No description
- CSS 35%
- HTML 18.8%
- JavaScript 18.5%
- Nickel 14.8%
- Fluent 4.6%
- Other 8.3%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
claude-bot
e883db8a88
`precompress-check` defends a claim about BYTES — a stale `.br` wins over a fresh source and the reader is served the old page at 200 with no way to tell — and it measured mtime. That proxy lied in both directions, and both were reproduced on a fixture before anything was changed. FALSE RED, and it is what made this visible: `touch` on a source, zero bytes changed, turns the gate red and orders a `brotli -q 11` over a file nobody edited. Not hypothetical — the `pre-commit` framework stashes and RESTORES every unstaged modification on each commit (its own log says so), so every derived file still uncommitted in a multi-commit session gets a new mtime with identical content. Today's five commits reddened this gate against themselves, twice, over a tree that was serving exactly the right bytes. FALSE GREEN, which is the hazard the gate exists for: mtime does not survive a clone, and a `git checkout` writes a source OLDER than the variant beside it. Both read as fresh under `>=`. Measured — a fixture whose source content differs from its variant and whose mtime is older — the OLD gate was run against it and printed `✓ every variant is at least as new as its source` while the tree served the wrong bytes. It was blind to precisely its own declared hazard, and no amount of running it would ever have said so. So the criterion is the content: each variant is decompressed and compared to its source. This is the rule the constellation already applies and this script had not caught up with — `onre sync assets` stamps a digest, and scripts/decks/lib/freshness.mjs says it in the same words one commit ago: «mtime no sobrevive a un clone ni a un checkout […] quien decide es el digest». THE VERDICT IS TERNARY (adr-088). An unreadable variant or an absent decompressor reports UNKNOWN, apart from STALE, carrying the tool's own words: a red that says «the subject is broken» when the truth is «the instrument could not look» is unactionable, and a reader who cannot act on a red learns to discount every red. The WRITER moves to the same criterion, and it must: a writer deciding by the clock under a gate deciding by content is two rules over one fact, and the divergence surfaces as a gate that stays red after the fix it prescribed. It also refuses outright when brotli or gzip is absent, instead of quietly writing nothing and shipping raw bytes with no content-encoding. OBSERVED FAILING BEFORE BEING TRUSTED, on a fixture, per adr-072 `falsify-against-the-new- capability`: touch → green (was red); real content edit → STALE; source older than variant with different content → STALE (old gate: green); truncated .br → UNKNOWN naming brotli's error, not STALE; deleted .gz → MISSING; brotli off PATH → UNKNOWN attributed to the instrument, and the writer refuses. Cost measured, both ways: the gate is 2.0s over 550 sources; `just precompress` after touching three derived files now reports `0 variant(s) rewritten` in 2.0s, where the clock ordered `brotli -q 11` on all three. 24/24 site gates green. ondaod — engaged tension: enforcement-vs-emergence, whose eighth reading (2026-08-17, the same day) names this column: gates whose blindness reports as the subject's fault. The floor does not move here — the gate stays Hard, stays in `check`, stays refusing. What changes is that it stops lying about where it is. The rule itself is NOT new and no ADR is opened: adr-072 `a-check-measures-the-capability-not-a-proxy` (Soft) already carries it, and the instance is recorded as that ADR's first amendment, in the spine. |
||
| presentations | ||
| scripts | ||
| site | ||
| .gitignore | ||
| .pre-commit-config.yaml | ||
| LICENSE | ||
| README.md | ||
ontoref outreach
Public-facing material of the ontoref project: articles and posts, presentations,
case files, the ontoref.dev site and the jesusperez.pro web, the mdBook, and
the study-group material.
| Path | What it holds |
|---|---|
site/ |
SSR site — ontoref.dev deploy target |
web/ |
static web — jesusperez.pro deploy target |
presentation/ |
decks (markdown source → published PDF) |
book/ |
mdBook surface |
study-group-ontoref/ |
study-group sessions and hosted examples |
scripts/ |
build and publication tooling |
License
Two regimes live in this repository, and the boundary between them is declared, not inferred — see LICENSE for the exact wording:
- Content (articles, decks, case files, site and web copy, documentation) — CC BY-SA 4.0. Reuse with attribution; derivatives share alike.
- Software (
scripts/**, build recipes, thesite/andweb/implementations) — MIT OR Apache-2.0, the same terms as the ontoref implementation incode/. - Trade marks and brand identity (the names "ontoref", "on+re", "onre", the logotypes and wordmarks, the visual identity) — expressly reserved, covered by neither of the above. Unaltered nominative and editorial reference is permitted; modified marks and any use suggesting endorsement are not.