Platform Updates

Repositories: Partners, Partners API and Rego. Status: coordinated database, API and frontend release completed on 29 October 2026. The original scoped certification is retained below.

What’s Changed: Separate early bird cards configure registration tiers, add-ons and dealer packages. Organizers set dates in the event timezone and choose manual prices, flat discounts or percentage discounts. Checkout uses server quotes, preserves paid tier credit and purchased gift prices, and rejects changed reviewed totals. Database pricing checks account for expiry after a lock wait and preserve idempotent retries.

Impact: Organizers can offer timed discounts while retaining regular prices. Attendees review discounts, tax and fees before submitting; free checkout does not require payment evidence. Failed pricing reads withhold totals.

Original validation: Local checks passed for 30,000 exact-arithmetic pricing cases, 2,591 API assertions, 266 recorded database checks and 16 mounted checkout browser scenarios. Source hashes and the detailed tested boundaries are recorded in the finance certification report. The full native Windows database sweep and the existing Partners parity build gate remain outside the passing local suite.

Release follow-up: The authoritative pricing migration and coordinated API/frontend release are complete; current native PostgreSQL certification passes 266 checks, including independent lock waits and replay. Current builds and browser regressions pass. Hosted configuration, later manual upgrade approval and payment, ordinary checkout database atomicity, and provider settlement still need production acceptance checks. The earlier source receipts remain scoped to their original snapshot; current release evidence is recorded in the workspace verification entry.

Repository: Rego. Status: deployed through the linked GitHub build.

What’s Changed: Included package counts now say “N attendee rego included” across public package cards, application choices, checkout previews and summaries, payment details and transaction receipts. Related attendee place/slot labels and assistant capacity notices use registration and assistant terminology.

Impact: Dealers can distinguish included attendee registration from table space. Table-only packages explicitly say no attendee rego is included. Prices, package capacity and registration behavior are unchanged.

Upgrade Notes: No action is required. Typecheck, production build and source wording checks pass. Hosted shared-component previews passed at 1280px and 390px, including package selection for zero, one, two and three included registrations and package details, with no browser console errors or horizontal overflow. This copy release did not repeat payment submission or financial transactions.

Repositories: Partners API. Status: correction Git-deployed and the fresh hosted publication/removal cycle passed. Original CMS 503 attribution remains open because Sanity’s refreshed export window still excludes October 27.

What’s Changed: Scoped content reads request gzip instead of forcing uncompressed transfer. The proxy bounds decoded content to 8 MiB and keeps its existing deadlines and safe-read retry limit. It never replays content writes. Private failure classifications now distinguish waiting for headers from reading the response body without recording content, credentials or caller queries.

Validation: A real workerd/HTTP throttled large response fails before the correction after 14.2 seconds, then succeeds with the full decoded document in under one second. Oversized decoded content remains rejected. All 32 transport groups, 65 proxy cases, exact compiler and Worker dry build checks pass. Provider timing also shows lower transfer cost with compression, but does not establish the cause of the historical small-document failure.

Hosted follow-up: Normal Retry recovered the retained editor. Native Publish and reference-checked Unpublish each returned 200, both linked site builds succeeded, and ordinary warm public probes passed 404 → 200 → 404, including authored content/media, aliases, listings, feeds and sitemaps. Final exact reads show the test article removed and one assigned draft retained without a schedule. Normal Partners handoff also renewed CMS after reload. Report remained accessible during the real outage; its existing issue17 receipt was independently reconfirmed without private contact leakage. No duplicate Report was sent, and no staff notification destination is configured. These receipts do not close the original provider-log attribution gate.

Repositories: Public Furries PH site, Partners API and Docs. Status: Git-linked builds recovered after a repository-fetch interruption. Normal API and Docs pushes deployed successfully, followed by a successful fresh public-site build. Earlier CMS publishing/removal checks remain valid within their recorded scope; original CMS503 attribution and full certification remain open.

Impact: Check the linked build before treating a CMS publication as publicly available. The installed Cloudflare GitHub app already has repository access; the fetch failure is not yet attributed. No reauthorization, repository permissions, shared app installation or project source settings were changed.

Validation: The initial public-site, Docs and API documentation-only builds failed at source checkout. Subsequent API/Docs pushes and the fresh public-site retry succeeded. Ordinary public probes confirm a new static generation, the expected removed test article and listing/feed/sitemap exclusion, homepage assets and search metadata. This release-only follow-up performs no new CMS content mutation. The deployed Report response-body repair and confirmed technical Report17 remain valid. The Report validation notes are also verified on the deployed Docs page. Successful builds do not establish why either upstream interruption occurred.

Repositories: Partners API and CMS. Status: the response-body timeout repair is deployed; the original CMS incident attribution remains open until Sanity’s export window includes October 27.

What’s Changed: Report creation and installation-token requests keep their 15-second and 10-second deadlines active through the complete provider response body. Oversized acknowledgements are rejected. A lost response remains uncertain and never causes automatic issue creation to repeat.

Impact: Keep the same report reference and retry it after recovery. CMS-only login names are browser hints; when the reporting API cannot verify the session, the form requires a private contact email. That address is excluded from public issues, diagnostics and saved Report JSON.

Validation: The mounted CMS passed a real 90-second outage with the actual Report delivery coordinator, SQL/KV storage, forced restart and native 60-second alarm. Two intake attempts reused one reference and produced one synthetic issue. Eight actual workerd/real-HTTP checks cover successful, stalled and oversized response bodies. The wider runtime suite also passes three restarts, private contact retention, duplicate suppression, signed-comment mail and MIME replies through synthetic provider transports. Its retired mail fixture was updated to the canonical EMAIL binding and comms sender after the same old failure was reproduced on the immutable baseline. No live mail was sent by these tests. Fresh hosted acceptance passed: one normal CMS Report submission returned HTTP201 and created issue17. The public tracker independently confirms the expected labeled test and excludes the private contact address and upstream provider URLs. The earlier baseline receipt remains issue16. These are technical validation reports; no staff action is requested.

Incident evidence: Authenticated Sanity Manage confirms this Free project exports request logs from the last seven days excluding today. Current-day incident data is unavailable through that export and live log delivery is not configured. No export quota, plan or permission setting was changed. The historical cause remains unproven; successful recovery tests do not close it.

Repositories: CMS and Partners API. Status: repair released through the linked Git repositories; the fresh hosted publication/removal cycle passed. Broader certification remains open because the original stalled upstream read lacks a retained trace.

What’s Changed: An accepted publication can now retain its site-refresh work when the provider response is lost. Private persisted intents and autonomous alarms compare exact current revisions and request the static-site rebuild after recovery. Read and hook failures retry; content mutations are never replayed. Manual and scheduled publication use the same recovery record. Hook delivery is at least once, so a lost build receipt can request a safe repeat build.

Impact: Keep the tab open during a connection problem and reconcile an uncertain save or publish against its current document state. A native publish error can remain after a lost response even when later reads confirm publication. The rebuild recovery window is 24 hours; unresolved cases then require operator review. A later failed Pages build still requires release investigation.

Validation: Mounted Studio passed a real 90-second outage with its editing buffer/session retained, then normal and lost-response publishing with one publication command and one refresh request each. The Report draft survived JSON save, reopen and retry; actual widget/intake checks reused one reference/digest and produced one modeled issue across an uncertain then confirmed response. Twenty workerd alarm/persistence/eviction checks, 48 scheduler cases, 31 transport groups, 65 proxy cases, source boundary checks and exact compiler/build checks pass. These outage providers and delivery receipts are local models. The earlier real Report receipt remains issue16; no staff destination is configured. See the workflow record for the retained evidence and limits.

Hosted follow-up: Normal signed-in Publish and Unpublish each returned HTTP 200. Both persisted recovery alarms obtained build-hook receipts, and both linked static builds succeeded. The warmed public article passed 404 → 200 → 404, including aliases, listing, feeds and sitemaps, with the expected authored content and first-party image. No manual purge or visitor cache busting was used. The technical test article is removed; one unscheduled draft remains in CMS Preview.

Repositories: CMS and Partners API. Status: recovery repair deployed; the earlier blanket certification is narrowed to its observed test cases.

What’s Changed: Exhausted temporary document/query reads pause with a visible connection notice. Retry connection resumes those reads without signing out or replacing the editor’s document store. The notice reserves space and leaves Report a problem accessible. Read retries are bounded; uploads, mutations and explicit conflicts are not replayed by this recovery layer.

Impact: Keep the editing tab open during a connection problem and use Retry connection when the service is available. A connection notice does not confirm that an outstanding save or publication succeeded. Reconcile uncertain writes with current document state before repeating them. A held autosave can briefly show the last saved title after retry before its pending edit commits.

Validation: The actual mounted Studio survives two exhausted 503 batches, retains its draft/session/buffered edit, and commits one deliberate held edit exactly once. Report opens while paused at 280, 824 and 1440 pixels. Production checks resumed a real document-pair read with HTTP 200 and verified the saved preview, image and Publish readiness. An initial transport adjustment failed against real compressed HTTP responses; it was rolled back and corrected, and the misleading synthetic compression fixture was replaced with a real HTTP origin test. Current focused regressions, type checks and builds pass.

The original generic 503 did not record its failed dependency, so that historical cause remains unproven. These results validate the specified recovery cases; they do not guarantee provider availability. See the corrected workflow record.

Repositories: CMS, Partners API and furries.ph. Status: earlier normal-flow results below passed. The later read-outage follow-up above supersedes their broader certification wording; publication/report receipts alone did not certify exhausted-read recovery.

What’s Changed: Scoped editors can save content, preview current drafts, upload and crop images, inspect and restore revisions, retain supported list settings, schedule/cancel publication, and unpublish through genuine reference checks. The Report launcher remains clear of editor footer actions. Public pages now follow current published content after publishing and unpublishing: builds exclude retained drafts and stale query caches, and canonical delivery selects the current static output.

Impact: Reopen CMS through Partners and allow the accepted site build to finish after publication changes. A failed content build keeps the previous deployment available rather than silently publishing empty content. Report a problem confirms its actual tracker receipt. Staff alerts require a configured destination; none was configured for this validation.

Validation: The affected signed-in partner completed editing, media, history restore, preferences, real elapsed Cron publication, manual schedule supersession and current-draft preview. Two complete warm public cycles passed 200 → 404 → 200 → 404, with successful CMS actions and linked Pages builds, matching listing/feed/sitemap/search changes, and no manual purge or visitor cache-busting query. The test article is removed; its assigned draft remains unscheduled. One normal report created independently verified issue 16. Focused regressions, compilers and production builds pass. Guide/event and failure-mode coverage remains bounded to the fixtures described in the workflow validation record.

Repositories: CMS and Partners API. Status: deployed through the linked GitHub production branches. Full end-to-end certification remains open.

What’s Changed: Scoped editor preferences persist across reloads. Revision inspection and restore use a scoped, sanitized history transport and Sanity’s supported Timeline. Scheduling works from published-only content, cancellation checks current revisions, calendar dates filter the list, and due publication uses current content instead of an older search snapshot. Manual publication clears its previous schedule so later edits remain drafts.

Impact: Partner editors retain supported list/timezone settings and can inspect previous content. Posts and guides retain authored fields through scheduling/cancellation; changed or invalid drafts are held for correction. Fully deleted document history and unavailable provider revisions remain errors.

Upgrade Notes: Reopen CMS through Partners. API 0111154b and CMS 471004f9 passed their Git-linked production builds. Live CMS assets contain the Timeline and scheduling changes; the reporter matches tested source and anonymous preference/history requests remain blocked. Actual Studio history restore, post/guide scheduling, cancellation, due publication and a later unscheduled edit pass with synthetic providers. Focused proxy, SQLite, installed-client, schema and compiler checks also pass. See the workflow validation record. No database migration or account-permission change is required.

Earlier acceptance checkpoint, superseded by the signed-in validation above: The affected signed-in partner session is deferred at the user’s request. A hosted upload/edit/restore, elapsed scheduled publication, resulting public content and a matching real report issue receipt are still required. Local fixtures do not establish production certification.

Repositories: CMS and Partners API. Status: fixes deployed through the linked GitHub production branches. Full end-to-end certification remains open.

What’s Changed: Report a problem recovers failed verification-script loading through Retry connection while preserving entered details. Native CMS hooks wait for confirmed publication; scoped public writes request the Worker refresh after commit. Draft autosaves do not request builds. Cloudflare now runs pnpm build so the reporter is present before Studio initializes, with Rocket Loader excluded.

Impact: Editors avoid permanently stuck verification retries and premature rebuilds. Initialization failures remain reportable. Verification alone sends no report.

Upgrade Notes: Reload CMS through Partners. If verification remains blocked, save the report and retry when the provider is reachable. API changes 0b64d19d and CMS changes 9a7e8e6b / 8e07f373 passed Git-linked deployments. Live reporter bytes match the tested API source; deployed CMS HTML contains one initial reporter. Fifty proxy groups, twelve native React/browser cases, continuous-stream editor save/reload/reconnect, failed-Studio startup, desktop/mobile report recovery, TypeScript checks and production builds pass within their documented scope.

Earlier acceptance checkpoint, superseded by the signed-in validation above: The affected signed-in partner session was deferred. Actual publication/public-page content and a matching real GitHub issue receipt remain required. History is unavailable in the scoped transport; stored preferences, scheduling/timezone behavior and revision restore remain uncertified. Local provider fixtures and anonymous checks do not establish full production acceptance.