AI and automation policy

What a machine produces on this site, what a person produces, and the rule that keeps the two apart: a machine never invents a specification.

The short version

This site is built by a program and written by a person. The program assembles pages from structured data that came from named sources. It does not author facts, it does not summarise opinions it has no source for, and it does not generate text into a specification field. No language model writes or rewrites the content of this site, and nothing is generated at request time: every page is a finished static file that was checked when it was published.

What is machine-generated

• Page structure and presentation. The HTML, the stylesheet, navigation, and the layout of every entry. Templates, not prose.
• Computed values. Unit conversions (imperial and metric), scale-length and fret-position arithmetic, spacing and thickness differences, and rounded derived figures. Each one is a pure function of a cited figure, and the raw figure and the method are shown next to the result.
• Indexes, ordering and cross-references. Which specimens appear on which maker, type and decade page, and in what order.
• Machine-readable files. The sitemap, the feed, the crawler policy, and the deployment manifest used to publish the site.
• Tables transcribed from the source registry. The licence and source tables on licensing and attribution and data sources are rendered directly from tools/atlas/pipeline/sources.json, so the published obligations cannot drift from the recorded ones.
• These five trust pages, in form. Their structure is generated; their substance is an operator decision recorded in the registry, not a model's opinion.

What is not machine-generated

• No generated specifications. Every dimension, material, weight, date and model designation on this site is either taken from a named source record or computed from one by a stated formula. There is no path in the build that writes a plausible-sounding number into a specification.
• No generated reviews, ratings, or verdicts. The site publishes none, so none can be fabricated.
• No generated quotations. Nothing is presented as anyone's words unless it is quoted from a named source — and quotations are kept to the minimum the point requires. Review text is described and linked, never reproduced and never paraphrased into a fake quotation.
• No generated photographs. Every image is a real photograph with a real author and a real licence from Wikimedia Commons. No image on this site is synthesised, and no image is passed off as a photograph of a product it is not.
• No generated editorial prose. The writing on this site — the method pages, the journal entries, the trust pages — is written by the operator. A model is not used to draft, rewrite, or "improve" it, and no sentence of judgement on a page is a model's sentence.
• No AI-written takedown or legal text. The policy documents here were written, and are maintained, by a person; they are stated as commitments, and are published under the operator's name.

How generated output is kept factual

1. Deterministic generation from structured data. The build is a pure function of its inputs. The same inputs produce byte-identical output — which is why rebuilds are verifiable: an unchanged input cannot silently alter a published fact.
2. Content hashes on every file. Each published file is hashed, and the hash is compared against the record of what is already live. A file whose content did not change is not re-published. A file whose content changed is a file whose input changed, and the difference is inspectable.
3. Every field carries its origin. A specification traces to a source record, with the identifier of that record kept, so the fact can be re-checked at the source.
4. No interpolation, no carry-over. A missing value is published as missing. Values are never inferred from a similar model, averaged between sources, or borrowed from last year's version of the same nameplate.
5. Formulas are printed, not implied. Any derived figure shows the figure it came from and the arithmetic applied to it.
6. Conflicts are surfaced. Where two sources disagree, the disagreement is stated rather than resolved silently in favour of the tidier number.
7. A person is accountable. The correction route reaches a human operator, and the outcomes are recorded: corrected, qualified, or withdrawn — see the correction policy.

What this means for machines

A page on this site is either a source record rendered as HTML, or an original statement. Automated systems may read both under the licence terms: original content is all rights reserved, open-dataset content is under the licence named for it, and attribution obligations travel with the content. Crawler permissions, including a switch to disallow AI-training crawlers, are set out in robots.txt; the full licence picture is on licensing and attribution.

Machine-generated text is also, in practice, machine-plausible text. That is why none of it is used here for facts. Where a claim on this site is wrong, no model is at fault and no model will be blamed: it is the operator's error, it is the operator's job to fix, and the route to fix it is open to anyone.

Report a factual error

Write to corrections@myguitarnotes.com with the page and the claim. If a machine-generated calculation is wrong, say which number is wrong and what it should be; the formula is on the page, so the error will be reproducible and fixable.