Repository navigation
The reader only accepts specVersion 0.3.0, so it cannot open half the registered feeds #11
Description
The bug
app.js decides what an OTE document is with an exact string comparison:
function isOteFeed(doc) {
return doc && doc.specVersion === '0.3.0' && Array.isArray(doc.events);
}
function isOteEvent(doc) {
return doc && doc.specVersion === '0.3.0' && doc.id && doc.name && doc.startDate;
}So the reader accepts only 0.3.0. The spec is at 0.6.0, and of the four registered adopter feeds, two already declare 0.4.0 — eventos.wiki and saraoia.com. The reader cannot open them today, and it will fail to open every feed published from now on.
This is not a 0.6 chore; it is a live failure against the real registry, and it is silent: the document loads fine, it is simply not recognised as an OTE document.
The fix
The version a document declares should decide which rules apply, never whether it is OTE. A feed is an OTE feed because it declares a specVersion and carries events[].
const SUPPORTED = /^0\.[1-6]\.\d+$/; // or: any 0.x, and say so when it is newer than knownBeing a reader rather than a validator, the tolerant reading is the right one: show the events, and if the version is newer than anything known, say so in the UI rather than refusing. A consumer that ignores fields it does not understand is explicitly correct in OTE — unknown fields have always been legal.
The inheritance list at line 352 is also short by two: contributing (new in 0.5) is inherited feed→event exactly like license, and the list should not need updating by hand each time — the schemas carry x-inheritsFrom on precisely the fields that inherit.
Worth having once it opens them
Nothing below changes whether a feed loads; they are what a reader can now show.
speakerswithaffiliation— "Ada Lovelace, Analytical Engines" is two fields, never one string.recordings— a past event with a video is the case a reader is most useful for.feed.covers— the window a feed claims completeness for. Directly relevant here: it tells the reader whether an event disappearing means "cancelled" or "aged out of the published window", which is the difference between dropping a saved event and keeping it.partOf.offers+offers[].scope— for a multi-part conference the admission lives on the parent, so a talk showing "free" because it has nooffersof its own would be wrong.
@opentechevents/schema@0.6.0 ships every version's schemas, so the reader can validate against whatever a document declares rather than hardcoding anything. ajvOptions is exported too — pass it whole; the bare strict: true the package used to document never compiled any OTE schema.
Activity
hhkaos commentedon Nov 4, 2026
Fixed in the working tree (0.10.0). Notes on what was found while doing it.
Confirmed against the live registry, 2026-11-04
specVersionDriven in a browser against the four real feeds: 25 events before, 701 after.
The one feed that did open was the one outside the support window, which is a
fair summary of the bug.
hydrateFeedCache()ran the same check on the IndexedDB cache, so a 0.4 feedwas also evicted from the offline cache on every start.
Decision: no dependency on
@opentechevents/validateReasoned before writing anything, since it conditions the rest.
The decisive argument is not size, it is semantics. The reader must be more
tolerant than a validator.
validateDocumentreports an out-of-window orunpublished version as an
error; gating the reader onreport.validwouldmake Community Builders' 0.3.0 feed unopenable — it works today — and would
break every feed again at 0.7.0. That is this issue's bug, reintroduced with a
better pedigree. A schema failure in one event would also invalidate a whole
feed, when the right behaviour is to show the other 484.
Size confirms it rather than decides it: ESM-only with per-version dynamic
import()aimed at a bundler, ~3.7 MB of compiled validators, against ano-build PWA whose entire
app.jsis 86 kB. Vendoring means megabytes in therepo and the service-worker precache; a CDN means an offline-first reader
depending on a third party to decide whether it can open a file.
So what is mirrored is the policy, not the schemas: window computed as the
last three minors, same vocabulary (
current/outdated/out-of-window),being behind is a note and never a defect. The cost of mirroring is drift, so
npm run checknow reads the npm registry and the published event schema andfails when the version list or the inheritance list disagrees with them. The
package stays the right tool where the job is judging a document.
One addition to its vocabulary:
future, for a version newer than this buildknows. A validator has nothing to check such a document against; a reader can
show it and say so, which is what this issue asks for.
Classification is by minor, not by exact string, so 0.4.1 is as supported
as 0.4.0 before this build has heard of it.
Two corrections to the issue
x-inheritsFromin thev0.6 event schema marks exactly five fields:
specVersion,organizers,textLanguage,license,contributing. Onlycontributing(new in 0.5)was missing. Same five in v0.5; v0.4 has four. The list still can't be read
at runtime without a build step, so it is checked at
npm run checktimeagainst the published schema instead.
covershas nothing to attach to yet. It matters where a reader dropsan event that disappeared from a feed — this one never drops anything
(
state.savedEventsis vestigial, nothing prunes). So there is no wronginference to correct today. Worth revisiting when pruning exists; building
it now would be building the consumer of a rule nobody applies.
Not done here, and why
speakers/affiliation,recordingsandpartOf.offers+offers[].scope(D032) are all rendering, and this app renders nothing — it sets
widget.eventson<ote-events>, already pinned at v0.6.0.inherit()copies whole event objects, so
partOfand the rest reach the widget intact;fixing the reader's own inheritance list is the only part of D032 that was
ever on this side. If an add-on price renders wrong on a 0.5 feed, that is an
embed issue.
UI
A feed on an older supported release says so when you open that feed, and
nothing on the aggregate screens — a legitimate choice shouldn't sit in the
way of every session. Out-of-window, newer-than-known and unpublished get
flagged in the sidebar too. Nothing refuses a feed, and no message asks the
publisher for anything.