Release history

Corrects. v0.1.7's release note and board card 010 both said 70 images were missing and that a re-run would recover them. The honest figures are 89 recovered, 16 never archived and 51 that were never this site's to recover. The wrong text stays on the card with the correction above it. Separately, two renderer defects found by this site's own prose: a code span containing markdown had that markdown rewritten — a board card explaining that image syntax had never been handled turned its own example into a link and broke the build — and the check that catches literal markdown then flagged the correct rendering of a code span as a defect. A renderer that cannot document its own syntax is one that will quietly rewrite any example anybody ever writes in it.

Corrects. A decision from v0.1.3. The archive pages said images were 'left exactly as written rather than silently repaired, because a rewritten archive is no longer evidence'. The principle is right and the application was wrong: it produced an archive nobody could read, which is not evidence either. Images are now served from this repository with the original URL kept in each image's title — rewritten AND recorded. Separately, pointing an <img src> at the local copies broke the authoring contract, because a declarative reference to a vault path 404s inside a sandboxed frame before the bridge installs; the validator caught it before it shipped. Images are emitted as links that JavaScript upgrades, which passes the contract, works in a vault, works on the static mirror, and degrades to a working link with scripting disabled.

Corrects. Two things this release found by having more than one article. With one article the audience ontology looked serviceable; with three, the founder receives everything and seven concerns have never fired once. Neither is necessarily a defect — a concern about funding programmes cannot fire on a newsroom that has not published one — and that ambiguity is exactly why they are published as questions for a human rather than auto-corrected. Separately, review.py's first diagnosis named the wrong culprit: a max-of-counts heuristic pointed at the concern that fires most, which is not the one that fired on the article that should not have been delivered. It now reports the whole distribution and lets a reader find it.

Corrects. Nothing from an earlier release, but the first run of join/v1 corrected three things in the audience ontology written hours earlier, and the page shows the before and after rather than the fix alone: a concern matching on Source fired for every article that freezes its sources and put the security practitioner at the top of a story about stage names at a startup event; one entity satisfying two overlapping concerns scored twice and put the executive above the founder on an event story; and a concern called 'something more likely or more expensive' fired on any event at all. Concerns can now be qualified, rank counts each entity once, and the practitioner is correctly withheld.

Corrects. v0.1.3 was announced as shipped. It was committed and pushed; it never deployed, and the site served v0.1.2 throughout. The failure was in the repository's own hygiene rather than in anything the release changed, which is the least interesting kind of outage and the easiest to repeat: the assertion added here exists so the next one fails at the desk instead of in CI.

Corrects. The whole argument of v0.1.0 through v0.1.2. It was a plausible thesis about feed readers in general and not a description of this project, and it was asserted on a site named after the project. The pages stay up, marked, because deleting them would remove the only evidence that this site corrects itself rather than quietly rewriting. Separately: v0.1.2's validator reported 19 pages passing while the build produced 38 — it maintained its own idea of what existed. It now reads the build's published inventory.

Corrects. v0.1.1's release note ends 'this release makes the site deployable; it does not make it deployed', and v0.1.0's says nothing is served yet. The first was wrong within ten minutes of being written — the deploy job it added is precisely what made the site deployed. Both entries stay as written; this is the correction above them.

Corrects. v0.1.0 described a four-step release ending in a live check, and shipped neither a deploy workflow nor anything to check with. The description was accurate about the intent and wrong about the repository.