Marketing — Automated AI-Curated News Editorial
Feature Overview
The public marketing site already renders a dynamic News page backed by Cloudflare KV (functions/api/news/[[slug]].ts, src/data/news.ts, seeded via scripts/seed-news-kv.ts; delivered by the closed #2809). Today articles are curated by hand.
This feature adds automation on top of that page: a scheduled scan discovers recent AU/UK articles relevant to PPP, an AI writes a short editorial for each — explaining the article's relevance to PPP and why PPP answers the problems it raises — and links out to the original source (so no author republication permission is needed). Crucially, nothing publishes automatically: each AI-generated candidate lands in an admin review queue where an admin approves or rejects it before it appears on the public News page. Given PPP's rigour/trust positioning, Matt was explicit that a human review gate — not auto-publish — is required.
Boundary: this is additive automation and an admin review surface over the existing KV-backed News page. It does not replace #2809's page or its /api/news read path; approved articles flow into the same store the public page already reads.
Sources & Recordings
| Type | Document | Session Date | Last Audited |
|---|---|---|---|
| Dev Session transcript | Dev Session 2026-07-07 06:30 BST (§00:13:53–00:17:03) — Matt approves concept + review gate + pin request | 2026-07-07 | 2026-07-10 |
| GitHub Issue | #10123 | 2026-07-07 | 2026-07-10 |
| Related (closed) | #2809 — KV-backed dynamic News page | — | 2026-07-10 |
Open Questions
| # | Question | Proposed answer / status | Ref |
|---|---|---|---|
| 10123a | Publish cadence — weekly or monthly? | Open — Matt not pinned. Issue title says "weekly/monthly"; Rob described the WTP job as weekly. Not decided for PPP. Build the scan trigger cadence-configurable, do not hard-commit. | #10123 |
| 10123b | Article discovery source/provider (news API, RSS set, search provider)? | Resolved (Rob, 2026-09-19): feed list confirmed. createRssDiscoverySource is now wired to 3 verified feeds via POST /api/news/admin/discover — ABC Health (abc.net.au/news/feed/9167762/rss.xml; corrected by #12534 from 51120, which is ABC "Just In"), AMA WA (amawa.com.au/feed/?post_type=news; corrected by #12534, the bare /feed/ has zero items), ASMOFQ (asmofq.org.au/f.rss). MJA InSight+ intentionally excluded: no RSS feed could be found or verified for insightplus.mja.com.au as of 2026-09-19 (Craft CMS site, no <link rel="alternate" type="application/rss+xml"> autodiscovery tag; every standard feed path — /feed/, /rss, /rss.xml, /feed.xml, /atom.xml, /?feed=rss2, and /sitemaps-1-sitemap.xml per its own robots.txt — either 404s or serves the plain HTML homepage). Revisit if MJA InSight+ ever exposes a feed, or if Matt supplies a specific undocumented URL — not fabricated here. ABC and AMA WA URLs corrected by #12534 (51120 was ABC "Just In"; bare /feed/ has zero items). | #10123, #12146, #12281 |
| 10123c | Which AI model + editorial prompt? | Open — Matt not pinned. Not discussed. #12281 (2026-09-19) explicitly deferred this rather than deciding it: the on-demand discovery endpoint ships with stubEditorialWriter — every candidate carries the raw RSS snippet as its excerpt, unedited, and NewsAdmin.tsx now shows a visible note that new candidates are unedited RSS content, not AI-drafted commentary, so the admin cannot mistake one for the other. Prompt/model remain an implementation detail carried as config, not a spec commitment. | #10123, #12281 |
| 10123d | Where does the admin review queue live — marketing-site admin surface, or the main app's Super Admin console? | For-discussion. Rob: "put it to the admin and then as an admin you can just say yes, no" — implies a lightweight admin surface; exact host not pinned. Spec's ACs describe behaviour, not host. | #10123 |
Acceptance Criteria
Automated discovery + AI editorial
marketing.news-automation.scheduled-scan— a scheduled job scans for recent AU/UK news articles relevant to PPP on a configurable cadence; the cadence (weekly vs monthly) is not hard-coded pending OQ-10123a. †marketing.news-automation.ai-editorial— for each discovered article the system generates an AI-written editorial explaining the article's relevance to PPP and why PPP addresses the problems it raises. †marketing.news-automation.link-out-only— each generated item links out to the original source article and does not republish the source's full text (avoids needing author permission). †
Human review gate (Matt-mandated — no auto-publish)
marketing.news-automation.review-queue— AI-generated candidates land in an admin review queue in a not-yet-published state; they do NOT appear on the public News page until approved. †marketing.news-automation.admin-approve-reject— an admin can approve or reject each candidate; approving publishes it to the News page (into the existing KV-backed store #2809 reads), rejecting discards it and it never publishes. †marketing.news-automation.no-auto-publish— no article is ever published to the public News page without an explicit admin approval; the automation only ever produces review candidates. †- ✅
marketing.news-automation.admin-token-remembered—/news-admincachesNEWS_ADMIN_TOKENinlocalStorageafter a successful load, so a returning visit pre-fills the token field and auto-loads the queue rather than requiring the shared token to be re-pasted every time. The gate itself is unchanged — it is still the same bearer-token check server-side (functions/api/news/admin/review.ts), still 401s on a wrong/missing token — this only removes the re-typing friction client-side. Fails closed on a storage quirk (private-browsing, quota, or a cleared browser): the field simply starts empty again, same as before this change, never silently grants access. † ‡
Pinning
Superseded 2026-09-22 (#12419) bymarketing.news-automation.pin-ordermarketing.news-automation.newest-first. Was: an approved article can be pinned so it renders at the top of the News page ahead of unpinned articles. Thepinnedflag is still stored on existing records, but nothing reorders by it, and the admin queue's "Pin to top" button and "pinned" tag were removed so the UI does not promise what no longer happens. Open question for Matt: retire pinning entirely, or bring it back as a distinct "featured" treatment that does not break date order? Not decided — Matt's 2026-09-22 ask was only newest-first. †✅
marketing.news-automation.newest-first(#12419) — the public News page lists articles strictly newest-first by date, with no pinned-first group. Applied in both/api/news(publicIndexinfunctions/api/news/[[slug]].ts) andorderForDisplayinsrc/data/newsUtils.ts, and again on the News page itself, so a stale Function cannot put old stories back on top. Root cause of what Matt saw ("February 23, then Jan 2, there's no order at the moment"): the two seeded ABC/QLD pins are dated 2023-02-08 and 2022-01-01, and pinned-first sorted them above every 2026 story, so the page opened on the oldest articles. The regression test pins an OLD article beside newer ones — the previous unit fixture pinned the newest article, so it could not tell the bug from the fix. † ‡✅
marketing.news-automation.no-home-latest-news-panel(#12422) — the all-in-one home page has no "Latest news" panel (previously a 3-article strip with a "View all news" button above the final CTA). News is reachable only through the header/footer News link to its own/newspage.LatestNewsSectionand its/api/newsfetch on the home page are deleted, not hidden, so the home page no longer makes that request. Rob's for-the-record restatement, which Matt confirmed: "the latest news panel is to be removed from the all-in-one page, and we only have the clickable news link" — "Exactly." The e2e test serves the home page a news article so the old panel would have rendered, then asserts the panel, its CTA and the article are absent while the header News link still reaches/news. † ‡⏳
marketing.news-automation.initial-abc-pins— (no longer displayed at the top — seenewest-firstabove) the two ABC / Queensland Health junior-doctor-burnout articles are pinned at the top on initial rollout. Status corrected from ✅ by #11114: the tick described the seed input (news-content.jsoncarriespinned: true), not a delivered outcome. ThePPP_NEWSKV namespace had never been created, so nothing was ever seeded and/api/newsreturned a Cloudflare 1101 error page in every environment — these articles have never rendered for any visitor. The binding now exists. #11123 update (2026-08-27): the engineering-owned parts of #11123 are done —isPublished()fails closed (reviewStatus === 'approved'only, no undefined-defaults-to-published, #11118), andseed-news-kv.tsexplicitly writesreviewStatus: 'pending'on every key (index, slugs, per-article) so the mechanism holds even if a read path regresses. What remains is not engineering work: Matt must approve the five seeded articles individually in the admin review console (/api/news/admin/review, #10123) — nothing publishes until that happens, by design. Still ⏳ pending that manual approval, in both UAT and prod. † ‡✅
marketing.news-automation.content-ci-validation—news-content.jsonis content-validated in CI. Resolved by #11123 (no new code needed):.github/scripts/check-proprietary-naming.mjs(added by #11112, wired intoci.yml'sProprietary naming guardjob) explicitly scansinstances/marketing/scripts/news-content.jsonfor banned proprietary product/feature naming (CAPE) on every PR.instances/marketing/scripts/verify-policy-fidelity.mjs— the other content-validation script in this package — is unrelated: it verifies the/informationpolicy pages against Kristin's source PDFs (POLICY_MAP) and has never readnews-content.json; it is not dead code, it is actively gating a different surface, and is correctly left unwired fromnews-content.jsonvalidation. ‡✅
marketing.news-automation.seeded-articles-live— the News page carries current, published content and the header News item is therefore visible. This closes the loopinitial-abc-pinsleft open: the KV namespace now exists AND approved articles are seeded, so/api/newsreturns a non-empty array for the first time. Three link-out articles were added (2026-09-01): MJA InSight+ on doctor burnout (2026-08-17) and on CPD requirements (2026-08-10), and the AMA (WA) Hospital Health Check (2026-05-05). Each was verified live before inclusion — title, author, date and quoted figures fetched from the source rather than taken from a search snippet — and each carries asourceUrllink-out with original "Why This Matters for PPP" commentary rather than republished source text. A well-ranked AHPRA-audit guide was deliberately rejected: it reads as neutral advice but is published by ClinicComply, a competing compliance-software vendor, and because articles link OUT it would have routed prospective members to a competitor. Approval route:initial-abc-pinsnames Matt approving in the admin review console (/api/news/admin/review); these three were approved by Rob, explicitly and on the record, in session (2026-09-01) and seeded withreviewStatus: 'approved'directly. Rob is an authorised approver for marketing content — the console is one approval surface, not the sole source of authority — so this is a valid approval, not a bypass. What the console would have added is an audit trail; that role is served here by this criterion and the commit history naming the approver and date. The review gate itself is untouched and still fails closed — the five legacy articles carry no status and correctly remain unpublished, verified live. UAT only; prod seeding is a publishing action and stays gated on Matt. † ‡✅
marketing.news-automation.rss-discovery-adapter— a realDiscoverySourceexists:createRssDiscoverySource(feeds, options)fetches RSS 2.0 feeds, parses items intoDiscoveredArticles and merges them, settling OQ-10123b's mechanism (Rob, 2026-09-01: "start with an RSS"). It plugs into the slotnewsAutomation.tsalready reserved, so the module's own warning — do not hard-code a source and ship it as "the design" — is respected: the interface is unchanged and the choice stays reversible. Deliberately no XML library: this is Workers-runtime code, guarded by thefunctions/runtime-purity gate, so a scoped regex over the flat RSS<item>shape is smaller and safer than pulling a parser into the wrangler bundle; it is not a general-purpose XML parser and must not be reused as one. Fails soft per feed (Promise.allSettled): one publisher being down cannot cost the run every other feed. De-duplicates bysourceUrlso a syndicated story yields one candidate, and drops items older than a configurable freshness window (default 90 days) so the page cannot fill with archive material; undated items are kept, since a missingpubDateis not evidence of age. Discovery still never publishes — candidates remainpendingand the review gate is untouched. Verified by mutation: removing de-duplication, the freshness filter, the incomplete-item skip, or downgradingallSettledtoalleach fails its own test and no other (11/11 green when restored). Not included: the feed list (Matt's editorial call, OQ-10123b) and the editorial writer (OQ-10123c, still stubbed — the model and prompt are Matt's decision). † ‡✅
marketing.news-automation.kv-binding-survives-sibling-writes—cf-ensure-pages-kv.sh(the script that bindsPPP_NEWSto the marketing Pages project on every deploy,initial-abc-pinsabove) now read-modify-writes the project's completekv_namespacesmap on every PATCH, instead of sending only the binding it is setting. The Marketing UAT CD job began failing 2026-09-15/17 (#12270): the script's own post-write assertion caught the pre-existingPPP_WAITLISTbinding disappearing after a single-key PATCH forPPP_NEWS— Cloudflare's Pages API was observed replacingdeployment_configs.production.kv_namespaceswholesale rather than merging it per key, contradicting the merge-per-key behaviour this script had previously verified and relied on. Left unfixed, this class of bug reopens the exact defectinitial-abc-pinsexists to prevent: a binding silently disappearing means/api/newsreturns the Cloudflare 1101 error again on the next deploy that binds a sibling first. Verified red-then-green: the pre-fix single-key payload against a fake Cloudflare API (holding a pre-existing sibling binding) reproduces the sibling's disappearance and the script's own assertion catching it; the fixed read-modify-write payload preserves both bindings. † ‡✅
marketing.news-automation.seeded-images-deduplicated— the 8 seeded articles each carry a distinct image (previously 5 of the 8 shared one generic PPP infographic, and 2 more shared a PPP app-screenshot, so News visibly "doubled up" as Matt described). For the 5 link-out articles, the image is the source publication's own lead/og:image, downloaded once intopublic/assets/news/rather than hotlinked, so rendering does not depend on the source's CDN staying reachable — this displays the same image a reader would see previewing the link elsewhere (a link-card image), not a republication of article body content. Every existing on-disk alternative to the shared PPP infographic was rejected: the only unused variants are byte-identical to what was already in use, or carry the retired "Athlete Doctor" name (docs/features/marketing.md's own Change Log records that name as retired, #7671) — reusing them would have been a regression, not a fix. This is the "safe, technical fix" half of #12268 only. The issue explicitly separated it from an editorial review Matt asked to do in person ("we need to go back and forth you and I a bit") — no new image sourcing, captioning or article content is decided here, and nothing about which stories to publish changes. † ‡✅
marketing.news-automation.rss-discovery-carries-image—parseRssFeednow extracts each item's own lead image, from whichever of the two common RSS conventions the feed actually uses: amedia:content/media:groupblock (medium="image"or an imagetype) or a plain<enclosure type="image/...">.DiscoveredArticle.imageUrlcarries it throughgenerateCandidatesinto the review-queue candidate'simagefield, replacing the unconditionalimage: ''every future auto-discovered candidate previously got regardless of what the source actually offered — the root causeseeded-images-deduplicatedabove only patches for the current seed set, not for articles this pipeline discovers going forward. No second fetch is added: both conventions are inline in the same item XML the parser already reads in one pass, consistent with the deliberate no-XML-library, Workers-runtime constraint on this module. An admin still reviews every candidate before publish (review-queue,admin-approve-rejectabove), so a missing or wrong source image is caught there, the same as any other field. Verified red-then-green: the two new extraction tests fail against the pre-fix parser (noimageUrlfield existed) and pass with the fix; a third test confirms a non-imagemedia:content(e.g. video) is not misreported as a lead image. † ‡✅
marketing.news-automation.native-articles-use-editorial-imagery— the 2 native (PPP-authored) seeded articles —doctor-burnout-research-2025andcpd-tracking-medical-board-requirements— were out ofseeded-images-deduplicated's scope (that AC covers only the 5 external link-out articles) and still pointed at repurposed marketing/product assets: a branded "From Burnout to Breakthrough" campaign infographic (with the PPP logo and campaign messaging baked in) and a product dashboard UI screenshot, respectively. Neither is appropriate as a neutral news-article hero image. Both now point at new, purpose-generated editorial-style photographs underpublic/assets/news/, matching the documentary/photojournalism tone of the 5 already-fixed link-out images — no text, no logos, no UI chrome. Unlike the link-out articles, there is no third-party source image to preserve here: these are PPP's own articles, so sourcing new imagery is our call, not an attribution question — the constraint is only that it must not be a recycled marketing-campaign or product-UI asset. † ‡✅
marketing.news-automation.on-demand-discovery— an admin can trigger discovery on demand from/news-adminvia a "Check for new articles" button, rather than waiting on a schedule (OQ-10123a's cadence question is unaffected — this is an additional manual trigger, not a replacement for a scheduled scan, which remains unbuilt).POST /api/news/admin/discover(instances/marketing/functions/api/news/admin/discover.ts) is gated behind the sameNEWS_ADMIN_TOKENbearer check asadmin/review.tsandwaitlist/admin/list.ts, callscreateRssDiscoverySourceagainst the 3 feeds OQ-10123b confirms above, and writes new candidates into the queue using the exact same mechanismseed-news-kv.tsand the approve/reject flow already use (per-articlenews:<slug>KV records, a maintainednews:slugslist, a rebuiltnews:index) — no second storage mechanism was introduced. Dedupes bysourceUrlagainst every existing record regardless of review status (pending, approved, or rejected) so a repeat scan or a re-syndicated story never adds a second candidate for the same article — the same signalcreateRssDiscoverySourcealready uses internally to merge one story across feeds. AI editorial commentary is explicitly OUT of scope (OQ-10123c, unchanged/still open): candidates carry the raw RSS snippet as their excerpt viastubEditorialWriter, andNewsAdmin.tsxshows a visible, tooltip-backed note that new candidates are unedited RSS content, not AI-drafted commentary, so the admin cannot mistake one for the other. Fails loud rather than silently: a feed fetch/parse failure, a KV failure, or a bearer-auth failure each surface as an explicit error response, and the admin UI shows a loading state while the scan runs and a clear error message (with the server'sdetailwhere available) on failure — never a silent no-op. Discovery still never publishes: every added candidate landsreviewStatus: 'pending', and the review gate (review-queue,admin-approve-rejectabove) is untouched. † ‡✅
marketing.news-automation.no-self-referential-launch-article—news-content.jsoncarries no article claiming the platform has already launched. Removedppp-platform-now-available("Professional Performance Portfolio Now Available for Medical Professionals",category: "Product Launch", dated 2026-03-01, in the past): it directed readers to sign up at a deadppp.embrace.technologydomain (not the liveprofessionalperformanceportfolio.com), was self-published with nosourceUrlunlike every other seeded article, and its "now available" claim contradicts the site's own pre-launch premise everywhere else (the waitlist copy, the Contact-page-era "since PPP hasn't launched yet" line). Seeded before the RSS-curated pipeline (rss-discovery-adapter,on-demand-discovery,add-by-urlbelow) existed, as a placeholder that was never revisited once real content arrived. † ‡✅
marketing.news-automation.add-by-url— an admin can hand-pick a single specific article Matt already has in mind (an AHPRA update, a story someone sent him) rather than waiting for it to surface via RSS discovery, independent ofon-demand-discoveryabove: a URL input + "Add article" button on/news-admincallsPOST /api/news/admin/add-by-url(instances/marketing/functions/api/news/admin/add-by-url.ts), gated behind the sameNEWS_ADMIN_TOKENbearer check asadmin/discover.ts/admin/review.ts/waitlist/admin/list.ts. The endpoint fetches the given URL server-side and extractsog:title/og:description/og:image(instances/marketing/src/data/urlArticleFetcher.ts), falling back to<title>and the first substantial<p>whenog:tags are absent — it fails loud (UnreachableUrlError/NotAnArticleError, surfaced as 502/422) rather than emitting a garbage candidate when the URL is unreachable or the response is not an HTML article page. No AI editorial commentary, matchingon-demand-discovery'sstubEditorialWriterdecision (OQ-10123c, unchanged/still open): the candidate's excerpt is the raw fetchedog:description/fallback, unedited —NewsAdmin.tsx's existing raw-excerpt note was broadened to cover both this form and the RSS button so the admin cannot mistake either source's excerpt for reviewed commentary. Still gated by review: the added article lands as apendingcandidate in the exact same queuediscover.tsandseed-news-kv.tswrite to (per-articlenews:<slug>KV records, a maintainednews:slugslist, a rebuiltnews:index) — this is not an instant-publish shortcut, and the approve/reject/pin flow (review-queue,admin-approve-rejectabove) is untouched. Dedupes bysourceUrlagainst every existing record regardless of review status, reusingdiscover.ts's own dedupe signal rather than a second implementation:discover.ts's previously-inlined auth/read/index/dedupe helpers were extracted intoinstances/marketing/src/data/newsQueueStore.tsso both endpoints share one implementation. A duplicate URL returns a specific409"already in the queue" message rather than a generic failure, checked before the network fetch so a known duplicate never costs a request to the source. Image is hotlinked, not downloaded: the fetchedog:imageURL is carried through as-is into the candidate'simagefield, the same conventionrssDiscovery.ts'sreadLeadImagealready uses for an RSS item's own lead image (rss-discovery-carries-imageabove) — there is no R2/object-storage binding anywhere in this Pages project for a runtime fetch to download bytes into;seeded-images-deduplicated's "downloaded intopublic/assets/news/" was a one-off manual step done offline against the 8 seed articles at commit time, not a runtime code path this endpoint could call, so this endpoint intentionally matches the one download/hotlink convention that actually exists in code today rather than inventing a second, different one. An admin still reviews every candidate's image before approval, same as any other field. † ‡
Current Status
⏳ Specified (draft). Spec-first; implementation tracked in #10123. Firm ACs trace to the 2026-07-07 Dev Session (concept approval, review-before-publish gate, initial ABC pins). Cadence (OQ-10123a) and AI model/prompt (OQ-10123c) remain Open Questions pending Matt and are deliberately left configurable rather than inferred. Discovery source (OQ-10123b) is now resolved — Rob confirmed the feed list 2026-09-19 (#12281): 3 verified RSS feeds (ABC Health, AMA WA, ASMOFQ); MJA InSight+ excluded for want of a discoverable feed, not decided against.
Implementation / Architecture
Additive to the #2809 KV-backed News page. Expected shape (subject to the Open Questions): a scheduled Cloudflare trigger drives discovery + editorial generation; candidates are written to KV (or a sibling namespace) with a review-status and optional pinned flag; an admin surface lists candidates and flips status to approved/rejected; the public /api/news read path returns only approved articles, pinned-first. NewsArticle (src/data/news.ts) gains the review-status and pinned dimensions.
Open GitHub Issues
Open issues on label:feature:marketing — no open issues. Fetched at: 2026-10-13T07:29:49Z — live list on GitHub.
Change Log
| Date | Change | PR/Issue | Reason |
|---|---|---|---|
| 2026-09-28 | Discover feed URLs corrected: ABC Health now abc.net.au/news/feed/9167762/rss.xml (channel titled "ABC Health") instead of 51120 ("Just In", all news); AMA WA now amawa.com.au/feed/?post_type=news instead of /feed/. utm_* tracking params are stripped from item links before sourceUrl dedupe, and the discover response carries a per-feed feeds breakdown (outcome, HTTP status, parsed/stale/duplicate/proposed counts). | #12534 | UAT test 2026-09-28: one run proposed 25 off-topic ABC stories and 0 from AMA WA / ASMOFQ. Verified live: 51120 is ABC "Just In"; AMA WA's default WordPress feed only carries the unused post type (valid RSS, zero items) while its articles are the news post type. ASMOFQ parses correctly — it publishes roughly monthly, so only one item (2026-07-31) is inside the 90-day window; the per-feed breakdown makes that visible instead of a silent "0". |
| 2026-09-24 | Public News page ordering is strictly newest-first (API function, orderForDisplay and the page itself); pinned-first removed, pinned still stored but no longer reorders, and the admin queue's Pin control removed. Home page no longer renders a Latest news panel. marketing.news-automation.pin-order superseded by newest-first. | #12430, #12419, #12422 | Matt, 2026-09-22 (~00:19:41, ~00:27:05): "reverse chronological order... there's no order at the moment"; the latest news panel is to be removed from the all-in-one page. Root cause of "no order": the two seeded ABC/QLD pins are dated 2023 and 2022 and pinned-first sorted them above every 2026 story. Whether pinning is retired or returns as a separate featured treatment is an open question for Matt. |
| 2026-09-20 | Removed the stale "Professional Performance Portfolio Now Available for Medical Professionals" seed article from news-content.json — it falsely claimed the platform had already launched, dated itself in the past, pointed at a dead embrace.technology domain, and was self-published with no sourceUrl unlike every other seeded article. | #12309, #11177 | Matt's 2026-09-19 review of the Task 2 walkthrough video flagged what he read as "AI generated news" on the site; investigation found this specific placeholder article, seeded before the real RSS-curated news pipeline (#12281/#12282) existed, still shipping as a permanent seed entry. |
| 2026-09-18 | De-duplicated the 8 seeded news article images (5 previously shared one generic PPP infographic) by downloading each link-out article's own lead/og:image into public/assets/news/; wired parseRssFeed to extract a media:content/enclosure image from future auto-discovered RSS items so the same gap does not reopen for articles this pipeline discovers going forward. | #12268 | Matt flagged on the 2026-09-16 weekly tech call that the News page was "doubling up on images" and showing "our old platform" screenshots as generic thumbnails. The issue explicitly separated this safe technical fix from an editorial review of image sourcing/captioning that Matt asked to do in person — that half stays open, tracked on #12268. |
| 2026-09-18 | Replaced the hero images on the 2 native (PPP-authored) news articles — doctor-burnout-research-2025 and cpd-tracking-medical-board-requirements — which were out of #12268's scope and still pointed at repurposed marketing/product assets (a branded campaign infographic and a dashboard UI screenshot) rather than editorial article imagery. | #12279 | Found while reviewing the local /news preview against #12268's shipped fix — those 2 articles are native content, not external link-outs, so #12268's og:image-download approach didn't apply. Rob confirmed we're free to source our own imagery for native articles, just not campaign-branded assets recycled as generic article thumbnails. |
| 2026-09-18 | The /news-admin review console now caches the NEWS_ADMIN_TOKEN in localStorage after a successful load, so a returning visit pre-fills the token field and auto-loads the queue instead of requiring the token to be pasted every time. Falls back to the old paste-required behaviour if storage is unavailable or cleared. | #12280 | Rob: "How do we avoid Matt having to paste a token?" Re-pasting a shared secret on every visit was pure friction with no security benefit — the gate is still the same bearer-token check server-side, just remembered client-side. |
| 2026-09-17 | cf-ensure-pages-kv.sh now read-modify-writes the project's full kv_namespaces map on every PATCH instead of sending only the binding being set. | #12270 | The Marketing UAT CD job began failing 2026-09-15/17: the script's own post-write assertion caught the pre-existing PPP_WAITLIST binding disappearing after a single-key PATCH for PPP_NEWS. Cloudflare's Pages API was observed replacing deployment_configs.production.kv_namespaces wholesale rather than merging it per key — contradicting the merge-per-key behaviour this script had previously verified and relied on. Sending the complete merged map is correct regardless of which behaviour the API exhibits. |
| 2026-08-27 | Settled both engineering-owned checkboxes from #11114's residual (#11123). Confirmed isPublished() already fails closed (reviewStatus === 'approved' only) per #11118's decision, and seed-news-kv.ts already writes an explicit reviewStatus: 'pending' on every KV key. Confirmed news-content.json is already content-validated in CI by .github/scripts/check-proprietary-naming.mjs (added by #11112, wired into ci.yml), so no new CI wiring was needed. verify-policy-fidelity.mjs is confirmed unrelated to news content (it verifies the /information policy pages against source PDFs) and is correctly left unwired from this file — not dead code, just a different gate. | #11862, #11123, #11114, #11118, #11112 | #11123 was opened as the owned follow-up for #11114's two remaining checkboxes: publishing seed content, and settling news-content.json's CI validation status. Both the fail-closed publish gate and the seeder's explicit reviewStatus default had already shipped in a prior PR by the time this issue was picked up, and #11112 (merged 2026-07-29) already wired a content validator for news-content.json — so the CI-validation checkbox needed confirmation and documentation, not new code. The remaining checkbox (publishing the 5 seeded articles) is an explicit manual admin-console approval action per Matt's #11118 ruling ('we only publish after approval in the admin console, no exceptions for seeded or imported content') — not something an engineering PR can or should do on his behalf. |
| 2026-07-28 | The PPP_NEWS KV namespace is now created and bound to both marketing Pages projects by CD on every deploy (and the binding is verified, not assumed), so /api/news serves instead of returning a Cloudflare 1101 "Worker threw exception" page; a missing binding now fails loud as a JSON 503 rather than an uncaught throw, and a deploy-time smoke test asserts the feed contract in both environments. | #11119, #11114 | #11114 — /api/news returned HTTP 500 on production AND UAT for the entire life of the endpoint. The KV namespace had never been created on the Cloudflare account and kv_namespaces was null on both Pages projects, so ctx.env.PPP_NEWS was undefined and .get() threw. Creating it was only ever a manual dashboard step nobody performed, and nothing in CD or Pulumi declared it. It stayed invisible because the CD health check only curls the site root — /news is a client-rendered SPA route that kept returning a healthy 200 shell while its data source was dead. |
| 2026-07-13 | BDD scenario coverage intentionally deferred — spec is status: draft (not yet built). No @feature:marketing-news-automation scenarios needed yet. | #10578, #10572 | Coverage/BDD gap audit (#10572) flagged this spec as having zero BDD tag match. Confirmed draft-stage per front-matter status:; BDD is written when the feature is built, not while the spec is still draft. This note stops future coverage sweeps from re-flagging the gap. |
| 2026-07-11 | Seeded and pinned the two ABC / Queensland Health junior-doctor-burnout articles at the top of the News page on initial rollout (link-out only, with a source-link render on the article detail page). | #10393, #10388 | Delivers marketing.news-automation.initial-abc-pins — Matt (Dev Session 2026-07-07): 'I'm keen to pin at the top definitely those ABC articles from Queensland Health... junior doctor's burnout... two articles.' The pin/review mechanism shipped in #10123/#10372; this seeds the deferred content once the two source URLs were confirmed. |