One definition of a category, the model and the test that apply it, and the questions it leaves open. The claims live in datasets, they are reviewed on a stated cadence, and they are corrected in public.
Autonomous Resource Planning is defined here in one sentence and taken apart clause by clause. Around that definition sit a loop of six stages; a maturity model of six levels graded per resource class; a test of eight questions, three of them load bearing; eight architectural principles stated as requirements; eight chains traced from a trigger to the entry they leave behind; an account of governance; twelve objections with two of them marked unresolved; a history of the eras this one follows; an account of how a transition of this size arrives; and a glossary of 59 terms.
The purpose is a vocabulary precise enough to argue in. A category with no agreed definition is a category in which every system qualifies, every claim is unfalsifiable and the word means whatever the last person to use it needed it to mean. So all of it is written to be failed: the definition excludes things, the test can be failed question by question, the maturity model places most systems below the line it draws, and the questions that have no answer here are published with the fact that they have none.
It is not a product, a buyer's guide, a ranking or a shortlist. No system is assessed here and none is recommended. Two pages name systems, the history and the transition, and a name appears on them as a plain fact about what shipped and when, carrying no adjective. Everywhere else the subject is architecture, and naming a product there would turn a question about how something is built into a question about who built it.
No publisher is named, and that is a decision, not an omission. Nor is any claim made about whose interests this serves, because such a claim would be a statement about people the page does not name and could not be checked by a reader either way. What can be checked is the material: whether the definition holds consistent everywhere it is used, whether the test can be failed, whether the open questions are genuinely open, and whether a correction sent here changes what is published. Those are the terms it asks to be read on.
The load bearing content lives in datasets, not in pages. The levels, stages, resource classes, principles, test questions, eras, scenarios, objections and glossary terms are each a record with named fields, and a page renders those records instead of restating them. One consequence is that a definition cannot say one thing on the page that defines it and another on the page that uses it, because there is only one copy. Another is that a change to a claim is a change to a record, visible as such.
A set of tests runs against every published version and enforces what has been promised: that every record carries the fields its type requires, that every id rendered by a dataset still has its anchor on the page that owns it, that the sitemap lists exactly the pages that exist, and that no page can quietly drop an entry the rest of it refers to. These are ordinary software checks doing an editorial job. They exist because an error of this kind is invisible to a reader and obvious to a machine.
Two rules govern what may be published. A figure taken from outside appears only with the source it came from and the date that source was read, and a figure without both is not published at all. A question that cannot be settled is marked unsettled where it is asked and never answered thinly, which is why two of the twelve entries on the objections page are marked open. Neither rule is a courtesy. Work of this kind is worth reading in proportion to what it declines to assert.
Everything is read end to end at least twice a year, and the review date shown on every page is the month of the version you are reading. A review is not a copy edit. It asks of each page whether the claim is still the claim, whether anything since has settled a question marked open, and whether a page has drifted from the definition while nobody was looking at both at once.
Corrections are accepted and wanted, including ones that unsettle a claim made loudly here. Send them to corrections@autonomousresourceplanning.com. A correction is most useful when it carries four things.
A correction that holds changes the page and the date on it. A correction that turns out to be a disagreement about the argument, and not an error in it, has a home too: the objections page exists for exactly those, and the strongest version of an objection is worth more than a weak version somebody here can answer. What will not happen is a claim being quietly softened. If something here is wrong, the page says what it said and what it says now.
If a citation has broken, that is also a correction. Send the URL you were pointed at and where you found it, and see how to cite for what the anchors promise.