytic standards similarly emphasize alternatives, contrary information, assumptions, and explanations for why judgments change.

Ideological arguments should be explicit rather than cosmetically neutral. A BuiltToResist essay might therefore say:

Content type: Normative argument

This essay argues from the premise that concentrated unaccountable political power is undesirable. The historical and empirical claims supporting the argument are subject to BuiltToResist's normal sourcing standards; the underlying value judgment is not presented as an empirical finding.

That distinction follows established recommendations to label advocacy and opinion distinctly. It also permits disagreement at the correct level: a reader can accept the facts while rejecting the value premise, challenge the empirical premise without denying the value, or contest the reasoning connecting them.

Historical analogies require particularly aggressive qualification. Analogies can be rhetorically potent; recent experimental political-science research found that historical analogies can increase public confidence in leaders' proposed foreign-policy reasoning, underscoring why an advocacy-oriented site should not treat resemblance as self-validating proof. Scholarship and historical-method discussions likewise caution that analogies are aids to interpretation rather than substitutes for historical comparison and explanation.

Every important analogy should therefore answer four questions:

What is actually analogous? Name the mechanism rather than merely invoking an evocative era: institutional capture, capital flight, military fragmentation, strike coordination, censorship, elite defection, inflation, constitutional breakdown, and so forth.

What is materially different? Different state capacity, communications technology, international environment, class structure, electoral institutions, coercive capacity, demographics, economic structure, or legal system may destroy the analogy.

What prediction, if any, follows? Often none. Historical resemblance can suggest hypotheses without establishing recurrence probabilities.

What other historical comparison points the other way? A reader should see whether case selection is being driven by the desired political lesson.

A recommended label is: “Historical analogy. Used to compare mechanisms, not to claim that history will repeat or that the same outcome is probable.”

That same rigor should govern futures work. The UK Government's Futures Toolkit explicitly says scenarios are neither plans nor predictions: they are multiple possible futures used to explore uncertainty and test decisions. It also stresses that no single scenario should be expected simply to “come true.”

BuiltToResist should consequently prohibit language such as “we estimate a 70% scenario” unless the work is actually a probabilistic forecast. A scenario estimate may contain quantities conditional on assumptions—“If national turnout falls to X and support among group Y changes by Z, this scenario produces …”—but those numbers do not give the scenario itself a 70% chance of occurring.

A forecast, in contrast, should disclose:

Forecast horizon: through December 31, 2027

Predicted event: X

Likelihood: 55–70%

Evidence confidence: Medium

Method: reference-class model plus specified adjustments

Key assumptions: A, B, C

Last updated: date

Resolution rule: exactly what outcome counts as correct

Performance history: prior forecasts made under the same methodology

The distinction between likelihood and confidence is especially important. Intelligence Community analytic standards treat them as separate concepts: likelihood concerns the chance of an outcome, while confidence concerns the quality and sufficiency of the evidentiary and reasoning basis for the judgment. The standards also call for identifying key assumptions, explaining uncertainties, considering alternative hypotheses, and acknowledging significant contrary information.

BuiltToResist should borrow that separation without copying government jargon wholesale.

A statement such as:

70% likely, low confidence

is coherent: available evidence weakly supports a probability estimate centered around 70%.

A statement such as:

High confidence

does not mean “90% probability.” It means the evidence underlying an evidentiary judgment is unusually strong.

For ordinary historical or factual research, probabilities should generally be avoided unless the methodology genuinely warrants them. A simple three-level evidence-confidence system is preferable:

Every label should have a one-sentence explanation. A naked “High confidence” badge invites authority theater; “High confidence because…” creates accountability.

Sources, citations, persistence, and legal evidence

The central citation principle should be claim-to-evidence proximity. End-of-page bibliographies are useful, but they are insufficient for controversial research because a skeptical reader should not have to infer which of 48 references supports a specific sentence. The Trust Project specifically asks whether sources are supplied for individual claims in controversial or investigative work and whether enough detail is given to check them.

BuiltToResist should therefore combine three citation layers.

Inline claim citations answer “what supports this sentence?”

The Evidence Spine answers “what sources does the overall conclusion depend on?”

The full source ledger answers “what exactly did the researchers inspect?”

The Evidence Spine is especially important. Every substantive research page should expose its controlling sources—meaning sources whose removal, falsification, or material reinterpretation would force a change in one or more central conclusions. A reader might see:

Controlling sources

Department dataset, release 2026-Q2 — establishes the denominator used in the central calculation.

District Court opinion, pp. 41–47 — establishes the court's actual finding.

Collective-bargaining agreement, Art. VIII § 3 — establishes the contractual rule.

Census microdata extract — establishes the demographic baseline.

This is better than a generic “Sources” box because it tells readers how each source controls the argument.

The underlying source record should function like a lightweight archival catalog:

Persistent identifiers should be preferred over ordinary URLs when the source ecosystem supplies them; DOI infrastructure is specifically designed around persistent identification and associated metadata. Web archiving should supply a second layer of resilience. Harvard's Library Innovation Lab describes its Perma-related tools as infrastructure developed with libraries, courts, and legal scholars to preserve citations, with emphasis on fidelity, specificity, and provenance; its capture tooling can create archival files with detailed provenance.

A content hash supplies a third, narrower safeguard. NIST's Secure Hash Standard explains that message digests can be used to detect whether content has changed after the digest was generated. A BuiltToResist SHA-256 fingerprint should therefore mean only: “this is the byte sequence we captured.” It must not be described as proving that a document is authentic, accurate, or truthful.

The architecture that makes citations survive URL changes is consequently:

Article citation → stable BuiltToResist source ID → source registry → current canonical location + persistent ID + archived capture + metadata.

The article should never depend solely on a third party's current URL. When the publisher moves a PDF, BuiltToResist updates the resolver record while the article's citation remains unchanged.

For politically sensitive web evidence, archived capture is particularly valuable because web content can disappear or change. But the archive should complement, not conceal, the original URL: readers should be able to see where the material originated.

Legal claims require their own citation discipline. A statement about “what the law is” should identify the jurisdiction and the date through which the law was checked. The citation should point to the actual constitutional provision, statute and subsection, regulation and section, controlling case with a pinpoint page or paragraph, official order, or other controlling legal authority—not merely a newspaper explanation of it. Public legal-analysis resources can then provide context; for example, the Library of Congress's Constitution Annotated describes its analysis as grounded in a comprehensive review of Supreme Court case law.

A legal-analysis page should display something like:

Jurisdiction: United States federal law

Law checked through: September 15, 2026

Authority status: Binding Supreme Court precedent / enacted statute / proposed bill / pending appeal

Controlling authority: full legal citation + pinpoint

Our interpretation: separately labeled

Open legal question: where courts or authorities disagree

Original text: accessible official source when available

This prevents another common political-research error: citing a bill as though it were enacted law, an allegation as though it were a judicial finding, a trial-court opinion as though it bound the entire country, or a commentator's interpretation as though it were statutory text.

Original reports should be reader-accessible whenever lawful and practicable. That means providing the publisher's original, a persistent identifier, an archived copy where lawful, and a direct pinpoint. Where the original is a difficult scan, BuiltToResist can offer an accessible transcript or extracted text clearly marked as a convenience copy, while retaining the scan as controlling. Where copyright prevents republication, it should provide precise metadata and the lawful publisher/library location rather than silently hosting a copied work.

Accessibility should also mean usable by disabled readers: evidence should not be trapped solely inside an untagged image or screenshot. At minimum, key charts need text explanations, images need meaningful alternatives where appropriate, and any BuiltToResist-created PDF should have a usable HTML equivalent whenever feasible.

Currentness, obsolescence, and corrections

Political research decays at different rates. An eighteenth-century primary document may remain evidentially stable for decades; a statutory explainer can become wrong the moment a court or legislature changes the rule; a polling analysis can be stale in days. A single universal “last updated” timestamp cannot express those differences.

BuiltToResist should expose three dates whenever they are materially distinct:

Published: when the page first became public

Last substantive review: when a human reviewer last reconsidered its accuracy

Evidence current through: the latest date covered by the underlying evidence

For law, use “Law checked through.” For databases, use “Data through.” For a historical project, use “Sources reviewed through.” For a scenario, use “Assumptions frozen as of.”

This is consistent with the broader journalism principle that information should be gathered, updated, and corrected throughout a story's life and that later, more complete information may need to be provided.

The visible currentness states should be:

Current — reviewed against the site's current standard and no known material superseding evidence.

Review due — the site's review interval or an event trigger has been reached; the material is not known to be wrong, but currentness has not recently been verified.

Partially outdated — some data or context has aged, but substantial historical content remains useful. The banner identifies which sections are affected.

Superseded — a later article, dataset, law, methodology, or correction materially replaces the analysis. The old page remains visible but points prominently to its successor.

Archived — retained for the historical record and not represented as current guidance or current analysis.

Obsolescence should be triggered both by time and events. Relevant event triggers include a controlling source being withdrawn, a law changing, a court reversing a decision, release of materially newer data, discovery that a source was misrepresented, or a forecast reaching its resolution date.

A superseded page should generally remain available. Trust suffers when old claims disappear without an audit trail. The page should instead begin with a conspicuous warning:

SUPERSEDED — Do not rely on this page as the current BuiltToResist analysis.

This article reflected evidence available through March 2024. New evidence published in January 2026 changed our assessment. See the current analysis. The original text and revision history remain available for accountability.

Corrections should be similarly explicit. Both Reuters and SPJ call for prompt, transparent corrections; Reuters specifies that errors should be rectified clearly and comprehensively, while SPJ recommends prominent acknowledgment and careful explanation. The Trust Project likewise treats quick, clear, prominent corrections as an observable trust indicator.

BuiltToResist should distinguish four classes:

A substantive correction notice should answer seven questions:

What did we originally say?

What was wrong?

What does the evidence now show?

What changed in the article?

When was it changed?

Did the correction change our conclusion or confidence level?

Which other BuiltToResist pages depended on the erroneous claim?

That last question is important enough to automate. Claims and sources should have internal stable IDs so that when Source S-248 is withdrawn or Claim C-110 is corrected, editors can identify all pages that inherit the problem.

The site should publish a global corrections ledger, sortable by date and severity. Readers should be able to see not only that BuiltToResist makes mistakes but how it reacts to them. The 2021 transparency-interface study specifically found interest in displaying both the existence and nature of corrections rather than treating correction merely as a binary badge.

Corrections should also preserve the distinction between “our evidence changed” and “we made an error.” Revising a forecast because new information arrived is not necessarily a correction; silently rewriting the old forecast to resemble the new one would be. CBO's practice of evaluating projections against eventual results offers the right model: predictive credibility depends not only on publishing current estimates but on confronting prior performance.

Public editorial and research methodology for BuiltToResist

The following is suitable, with organizational details filled in, as the site's public methodology. It communicates research standards without publishing private prompts, internal security controls, source-protection procedures, unpublished notes, or system-steering documentation.

BuiltToResist Editorial and Research Methodology

Our standpoint

BuiltToResist publishes research from an openly stated political perspective. We do not claim that researchers are free of values, interests, or prior beliefs. We do claim that factual evidence should be treated consistently whether it strengthens or weakens our political arguments.

We distinguish empirical reporting from interpretation, interpretation from normative argument, and all three from speculation about the future. Advocacy does not lower our evidentiary standard.

What readers should be able to audit

Substantive research pages identify their author or accountable editorial author, publication date, evidence cutoff, currentness status, content type, controlling sources, and material corrections. Where an article depends on a specialized method, dataset, model, calculation, or selection procedure, we explain that method at a level sufficient for an informed outside reader to understand what was done and, where practical, reproduce the result.

We publish information about our mission, funding, governance, conflicts policy, correction policy, and editorial standards. These disclosures describe the rules governing our published work; they do not require publication of private source identities, confidential communications, security procedures, unpublished editorial deliberations, proprietary credentials, or internal system instructions.

How we classify statements

We distinguish documented facts, attributed claims, disputed claims, analytical inferences, quantitative estimates, legal interpretations, normative arguments, forecasts, and scenarios.

A citation to a source does not automatically turn an allegation into a fact. When a document proves only that an actor said something, we characterize it as an attributed statement unless independent evidence establishes the underlying proposition.

How we source factual claims

We prefer original records and primary evidence when they are available and relevant. Important claims should be traceable to the evidence that supports them. For consequential, surprising, accusatory, or politically contested claims, we seek independent corroboration where practicable and disclose when meaningful corroboration is unavailable.

We evaluate sources according to directness, expertise, independence, documentary support, currentness, incentives, track record, and whether apparently separate sources actually depend on a common origin.

Secondary journalism and scholarship are valuable for context, interpretation, and discovery, but we attempt to trace consequential claims back to their underlying records rather than citing a chain of summaries.

How we handle anonymous and confidential sources

Named sourcing is preferred. We grant confidentiality when disclosure could create a meaningful risk or when important information unavailable by reasonable open means justifies it. When source protection permits, we describe enough about an anonymous source's position, access to the information, potential interests, and basis of knowledge for readers to evaluate the claim.

A claim whose credibility depends heavily on one anonymous source will be identified as such and normally receives lower evidence confidence than the same claim supported by multiple independent records.

How we present disagreement

We do not hide material contrary evidence. When credible evidence supports competing explanations, the article identifies the disagreement, states what evidence favors each explanation, and explains our assessment.

We do not assign equal weight to every conceivable view merely because somebody holds it. Alternatives receive attention in proportion to their relevance, evidentiary support, and capacity to change the conclusion.

When the evidence does not permit a responsible resolution, we say that the issue remains unresolved.

How we separate politics from evidence

Essays whose principal purpose is to advocate a political, moral, institutional, or strategic position are labeled as argument or commentary. Their empirical premises remain subject to the same sourcing rules as research articles.

Normative conclusions are identified as such. We do not describe a value judgment as though it were produced by data alone.

How we use history

Historical claims cite historical evidence. Historical analogy is separately labeled.

When a present-day analysis invokes an earlier revolution, movement, collapse, strike wave, constitutional crisis, authoritarian transition, reform campaign, or other historical episode, we identify both the relevant similarities and the important differences. An analogy is not, by itself, evidence that the same outcome will recur.

Where an analogy supports a causal argument, we state the proposed mechanism and consider alternative comparisons. We do not convert historical resemblance into a numerical probability without an independent forecasting methodology that warrants doing so.

How we discuss the future

A scenario is not a forecast. Scenarios describe conditional possible futures in order to test assumptions, consequences, vulnerabilities, or strategies. Unless explicitly supported by a forecasting method, scenarios are not assigned probabilities.

A forecast makes a claim about the likelihood of a future event. Forecasts identify the prediction, time horizon, methodology, assumptions, probability or range where appropriate, evidence cutoff, and resolution criterion. When forecasts resolve, we preserve and evaluate their original form rather than rewriting them retrospectively.

How we express uncertainty

We distinguish confidence in the evidence from probability that a future outcome will occur.

“High,” “Medium,” and “Low” evidence-confidence labels describe our assessment of the quality, directness, independence, completeness, and consistency of the evidence supporting a judgment. They are not mathematical probabilities. Important confidence labels include a brief explanation.

We disclose consequential assumptions, missing evidence, unresolved contradictions, and information that could change our assessment.

How we cite law

Legal claims identify the relevant jurisdiction and, where current law matters, the date through which the law was checked. We cite controlling constitutional, statutory, regulatory, or judicial authority directly whenever feasible, with section, paragraph, page, docket, or equivalent pinpoint references.

We distinguish enacted law from proposed legislation; binding authority from persuasive authority; allegations from judicial findings; and the text of an authority from our interpretation of it.

How we preserve sources

Our citations are attached to stable source records rather than depending exclusively on live URLs. Source records may include bibliographic information, document identifiers, persistent identifiers, original locations, archived captures, version information, pinpoint locations, access dates, and file fingerprints.

When a web address changes, we update the source record without changing the identity of the citation. When legally permitted and useful, we preserve archival captures of web evidence. A file hash records the identity of the captured digital file; it does not independently establish the truth or authenticity of its contents.

How we make evidence accessible

Where practicable, readers can reach the original report, dataset, legal authority, document, transcript, or archive behind a consequential claim. We prefer public and accessible versions where equivalent authoritative copies exist.

When an original source cannot be redistributed because of copyright, confidentiality, safety, or access restrictions, we provide as much bibliographic and verification information as responsibly possible and explain the restriction.

How we keep material current

Substantive pages display when they were published, when they were last substantively reviewed, and the date through which their evidence is current when that distinction matters.

Pages may be labeled Current, Review due, Partially outdated, Superseded, or Archived. Superseded research remains available when possible so that readers can reconstruct the development of our analysis.

How corrections work

We correct substantive errors openly. A factual correction states what was wrong, what has changed, and the date of the correction. When the correction materially changes a conclusion or confidence assessment, we say so prominently.

We distinguish ordinary copyedits from clarifications, factual corrections, and major corrections or retractions. We do not silently alter substantive claims in a way that conceals a prior error.

We maintain a public corrections record and provide a mechanism for readers, subjects, and experts to challenge our evidence.

How our judgments change

Changing a conclusion in response to stronger evidence is not a failure of consistency. It is a requirement of evidence-based research. When a major BuiltToResist assessment changes, we explain what changed in the evidence, assumptions, or reasoning.

Our test for publication

Before publishing a consequential research conclusion, we ask: What evidence supports it? What evidence weighs against it? What is the strongest plausible alternative explanation? Which assumptions are doing the most work? Are our sources independent? Are we describing allegations as facts? Are we confusing historical analogy with prediction? What remains unknown? What evidence would cause us to change our mind?

If those questions cannot be answered honestly, the claim is not ready to be stated with confidence.

That policy draws from several durable credibility norms: original sourcing and attribution from SPJ; transparent sourcing, cross-checking, uncertainty, and correction from Reuters; visible labels, methods, authorship, references, and feedback from the Trust Project; explicit methods, underlying evidence, alternatives and uncertainty from CBO; and the analytic separation of confidence, likelihood, assumptions, and alternative hypotheses used in formal analytic standards.

The final design principle for BuiltToResist is therefore simple:

Do not ask skeptical readers to trust the site's conclusions. Give them enough structured evidence to audit how the conclusions were reached.

For a radical publication, ideological transparency plus methodological constraint is stronger than a performance of neutrality. The site becomes credible when a hostile reader can reproduce a calculation, locate a controlling document, identify where interpretation enters the argument, discover the strongest counterevidence, see that a prediction was actually made before the event, find an old error rather than watching it disappear, and determine exactly how new evidence changed the site's mind. That is the difference between a political website with citations and a genuine research institution.