Checked Before Published

How we check our work

A resource is only as trustworthy as its sources and its honesty. A single wrong verse reference or misattributed quotation undermines the whole — especially with the thoughtful skeptics we most want to reach. So our content is held to a deliberate review standard before it goes live.

We call that standard Checked Before Published. It is not a slogan; it is the set of steps below, and the figures further down are counted from our own repository rather than asserted. Where we have fallen short of it, the corrections are published at the foot of this page.

The review process

Each long-form Evidence Library essay passes through several independent stages before publication:

1. Drafting to a scholarly standard. The argument is stated carefully, its history traced, each premise defended in depth, the strongest objections answered by name, the state of the scholarly debate represented honestly, with numbered footnotes and a bibliography.
2. Independent fact-checking. Every quotation, date, statistic, scholar attribution, and Scripture reference is verified against primary and scholarly sources. Quotations are checked for accuracy; citations are checked for existence and correct attribution.
3. Scholarly & orthodoxy audit. A separate, automated review checks the essay for rigor, for fair representation of critics (steelmanning rather than strawmanning), and for conformity to historic Christian orthodoxy and our denominational neutrality.
4. Source verification & correction. Every flagged issue is corrected before publication. Where a claim cannot be sourced to a real, verifiable work, it is removed or hedged.
5. Editorial sign-off. Nothing is published until it has cleared the stages above and been signed off.

This is a multi-stage editorial review rather than outside academic peer review — we want to be precise about what we do and don't claim. What it guarantees is that our published essays have been independently fact-checked, audited for rigor and orthodoxy, and sourced to real, citable scholarship.

Where the review stands today

Figures counted directly from the site, not typed by hand — a build check fails if this list drifts out of step with what is actually published:

• 93 deep-dive essays and 110 answer pages published, each carrying a dated record of the reviews it passed — 110 of the 110 answers record both an argument and an orthodoxy review.
• All 156 passages in our public-domain source library have been checked word-for-word against the original text — and only checked passages may be quoted. A passage that fails the check is held back or removed rather than published: one was removed in July after the check found its wording spliced together from two different editions.
• 97 argument briefs — short summaries of our house framing that the assistant may draw on — every one of which cleared both review gates before it could be used.
• 117 automated checks run on every change to the site, alongside a separate scan of every published page of prose against a list of known heterodox phrasings. That scan is a coarse net rather than a judge: it fails the build when a new match appears, so a person has to look.

Who — and what — does the checking

You should know this before you weigh anything above. These reviews are run by purpose-built AI reviewers working to a fixed brief — the Nicene and Apostles' Creeds and our written statement of faith — under human supervision, with every finding applied by hand. That is how a very small team can gate everything in its two main content lines in full, rather than spot-checking it, and the reviewers are deliberately adversarial — one exists only to attack our own drafts for overstatement, another only for doctrinal drift. Their findings are frequently uncomfortable; several of the corrections listed further down came from exactly that.

We would rather tell you that plainly than let “independent review” imply a newsroom. Three honest limits follow from it:

• It is not outside academic peer review, and we do not claim it is.
• It is not pastoral sign-off. We are seeking a standing pastoral reviewer; until one is in place, we will not claim it.
• The doctrinal gate runs on every essay and answer page before it ships — not on every page of the site. One section, our conversion-story tab, has deliberately not yet had the full sweep, and the mastery drill pages are reviewed as they are next revised.

What our AI can and cannot do

The reasonable worry about any AI and Scripture is that it will invent a quotation, misattribute a scholar, or simply tell you what you want to hear. One of those we have addressed by construction; the others by review and by explicit instruction — and it is worth being precise about which is which, because the difference matters.

By construction: quotations from the creeds and the Church Fathers. When the assistant quotes a historic source, the only such text placed in front of it is drawn from passages our citation check has already verified word-for-word against the original. Unverified passages are not merely discouraged; they are absent, and the assistant is instructed to quote a Father or a creed from that list and from nowhere else. Construction guarantees what it is given; that it never reaches past that list is an instruction, and we count it as one. Every one of those passages is public-domain and readable in full on our Source Library page, so you can check any quotation against the original yourself — which is the point.

By instruction: modern scholars. For living scholars there is no such list, so there is no architectural guarantee — only a standing instruction to cite a work only where it is genuinely confident the work exists, and otherwise to name the author and the topic rather than invent a title. That is weaker than the constraint above, and we would rather say so than let you assume otherwise.

Its instructions are reviewed, and where we have house framing it draws on that. The instructions governing the assistant that answers your questions pass the same argument and orthodoxy review as any published page. Alongside them we supply short briefs distilled from essays that have already cleared those checks — but they are offered as optional background on the topics they cover, not as a script, and on most questions no brief applies at all. So the honest description is not that it answers from our material: it answers within reviewed instructions, informed by our reviewed framing where we have it.

What it is not. It is not a pastor, a counsellor, or a substitute for your church. Where a question is a cry for help rather than a request for an argument, the honest response is a person, not a paragraph — so it is built to say so, to point toward Christ's presence, and to point toward real care: a trusted friend, a pastor, or a professional counsellor, and findahelpline.com or your local emergency number if you are in danger.

What these checks have caught

A review process is only worth describing if it actually stops things. Publishing what it caught is better evidence than any assurance we could write, so here are real errors we found in our own published work and corrected, together with ones caught before they shipped:

• A comparison stated the wrong way round. An origin-of-life essay weighed an improbability against the universe's available trials backwards. It read persuasively and was simply incorrect. Corrected.
• Paraphrases dressed as direct quotations. Several argument cards put a scholar's position inside quotation marks when the wording was ours, not theirs. Every one was either sourced to the exact wording or rewritten plainly as our own summary.
• A creed labelled public domain that wasn't. A modern copyrighted translation of the Apostles' Creed was carried both in our source library and live on our statement-of-faith page, and a quotation from Ignatius had been spliced together from two separate passages. A citation check caught both; the creed was replaced with a verified public-domain text.
• Conceding more than the evidence required. Pages answering other worldviews had granted the opposing case too much. One wrongly credited a group that denies Christ's deity with reasoning it called biblical and coherent; another wrongly granted, of the pagan-myth parallels, that “the parallels are real.” Both conceded the conclusion and not merely the facts. They were rewritten to concede what a person sincerely believes and what a text actually says — never that a mistaken reading is sound.
• A scholar given the wrong credentials. A scholar we cite was described as a historian rather than as an apologist. Small, and exactly the sort of thing a critic is right to catch.
• A review that certified the wrong copy. The least flattering of these, and the most instructive. Each answer page exists as three copies that must agree — the visible page, the underlying data, and the structured data search engines read. Six pages had their visible text edited by hand without the other two being updated, so our review had certified a version that no longer matched what a reader saw: that text was, in effect, unreviewed. All six were reconciled and re-checked — the re-check found no doctrinal error, though we could not have known that in advance — and the generator now audits for this mismatch on every run, so the next build surfaces the mismatch instead of passing it.

None of this is comfortable to publish. It is here because a case that overstates itself will fail the first serious objection a reader meets in the wild, and confidence built on that is worth less than none. What remains has been through the checks that caught these.

Our standing commitments

• Cite real, verifiable sources. We link to primary texts and peer-reviewed scholarship where possible, and we never fabricate a quotation, source, or figure.
• Steelman the other side. We present objections in their strongest accurate form and name the scholars who hold them — charity means accuracy, never inflating the other side's case.
• Concede what should be conceded. We distinguish what an argument proves from what it merely suggests, and we flag genuinely contested claims as contested. We concede facts, not conclusions.
• Represent Scripture accurately. Verse references are checked to exist and to support the point they are cited for, in context.
• Gentleness and respect. Following 1 Peter 3:15, we make the case without sneering at those who disagree.
• Correct our mistakes. When we get something wrong, we fix it. If you spot an error — a misquotation, a bad citation, an unfair characterization — please tell us, and we will check it and correct it.

Found a problem?

We would genuinely rather hear it than not. Use the Asked & Answered page to flag any inaccuracy, and we will review it. Accuracy is a moving target, and we treat corrections as part of the work, not an embarrassment.