> For the complete documentation index, see [llms.txt](https://hetcreep.gitbook.io/hetcreep-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://hetcreep.gitbook.io/hetcreep-docs/changelog.md). # Changelog Versions follow [Semantic Versioning](https://semver.org/) applied to a standard: * **MAJOR** — a locked rule changes meaning, or a published tolerance moves. Conforming work may stop conforming. * **MINOR** — a rule or register entry is added; existing conforming work still conforms. * **PATCH** — wording, citations, corrections that do not move a value. ## 2.0.2 — 2026-08-11 **The address this document told everyone to cite could not resolve a single rule.** 2.0.1 published at a documentation site and said "rule ids are stable anchors: `#L1`, `#E3`, `#A2`". They are not, there. That renderer strips explicit anchors and generates its own slugs, so every `#L1` in a citation landed at the top of the page — silently, which is the worst way for a link to fail. Measured on the served HTML of both surfaces, not inferred: ``` documentation site, rendered 0 of 13 rule anchors present the file at a signed tag 13 of 13 present ``` An earlier check appeared to show the anchors surviving. It was wrong, because the tool used converts a page to markdown before anyone looks at it — so it was reading the source both times rather than the render. That is recorded because the mistake is the instructive part: a measurement is only as good as the thing it actually touched. **Changed** * Citations now point at `SPRITE-DESIGN-DATUM.md` **at a signed tag**, which is where the anchors resolve. `CITATION.cff`, the front matter and `README.md` all say the same thing. * The documentation site is described as what it is: a reading copy, convenient and **not** citable. Not hidden, not apologised for — a reader is told which surface answers which need. * The repository is **public**. That is what makes a tagged file a citable address at all, and it closes the oldest gap here: a document that has instructed readers to "LINK to it and cite it" since 1.0.0 finally has somewhere to point that works. * `.gitbook.yaml` and `SUMMARY.md` bound what the site publishes. It had been publishing every tracked file — including the caretaker's own seat definition and an internal audit report — as public pages nobody wrote for readers. * The `check.yml` header is repaired. It had been wrong twice in opposite directions, first claiming no remote existed and then left half-written during that repair, and now records both. The standard is now published at ****, which is the first time the instruction it has carried since 1.0.0 — *"LINK to it and cite it"* — has had anything to point at. The front matter and `CITATION.cff` now carry that address, and the front matter states what a citation should look like: the id, the version, never the heading text. PATCH, and the version moves at all only because `GOVERNANCE.md` says a released version is immutable. 2.0.0 was tagged and pushed before these edits; changing its text and leaving the number alone would have made the tag and the document disagree, which is precisely the failure the immutability rule exists to prevent. **Corrected — four places where a document described a world that had changed** * `ERRATA.md`'s summary table still marked `E-001` and `E-002` as **reported** while their own entries said **corrected in 1.1.0**, and `E-003`'s state cell was empty. The table is what a reader scans first, so it was the copy most worth keeping right and the one that drifted. * `.github/workflows/check.yml` opened by declaring the repository had no remote and the workflow had never run. Both stopped being true the moment it was pushed. * `CITATION.cff` warned that nothing enforced the three copies of the version string. `check.mjs` check 4 has enforced it since the day it was written. * `REVALIDATION.md` had an empty Sweeps section and said nothing about being empty. It now says so outright: no claim has been re-checked since publication, and an empty register states that better than a green tick would. All four were found by the caretaker running offline, and all four are the same species — a sentence that was true when written and quietly stopped being true. That is the thing this repository exists to catch, so finding four of them inside its own files on day one is the expected number, not an embarrassing one. **`E-003`.** Layer B and the unbounded register were sorted on two different axes and the document never said so, which made three quantities appear in both lists and read as a contradiction. For two of them it was not one. An opinion board of four independent seats — one deliberately blind to the framing, the options, and the proposal on the table — was asked where the three belonged. All four rejected the question. The discriminator in use, *did you measure it*, is not what separates these cases. **Changed — the Layer B entry condition** Was: a value enters Layer B only with a corpus measurement behind it. Now: a value enters Layer B only if a **wrong** value would be a demonstrable error against something that exists independently of the declaration. A canvas declared at 63 px when the file ships 64 px is wrong, and anyone can open the file and show it. Eight frames rather than twelve is not a wrong count — it is a different animation. The old test admitted the second, because you can always count files; the new one does not, because counting them proves nothing about which count was right. The document also now states the axis outright: **Layer B asks whether the adopting project can measure a value; the unbounded register asks whether anyone outside bounds it. A value may sit on both**, and a frame count does. **Changed — the slot list** The direction-to-index mapping is no longer a Layer B slot. Nothing in the art can be measured to produce "0 means south"; it is a label. It carries a warning the board surfaced that nobody had raised: it is **free of external constraint and not free of cost**, because every filename already delivered encodes it. Changing it is a breaking migration. Each remaining slot now states its answer on both axes instead of sitting in a bare list. **Changed — the promotion clause** Promotion out of Layer C now takes the same evidence the entry condition asks for, so the two cannot drift apart. And it says plainly that **promotion is not exclusion** — a value that earns Layer B still belongs in the unbounded register if nobody outside bounds it. Without that sentence, the old wording quietly recreated the same mis-sorting from the other direction. **Added** * A register row for the **number of directions in a set**. It had appeared exactly once in the whole document, in the slot list, and any repair that only moved slots would have lost it entirely. **Why MAJOR** The stated test is that a locked rule changes meaning **or** a published tolerance moves. The first clause fires: the Layer B entry condition admits and excludes different values than it did. The second fires too — a project that filed the direction-to-index mapping at Layer B, which the slot list told it to do, stops conforming. An earlier draft of this entry called it MINOR by arguing only the second clause and skipping the first. The rule uses *or*, not *and*. Calling it MINOR because the document is one day old and nobody outside has adopted it yet would have been choosing the number by convenience rather than by the test — which is the habit this document exists to make harder. ## 1.1.0 — 2026-08-11 Two errata verified against their primary sources and corrected, plus the taxonomy defect that let one of them hide. `ERRATA.md` keeps both entries permanently — 1.0.0's text still says the wrong thing, and anyone holding that version needs to be able to find out why. **Corrected** * **E-001** — the Unreal texture-ceiling row said 8192 was available "without an engine configuration change". Epic's own documentation says the opposite: it needs `MaxLODSize` set in `BaseDeviceProfiles.ini`, and without that an imported 8192 texture renders at 4096 **silently**. The row now states the condition and names the silence. * **E-002** — the half-texel row attributed its explanation to Microsoft Learn's *Bilinear Texture Filtering*, a Direct3D 9 conceptual page that does not contain it. The statement was correct; the citation was not. Now sourced to *Coordinate Systems* (Direct3D 10), which states `Left Texel # = floor(U − 0.5)` verbatim, and to the Vulkan specification. This was the second occurrence of a failure this document already records committing once, which is why both rows now carry their citation class. **Changed — the taxonomy that let E-001 hide** * The register section headed *"Hard — a gatekeeper or an API rejects violations"* is now *"Hard — the violating result cannot exist"*, and states which of two mechanisms each row relies on: a gatekeeper refusing the upload, or the engine silently overriding you. The old wording tested for **rejection**, so a row whose engine quietly clamps you to half the size read as correctly filed. It was not, and that wording is what let it sit there unnoticed. * The pixel-exact rendering row moved from *Hard* to *Recommendation*. Nothing rejects, clamps, or detects a mismatched Pixels Per Unit — it renders, and it looks wrong. That is the Recommendation tier's definition, not Hard's. **Why MINOR and not MAJOR** The stated test for MAJOR is that a published tolerance moves and conforming work may stop conforming. Neither happened. 8192 is still 8192; what changed is the condition attached to it and the honesty of its strength class. Moving a row out of *Hard* makes it **less** binding, so a project that met the stricter reading still conforms. A correction that only loosens, or that adds information about a third party's behaviour, does not break anyone downstream. ## 1.0.0 — 2026-08-11 First standalone publication. The document existed before this as an internal design lock inside one game project; this release is that document with every project-specific reference removed, so it stands on its own. **Added** * `README.md`, `AGENTS.md`, `CHECKLIST.md` — one entry point each for a human reader, an AI agent, and the person actually delivering art. * The `datum` framing: the term is taken from dimensioning and tolerancing (ASME Y14.5, ISO 1101), where a datum is the exact reference every other feature is located from. A 2D character's is its foot line. **Changed** * `E2` and `E3` restored to reading order. `E3` had been appended ahead of `E2` in the source document. * Layer B is now described as the *adopting project's* measurement rather than one specific project's. * The conformance section now describes how any project builds its own record, instead of pointing at one particular file. **Licence** * Standalone CC BY-NC-ND 4.0 rather than a carve-out from a surrounding MIT repository. The standing grant to Katomnoi Studio carries over unchanged, still without the right to sublicense or redistribute. * Two limits added to the scope section that the earlier draft did not state: facts and measurements are not copyrightable, so independent re-measurement is expected and welcome; and short quoted passages from vendor documentation remain their owners' property, cited under fair dealing. **Carried over unchanged from the source document** Every number, every citation, every layer assignment, and every annotation of a point where an earlier edition was wrong. The provenance is in `README.md`: 96 claims through adversarial verification (44 held, 50 corrected, 2 unverifiable), then a second sweep finding 172 published values, 72 quantities with no published source, and 12 overclaims caught. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://hetcreep.gitbook.io/hetcreep-docs/changelog.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.