Editorial policy

Editorial policy and AI disclosure

This is the standard everything on LedgerTouch Intelligence has to clear before it is published, and an honest account of how it gets written. It is written down because a standard nobody can read is not a standard, and it is enforced by code because a standard nobody enforces is a preference.

Editorial standard version 2026-07-13 · This page last updated 2 August 2026

Who is responsible

Tayyub Yaqoob owns LedgerTouch and is the responsible editor for everything the desk publishes. Publication is his decision and the accountability is his. Editorial mail reaches him at editorial@ledgertouch.com.

The rule everything else hangs off: nothing goes out unverified

Every numeric claim in an article must trace to a source that was actually fetched during research. The writing stage receives a set of retrieved facts, each carrying the URL it came from, and may use only those figures. A number that appears in a draft but not in that retrieved set fails a check and the draft does not publish. Figures are never supplied from anyone's memory, human or model.

Sourcing rules

• Primary sources. The filing, the statistical release, the raw dataset, the paper, the statute — not a news story about it, and not a company's investor homepage.
• At least six distinct references per article, each a working deep link. A link whose path is a homepage or landing page is rejected automatically, because a homepage supports nothing.
• The link must return successfully and must contain the claim. It is fetched and checked. A live URL that does not support the claim attached to it is a failure, not a pass.
• No search-engine redirect URLs as citations. Grounding and redirect wrappers are research plumbing, not sources, and are rejected by name.
• Rolling endpoints must be labelled. A URL that always serves "the latest issue" will quietly serve different numbers next month. Those links are allowed only if the source label discloses the as-of date.
• No data may postdate publication, and every figure carries an as-of date. A figure more than one reporting period old for its series is rejected.
• Attribution must be exact. If an institution is named, that institution must have published the figure.
• No hotlinked third-party images. Images are self-hosted, and alt text must describe the actual image rather than repeat a template.

How an article is made

1. Research. Sources are located and fetched. Figures are extracted together with the document, page or table they came from.
2. Drafting. The draft is written with AI assistance, working only from that retrieved fact set. See the disclosure below.
3. Verification. Every figure is checked back against the primary source. For the guide series this is written down: each guide has a verification report that lists every number in the piece against the document, page or table it came from, and records which figures were dropped because they could not be verified.
4. Automated gates. The draft is run through the gate suite described below. Any failure blocks it.
5. Publication. Only a piece that has passed is published, under the responsibility of the named editor.

AI assistance: what we use it for, and what it is not allowed to do

The automated gates

Before a draft can be imported for publication it is run through a suite of more than thirty machine checks. A draft that fails any one of them is reported and skipped — it is not imported with a warning, and it is not published with a note. Among the things checked:

• every source resolves, is a deep link, and is not a redirect artifact;
• at least six distinct primary references;
• length is inside its band — 700–1,100 words for a market note, 1,800–3,000 for a guide;
• three to five key takeaways at the top, each carrying a specific number;
• the body is structured HTML with descriptive headings, balanced tags, and no unrendered template markers;
• title at most 60 characters; meta description 140–160 characters and a complete sentence;
• the slug is short, keyword-bearing, and not dated;
• a list of banned phrases and template stems, and a separate list of advice constructions, either of which rejects the draft;
• chart values trace to retrieved data, with one unit per axis;
• images are self-hosted and the alt text is not boilerplate;
• at least two internal links, each pointing at a canonical article URL;
• for guides: the strongest counter-argument is engaged, the piece states what would change its conclusion, and the product is mentioned at most once;
• for guides, measured voice thresholds — average sentence length at most 23 words, no more than 20% of sentences over 32 words — because a fact-checked draft drifts toward unreadable prose unless something measures it.

These gates exist because of specific failures, not as a wish list. The standard was written after an internal audit of the then-published archive, and each rule in it traces to something that had actually gone wrong.

The line between information and advice

These topics sound like advice questions. We answer them as evidence questions. That is a deliberate line, and it is tested rather than trusted.

• Describe, do not instruct. Not "rebalance annually", but "annual rebalancing captured most of the benefit in Vanguard's data, at a fraction of the trading cost". The test we apply: could a reader reach a different conclusion from ours and still be reading us correctly? If not, we have instructed rather than informed.
• No guaranteed outcomes. "Has historically" is not "will". Constructions that promise a market outcome are rejected.
• Blocked constructions. "You should", "we recommend", "investors must", "you need to", "the right move is" and similar phrases reject a draft outright.
• A disclaimer does not cure an instruction. Removing the imperative does. The disclaimer is there as well, but it is not the mechanism.

LedgerTouch is not a financial adviser, is not regulated as one, and nothing published here is a personal recommendation.

Nuance, limits and counter-evidence

• Every piece must state where its framework breaks, what would falsify it, and what the data cannot tell you — in the article itself, not in a footer disclaimer.
• Every guide must give the strongest objection its best case rather than a strawman. A piece that does not engage the serious counter-argument is marketing.
• No fabricated precision. If a study says "roughly 1–2% a year", we write that. We do not write 1.7%.
• Before publication a thesis is checked against the live archive, so that two articles do not tell readers opposite things about the same release.

Independence

Revenue comes from subscriptions to the LedgerTouch product. There is no advertising, there are no affiliate links, and there are no sponsored, paid or gifted articles. No third party pays for coverage or sees an article before publication. An article may mention the product at most once, and an automated check enforces that; the product is never permitted to be the conclusion of a piece.

Corrections

When we get something wrong we correct it in the open. How to report an error, what happens next, and where corrections are recorded are all set out on the corrections page.

What this policy does not claim

It does not claim we are never wrong. A process that catches most errors still lets some through, and a figure that was correct on the day of publication can be revised by the body that issued it. It also does not claim every article on the site is equally current: market notes are tied to a moment and age accordingly, which is why each carries its publication date. Where a piece has been overtaken by events, the honest response is a correction or an update, and both are covered by the corrections policy.