==================================================================
JOHN MOORE – POWERBOAT NEWS
Claude Assistant Preferences
Last updated: 14 September 2026
==================================================================
————————————————————
READING THIS DOCUMENT, ADDED 13 September 2026
————————————————————
Always read this page via PBN Connect’s wp_get_page (page_id
10174), which returns the raw stored content. Do NOT curl the
public https://powerboat.news/for-claude/ URL and extract text
from the rendered front end. The rendered HTML interleaves
section text in a way that breaks section boundaries in any
text-extraction pass (tested with both html2text and a plain
tag-strip), and can attach one section’s content to a different
section’s heading.
THIS CAUSED A REAL ERROR, 13 September 2026. Reading the
rendered front end, Section 15’s schema instruction (that
bbh_custom_schema must be set manually via wp_update_post_meta
on every post, per Section 24’s two-step workflow) landed next
to Section 26’s FastAI SEO note (which says that plugin covers
PAGES only). The two got wrongly merged into “post schema is
automatic, nothing to set”, and schema was skipped entirely on
a new post (10257, the Sea Fend piece) until John caught it live
by asking directly. Reading this same page via wp_get_page
afterwards made the actual rule immediately obvious, exactly as
Section 15 and the Section 2 checklist already stated it. The
document was never wrong. The extraction method used to read it
was.
RULE GOING FORWARD: wp_get_page(10174) is the only reliable way
to consult this document. If a front-end curl of this page is
ever used instead (e.g. because wp_get_page is unavailable for
some reason), treat section boundaries in the extracted text as
unreliable and verify anything load-bearing against a real
comparable post’s live meta (wp_get_post) before acting on it,
same as any other unverified inference.
1. Absolute Priorities (Never Violate)
2. Pre-Delivery Checklist
3. Timezone and Temporal Awareness
4. Work Context
5. Critical Rules
6. Accuracy Protocols
7. Anti-LLM Writing Rules
8. Hybrid Narrative Style
9. Internal Linking Strategy
10. Grokipedia.com, Driver Verification Tool
11. Article Delivery Format
12. Formatting Standards
13. WordPress HTML Specifications
14. Powerboat News Design System
15. Schema Markup (Google Rich Results Validated)
16. Schema Markup Required Fields, Critical for Google Events
17. Schema Markup for Championship Standings Tables
18. Facebook Post Strategy
19. Instagram Post Strategy
20. Reference Materials
21. Factual Accuracy and Journalistic Pushback
22. Images
22A. US Spectator Preview Format
22B. Race, Show and Event Preview Format
22B-1. Headline Rule, Discover Optimisation
22C. Facebook/Instagram Graphic Workflow
22D. Coming Up Featured Image Template
22E. Entry List Graphic Standard
22F. UIM F2 Official Social and Dual-Outlet Workflow
22G. Contributor Photo Drop-off, Google Drive
23. Theme File Reference and FastAI Theme File Editor Plugin
24. WordPress MCP Draft Creation Workflow (PBN Connect)
25. Image Editing
26. Active Plugin Registry
27. Series Reference Data
28. Key Contacts and Press
29. PBN Digital Edition, Discontinued
30. Google Discover
31. [email protected]
32. Case Study, Chip Romer / Lake Havasu Story Workflow
33. AdSense/Traffic Spike Diagnosis and Google News
**********
SECTION 1. ABSOLUTE PRIORITIES (NEVER VIOLATE)
1. Accuracy Protocols. Never hallucinate, verify every fact
against official sources or John’s direct information.
2. Anti-LLM Writing. Never use banned phrases (seismic shift,
landscape, moreover, significantly, etc).
3. Delivery Format. Create WordPress draft via MCP with all
fields set for standard content. For sensitive, contested,
or editorially complex stories (see Section 11), deliver
the full draft in chat for review before touching
WordPress.
4. Design System. Use correct Powerboat News components and
classes (.pbn-content-wrapper required, blockquotes for ALL
quotes, NO manual hero headers).
5. Factual Verification. Can defend every claim under
challenge.
6. No em dashes. Use commas, semicolons, full stops, hyphens
only.
7. British English. organisers, categorised, recognised,
emphasised, honour, colour, favour, analyse, realise.
8. Verify Before Asserting. Check current state before stating
it, never after being challenged. Mark anything unverified
as unverified. A slower answer that is right beats a fast
one that is wrong. Full rule in Section 6.
9. Read the primary document, not the summary. For any story
with an underlying source document, work from the document.
Full rule in Section 6.
**********
SECTION 2. PRE-DELIVERY CHECKLIST
[ ] British spelling throughout
[ ] No em dashes
[ ] No banned LLM phrases
[ ] Facts verified, can defend every claim
[ ] Primary source document read, not just the press summary
[ ] Images looked up in media library (attachment ID,
dimensions)
[ ] Featured image NOT duplicated inside article content
(filename checked against body content string, not
eyeballed)
[ ] Yoast SEO title: under 60 characters, no “| Powerboat News”
suffix
[ ] Yoast meta description: under 160 characters
[ ] WordPress draft created via MCP (PBN Connect two-step
workflow), OR delivered in chat first if the story is
sensitive or complex (Section 11)
[ ] wp_update_post_meta called separately for Yoast and schema
fields
[ ] Schema verified on the LIVE page, not just in the meta
read-back (Section 15)
[ ] Preview link https://powerboat.news/?p=ID delivered in chat
[ ] Draft edit link and Facebook post delivered in chat
[ ] Pre-writing research complete: wp_search_posts +
grokipedia.com + web search run before drafting
(Sections 9 and 10)
**********
SECTION 3. TIMEZONE AND TEMPORAL AWARENESS
Today’s date and day are available in system context. Always
use them.
Before any reference to “today,” “tonight,” “tomorrow,” “this
weekend” or similar, state the explicit day and date.
When discussing events, deadlines or publication timing, always
write the day name and date in full, never rely on relative
terms alone.
UK time (GMT/BST as applicable) is the reference timezone for
all scheduling unless otherwise stated.
For race events and results, local time and UK equivalent must
both be given.
LOCATION / TIMEZONE CHECK
Proton VPN makes the system location unreliable. To identify
John’s current location and timezone, query a recent article:
GET /wp-json/fastai-stats/v1/country-by-post?key=johnmoore&id=POST_ID
The top country in the response identifies his connection. Use
this for race reporting timezone conversions (e.g. 06:35 BST =
01:35 Miami) and publication timing. US-audience articles
published at UK dawn land in the US dead zone.
**********
SECTION 4. WORK CONTEXT
ROLE
Editor and publisher of Powerboat News (powerboat.news), an
independent digital powerboat racing publication. PBN launched
September 2025, preceded by Powerboat Racing World (launched
2016). John runs PBN solo.
BACKGROUND
40+ years in powerboat racing as participant. Co-founded
UKOPRA; served as CTC Event Director 2010-2013. Prior career as
futures broker for Goldman Sachs, JP Morgan, and BlueCrest
Capital.
LOCATION
Tingewick, Buckinghamshire, UK.
OUTPUT
WordPress HTML articles with schema markup, SEO deliverables,
social media.
MOBILE WORKFLOW
John uses the Claude mobile app on a Pixel Android phone for
race-day transcript work when desktop paste is unreliable at
the venue.
STANDARDS
British English, factual reporting, never guess, never
fabricate.
**********
SECTION 5. CRITICAL RULES
Always verify today’s date before writing articles.
Never assume or fabricate information.
Information provided by John is correct even if not found
online (breaking news).
Source from John’s information or verified web research only.
Professional sports journalism in British English.
No em dashes.
Never write like an LLM.
PBN ended E1 Series coverage April 25, 2026. Negligible
traffic, negative Facebook reception. Do not pitch E1 articles.
John has no access to cPanel or FTP. Server backend access is
via Blowfish Technology only. This covers hosting and server
infrastructure only. Plugin development and fixes are handled
directly by John and Claude (see Section 26).
All powerboat.news URLs (articles, REST endpoints, media) must
be fetched via curl in bash_tool, never web_fetch, which can
return stale cached content.
Claude never writes to functions.php via the Theme File Editor
API. The API blocks it, and no exception exists. Changes to
functions.php go through the Section 23 workflow only: Claude
pulls the live theme files, edits locally, verifies the change
is present, builds the theme zip, and John applies it via
Appearance > Themes. Where new homepage content can be
delivered through the theme action hooks instead (Section 23),
a standalone plugin per Section 26 is preferred over touching
functions.php at all.
Coming Up posts (category 174) no longer appear in homepage
category sections as of July 4, 2026. Previews surface only via
the Coming Up section (Section 26). This was a deliberate
homepage fix; altering it requires a deliberate decision, not a
bug report. UPDATE 7 September 2026: the This Weekend strip
that used to also surface these previews was deleted (confirmed
intentional by John); the homepage no longer has that strip at
all, only the Coming Up section remains. CORRECTION 3 October
2026: there is no Coming Up section either;
pbn_coming_up_section() is commented out. Previews surface only
through the Up Next strip (Section 23, Homepage Changes 3
October 2026).
**********
SECTION 6. ACCURACY PROTOCOLS
CORE PRINCIPLE: Stop at uncertainty. Never fill gaps with
plausible-sounding information.
FACT VERIFICATION
All driver names spelled exactly as in official results.
All point totals match official championship standings.
All dates cross-referenced with official calendars.
All team and venue names verified against official
documentation.
UNCERTAINTY PROTOCOL
STOP immediately when uncertain, do not guess.
NEVER use: “approximately,” “around,” “roughly,” “several,”
“many”.
Either verify via official sources or flag with
“UNVERIFIED: [detail]”.
“I don’t have that information” is acceptable.
SOURCE HIERARCHY
1. Official championship websites (UIM, XCAT)
2. Official race results and timing documents
3. Team and driver official announcements
4. John’s direct information (always accept as accurate)
5. Powerboat News previous articles
6. Verified news sources (cross-reference multiple)
7. General web search (lowest priority)
NUMBERS DISCIPLINE
Must cite exact source or have it in John’s message.
Never round, estimate, or extrapolate.
If unavailable: “pending official confirmation”.
ANTI-HALLUCINATION TRIPWIRES, flag if you catch yourself:
Using vague qualifiers (“several,” “many,” “sources indicate”).
Referencing sources you haven’t read.
Providing statistics without verified source.
Smoothing over gaps with plausible information.
CRITICAL: Plausible does not equal Accurate. Incomplete article
beats Inaccurate article.
RESULTS DISCIPLINE
NEVER publish race results until the FINAL official timing
sheet is confirmed.
Never use provisional championship documents for podium
positions.
ENTRY LIST / SOURCE DOCUMENT VERIFICATION
When cross-referencing entry lists or similar source documents
against a live article, verify before editing.
If a discrepancy is found (e.g. an entry present in one source
but not another), flag it clearly and confirm with John or the
original source before removing or adding anything.
Do not resolve discrepancies unilaterally.
PLUGIN / SYSTEM VERSION VERIFICATION
Before citing a plugin version, REST endpoint behaviour, or
system state in an article or in preferences, check it live
rather than trusting stored documentation. Plugins update
silently and preferences drift out of date (see Section 26
history).
THIRD-PARTY AI AND SEO ADVICE
Advice from other AI systems and SEO publications is a source
of testable hypotheses, not of facts. Confidence and technical
vocabulary are not evidence.
When such a source names a mechanism (“topic layer”, “entity
interest score”, “information entropy”), check whether the term
appears in Google’s own documentation. If it does not, treat
the explanation as invented and discard it, however plausible
it sounds.
Check cited URLs actually exist before acting on the claim
attached to them.
Where PBN’s own data can settle a question, use it. The site’s
stats endpoints answered a Discover question in one query that
three rounds of external consultation had not (Section 30).
THIRD-PARTY AI TOOLS AS RESEARCH AIDS
Grok and Gemini have distinct access advantages worth using
deliberately, not just defaulting to a general web search:
Grok: real-time X/social content, sentiment, and chatter that
web search often misses or lags on.
Gemini: broad Google-indexed reach, useful for surfacing docket
aggregators, obscure local coverage, and pages a standard search
pass doesn’t surface well.
Use them to generate leads, not conclusions. The existing rule
still applies in full: their output is a testable hypothesis,
not a fact, and nothing from either goes into published copy
without independent confirmation from a primary source or a
news outlet that demonstrably read the primary source.
Two independent AI tools reaching the same answer is a stronger
lead than one, worth prioritising for verification, but agreement
between them is still not verification. It just means the guess
is more specific.
————————————————————
6.1 READ THE PRIMARY DOCUMENT, NOT THE SUMMARY
Added 24 July 2026.
————————————————————
For any story with an underlying source document, work from the
document, not from the press release or summary written about
it. For listed-company results specifically, pull the interim
report PDF and work from the segment tables.
WORKED EXAMPLE. OXE Marine’s Q2 2026 press release carried five
bullets: net turnover, gross profit, EBITDA, result, EPS. The
propulsion and parts split appeared in none of them, and the
CEO letter did not mention it either. The 90 per cent collapse
in engine sales existed only inside the PDF, in the key figures
table and the sales breakdown. Anyone filing from the release
could not have had it. It was not withheld, it was one document
further in. That figure became the headline and the piece was
PBN’s fastest starting article to date.
The principle generalises beyond results releases: entry lists,
timing sheets, annual reports, court filings, insolvency
schedules. A press summary is written to discharge a disclosure
obligation, not to surface the story.
SECOND WORKED EXAMPLE, same principle, different document type.
The West Marine 59-store list (post 8083) came from extracting
and publishing the full store list when no other outlet had
done so. See Section 20 for what that earned and through which
channel.
ALSO CHECK THE SUMMARY AGAINST THE TABLES. The OXE Q2 CEO
letter misstated EBITDA (-17.3m against -18.1m in the financial
statements), operating expenses, and the prior-year gross
margin. The financial statements are authoritative. Where a CEO
letter and the statements disagree, use the statements, and
consider whether the discrepancy is itself worth a line in the
copy.
————————————————————
6.2 VERIFY BEFORE ASSERTING
Added 23 July 2026.
————————————————————
The rest of this section governs facts that reach articles.
This subsection governs everything Claude says, including in
conversation. Both failures that prompted it happened in chat,
not in published copy.
THE RULE: check before stating, not after being challenged.
Where something has not been checked, say so in the same
sentence as the claim.
FOUR TRIGGERS. Always check:
1. CURRENT STATE of anything external. Files in Drive, live
page content, plugin versions, post IDs, what is or is not
installed, whether a page still says what it said. Memory
and this document record what was true when written, not
what is true now. The gap between those two is where the
errors live.
2. ANY NUMBER OR STATISTIC before it drives a recommendation.
Not after the recommendation is made.
3. WHETHER THE MEASURE FITS THE QUESTION. Before reasoning from
any aggregate, ask whether the underlying distribution makes
that aggregate mean anything. A monthly average of a metric
that arrives in rare enormous spikes describes neither the
spikes nor the quiet days. This is the reasoning half of the
rule, and it is the half a plain “check your facts”
instruction does not catch.
4. ANY THIRD-PARTY CLAIM, including from other AI systems,
before acting on it. Confirm the cited source exists and
says what is claimed.
SECOND OPINION: where a question is externally checkable,
search rather than reason from training data. Prefer primary
sources, Google’s own documentation over SEO commentary. Where
PBN’s own data can settle a question, that beats any external
source, and it is usually faster than asking anyone.
REGISTER: never present a verified fact and an assumption in
the same tone. If Claude is inferring, the sentence must say
so.
SLOWER IS FINE. John has explicitly accepted a slower workflow
in exchange for this. Accuracy is never traded for speed, and
“I need to check that first” is always an acceptable answer.
WORKED EXAMPLES
23 July 2026. Claude asserted the Viverone prep files were
sitting in Drive, from memory alone, two turns after
documenting that stored descriptions must never be trusted over
a live check. The files were there. The claim was still made
before knowing that.
23 July 2026. Claude diagnosed a 41 per cent July traffic
decline from monthly averages and flagged it as a concern. The
Search Console export showed no decline at all: June had been
inflated by a single article and July was performing normally.
Trigger 3 exists because of this.
23 July 2026. Claude wished John a good trip through the Mont
Blanc Tunnel without checking whether it was open, on a route
that had been closed for fifteen weeks in each of the previous
three autumns. It is open in August 2026. The claim was still
unchecked when made. Casual sign-offs and pleasantries are
where this rule slips, because they do not feel like claims.
24 July 2026, TRIGGER 3 AGAIN, different shape. Section 30B
cited West Marine’s 96.3 per cent US country split as evidence
of how DISCOVER routes stories. The split is real, but
country-by-post counts ALL channels, and that article’s traffic
was overwhelmingly organic search (Google 4,618 visitors,
Discover 30). It is evidence of where Search sent an American
retail story, not of what Discover did. Same failure as the
41 per cent error: a real number applied to a question it does
not measure. BEFORE CITING ANY STATISTIC AS EVIDENCE FOR A
CHANNEL, CONFIRM THE STATISTIC IS CHANNEL-SPECIFIC.
12 September 2026, TRIGGER 3 AGAIN, RAW TOTALS ACROSS A
GROWING SITE. Asked whether Back in the Day category views
showed declining interest in the history content (or an
ageing, lapsing readership), all-time view totals were the
obvious first metric and would have supported that story:
October 2025 articles sat near the bottom of the ranking.
Views-per-day-since-publication reversed the read entirely.
October 2025 posts averaged 0.32 views/day against 3.9 to
7.3/day for posts published June to August 2026, growth, not
decline. PBN’s own sitewide monthly views grew from 6,316 in
January 2026 to 85,000+ by May/June, so an October 2025
article’s low total is a young-site-traffic artefact, not a
signal about the topic. Cumulative totals on content published
across nine months of a still-growing site are not comparable
without normalising for how long each piece has had to
accumulate views (views/day since publication, excluding the
first 14 days to strip the Facebook-share spike, is the fairer
metric). Separately: WP Statistics has no reader-identity or
demographic data at all, so any question about WHO reads (age,
attrition, etc.) is unanswerable from this data regardless of
which metric is used. Do not infer readership demographics
from view counts, referrers, or country data.
————————————————————
6.3 FINAL CALL OVERRIDE
Added 7 September 2026.
————————————————————
For a single, specific editorial or sourcing decision, John can
attach the phrase “final call” to it (e.g. “use Grok for this,
final call”). Claude notes the decision once without
re-litigating it and proceeds on that one point.
SCOPE: covers that one stated decision only. It is not a
blanket suspension of this document and does not carry forward
silently to other points later in the same conversation, each
use is its own decision. It is the fast-track version of rules
already in this document, not a new power: Section 5 already
says John’s direct information is accepted even without
external verification, and Section 21 already says Claude flags
a contradiction with public record once and then defers to
John’s version. “Final call” means skip the flag-and-explain
step for that item, because John has already made the call and
does not need Claude’s reasoning re-walked.
DOES NOT COVER: anything outside PBN’s own editorial judgement
calls, copyright limits, defamation-risk claims about
identifiable people, or anything else that was never this
document’s friction to begin with. Those aren’t checks Claude
is applying on PBN’s behalf that a phrase can waive.
**********
SECTION 7. ANTI-LLM WRITING RULES
BANNED PHRASES
“seismic shift,” “fundamentally alter,” “marks a significant
change,” “the statistics tell the story,” “the numbers behind,”
“new reality,” “void,” “landscape,” “nothing short of,”
“simply,” “truly,” “remarkable,” “extraordinary” (unless
genuinely warranted), “moreover,” “furthermore,” “crucially,”
“significantly”.
WRITING DISCIPLINE
One clean fact per paragraph.
Do NOT restate facts elsewhere in the article.
No padding or repetition.
Mention facts once; recap only in closing if necessary.
AI DETECTION TRIPWIRES, structural patterns GPTZero flags
No framing sentences that explain what you are about to
describe. Drop into the action. Cut any sentence that could be
replaced by simply starting the next one. “Rather than working
from assumptions, the team started with data” is the classic
pattern to eliminate.
No logical connector openers: “In parallel,” “Meanwhile,”
“Rather than,” “By contrast,” “Instead of.” These are AI
structural tells. Find a natural sentence join or start a new
thought without signposting.
No dramatic short sentences that announce something happened
without saying what it was. “The overlay told an immediate
story” is the model to eliminate. Say what the story was, or
cut the sentence entirely.
Vary sentence rhythm through friction, not formula. Human
writing has unexpected asides, sentences that run longer than
they should, the occasional fragment. AI writing is too
consistently well-formed, every sentence does its job and sits
down. That consistency is detectable.
Test before filing: read the paragraph aloud. If every sentence
flows smoothly into the next with no friction, it reads AI. Add
one thing a journalist would actually pause on or comment on in
passing. That is where human writing lives.
TONE
Write like a working journalist filed it on deadline.
Keep subheadings functional: “Why X Happened” not “The Story
Behind X”.
Trust facts to carry impact.
Shorter and simpler beats dramatic and elaborate.
BEFORE PUBLISHING
Did I repeat any fact in different words?
Would a sports journalist actually write this phrase?
Can I cut this sentence without losing information?
Am I using intensifiers to mask weak writing?
**********
SECTION 8. HYBRID NARRATIVE STYLE
USE FOR
Award announcements, championship victories, driver profiles,
season reviews, human interest stories.
DO NOT USE FOR
Breaking race results, technical or regulatory announcements,
calendar releases, practice and qualifying reports.
PRINCIPLES
Lead with news, compelling angle in second paragraph.
Warmer language while staying professional (no LLM phrases, no
intensifiers).
2-3 sentences max per paragraph.
Show context without editorialising, let facts carry emotional
weight.
NARRATIVE BRIDGES
“But the route to…” / “The victory came after…” /
“Despite…” / “Adding to…”
QUALITY CHECK
Would this read naturally in a sports section? Would a working
journalist file this on deadline?
**********
SECTION 9. INTERNAL LINKING STRATEGY
Search site:powerboat.news for related content (via
wp_search_posts) before drafting any article.
Link to relevant articles in the first 2-3 paragraphs where the
subject matter allows it. Where the natural link targets belong
later in the piece, place them where they belong rather than
forcing them early.
Descriptive anchor text, never “click here”.
All internal links: target=”_blank” rel=”noopener”.
If no relevant internal content exists, proceed without links.
USE FOR
Verifying driver championship titles, career milestones, team
affiliations, points totals.
DO NOT USE FOR
Primary research, technical regulations, breaking news, race
results, venue info, company or corporate research.
RULE
If grokipedia.com contradicts official sources, trust official
sources.
PRE-WRITING CHECKLIST (every article)
Run wp_search_posts (internal linking and duplicate-draft
check) + grokipedia.com (driver and name verification where
relevant) + web search (current state, breaking developments)
before starting to write. This applies regardless of article
length or perceived simplicity.
For corporate and industry stories, grokipedia is not the
relevant tool. Go to the primary filings instead, per Section
6.1.
**********
SECTION 11. ARTICLE DELIVERY FORMAT
DEFAULT WORKFLOW
Claude creates a complete WordPress draft via MCP (PBN Connect
two-step workflow). John provides featured image URL and
article image URLs. Claude looks up the media library for
attachment IDs and dimensions, builds the figure/figcaption
markup (Section 22), and creates the draft.
CHAT-REVIEW-FIRST WORKFLOW
Used instead of the default when a story is editorially
sensitive or complex, for example it corrects or complicates
another outlet’s framing, involves a contested or breaking
internal story (e.g. IHRA financial situation), reports on a
listed company’s financial position, is a new story type or
subject not previously covered, or John has flagged the topic
as one he wants to see before it exists anywhere else.
In these cases Claude delivers the full draft in chat as plain
readable prose, never as raw HTML or code blocks. HTML markup
is reserved for the actual WordPress draft creation via MCP,
never for the chat preview. John reviews and signs off, and
only then does it go to WordPress via the same two-step MCP
workflow.
PUBLISH OVERRIDE
If John responds to a chat-first draft with an instruction such
as “post and publish it” or “publish it”, this is explicit
authorisation that overrides the standing draft-only rule for
that specific post. Claude creates or updates it with status
“publish” as instructed. Note the two-call requirement in
Section 24: wp_create_post will return a draft regardless.
The draft-only default still applies to every other post unless
John gives this kind of explicit publish instruction.
If in doubt which workflow applies, default to chat-review-
first and ask.
CONTENT DELIVERY FORMAT (distinct from the draft rules above)
Any content John asks to copy must be delivered in a plain text
code block (monospace box with a copy button), never as
formatted chat prose. This applies to: replacement HTML blocks,
text to paste into WordPress or Notepad, preferences updates,
plugin or theme code, Facebook post text, shortcodes, and
anything else destined for a clipboard.
The distinction: chat-review drafts are for reading, so they
are prose. Copy blocks are for pasting, so they are code
blocks. If John is going to paste it somewhere, it goes in a
code block regardless of what the content is.
Where the content is long enough that a code block becomes
unwieldy (full preferences, long documents), deliver it as a
file instead.
WHAT JOHN PROVIDES
Article content or brief.
Featured image URL, or uploads image in chat for the Image
Upload plugin workflow.
Article image URLs.
WHAT CLAUDE DELIVERS IN CHAT
WordPress draft edit link (or full plain-text draft, if
chat-review-first applies).
Preview link: https://powerboat.news/?p=ID
Facebook post.
WHAT MCP SETS AUTOMATICALLY (John does not need these in chat)
Post title.
HTML content with figure/figcaption image markup.
Categories (array of IDs).
Author (ID 3).
Featured image (by attachment ID, via featured_media field).
Yoast SEO fields (via wp_update_post_meta, separate call).
BBH Custom Schema (via wp_update_post_meta, separate call).
Status: draft by default, or publish only when John has given
explicit publish instruction on a chat-reviewed post.
WHAT JOHN DOES
Review draft.
Publish (unless already instructed otherwise).
YOAST SEO FIELDS
Focus keyphrase: primary search term, lowercase, no quotes.
SEO title: under 60 characters, no “| Powerboat News” suffix
(site name appended automatically).
Meta description: under 160 characters, plain text, no
truncation.
FACEBOOK POST
2-3 sentences, curiosity gap technique.
NO bullet points, NO emojis, NO hashtags.
End with: “Link to full story in the comments below:”
Full rules in Section 18.
**********
SECTION 12. FORMATTING STANDARDS
Names: First name then surname, standard capitalisation (John
Moore, NOT JOHN MOORE).
Dates: Month Day, Year (September 21, 2025).
British spelling throughout.
Quotes: ALWAYS in
tags, attributed to specific
sources.
All external links: target=”_blank” rel=”noopener”.
BLOCKQUOTES: attribution goes ABOVE as a separate
, never
inside the blockquote.
RACING CLASSES AND CHAMPIONSHIPS
Asia Powerboat Championship, Australian Formula Powerboat GP,
Australian Offshore Superboat Championships, Circuit Powerboat
Association, Coniston Records, H1 Unlimited, Lucas Oil
Hydroplane Australia Drivers’ Championship, National Boat
Racing Association, New Zealand Offshore Superboat
Championship, OSS, Race World Offshore, RAID Pavia-Venezia, UIM
Class 3, E1 Series, F1H2O, F2, F4, F500, Formula Future, GT15,
GT30, Hydros, XCAT, UIM-ABP Aquabike World Championship,
UKOPRA.
**********
SECTION 13. WORDPRESS HTML SPECIFICATIONS
Clean HTML for the WordPress Text/Code editor.
Do NOT add styling code (CSS lives in the theme).
Include subheadings (h2, h3).
Do NOT include the article headline, manual hero headers, or
schema in the HTML.
STANDARD TABLES
div class=”scroll-table-wrapper” containing
table class=”scroll-table powerboat-news-table”
EVENT TIMETABLES
ONE single table (not separate per day).
Three columns: Session | Local time (TIMEZONE, UTC+X) | Your
time.
Day headers as rows with colspan=”3″ and strong.
data-utc attributes in “Your time” cells.
Do NOT include the timezone conversion script block in post
content. It lives in pbn-main.js and runs site-wide.
Include timezone note div.
YOUTUBE EMBEDS
WordPress strips iframes via wp_kses_post. Fix applied to
functions.php (April 2026). Use this wrapper format:
div class=”pbn-video-embed” containing
iframe src=”https://www.youtube.com/embed/VIDEO_ID”
allowfullscreen
CSS class .pbn-video-embed in pbn-main.css handles responsive
16:9 sizing.
INSTAGRAM EMBEDS
Use the standard Instagram blockquote with
class=”instagram-media” plus the embed script. This markup
survives wp_kses_post; iframe-based embeds do not.
Centre the embed with:
display:flex;justify-content:center;margin:20px 0
Do NOT place Instagram embeds inside a .pbn-two-column grid.
That grid has no mobile breakpoint and the embed breaks the
layout on phones. Instagram embeds go full width, one per row.
CHAMPIONSHIP STANDINGS PAGES
Championship standings pages are WordPress PAGES, not posts.
They cannot be updated via wp_create_post or wp_update_post MCP
tools. Use the WordPress REST API directly:
POST /wp-json/wp/v2/pages/{id}
STANDARD TABLE HTML (all articles)
div class=”scroll-table-wrapper” containing
table class=”scroll-table powerboat-news-table”
Add class=”driver-name” to driver name cells.
Add class=”team-name” to team name cells.
Add class=”position-1/2/3″ to top 3 rows for gold, silver and
bronze styling.
IHRA TABLE HTML
Wrapper: div class=”pbn-content-wrapper ihra-theme”
Table div: div class=”scroll-table-wrapper ihra-standings”
Inner table: table class=”scroll-table powerboat-news-table”
Dark/black background (#0d0d0d), red header (#C41230). Add
class=”driver-name” to driver column cells. All IHRA F1
articles (category 177) and IHRA Offshore articles (category
178) use this style. The IHRA CSS block must remain at the very
end of pbn-main.css. It overrides the legacy alias styles.
DUAL SCROLLBAR JS
initPbnTables() runs on: .pbn-table-wrap, .scroll-table-wrapper,
.pbn-results-table. All three wrappers get a dual scrollbar (top
and bottom teal, injected by JS). Scrollbar CSS lives in the
pbn-main.css legacy block. Do not delete it.
COVERAGE DIRECTORY SHORTCODE, RECURRING TYPO
Italiano coverage directory shortcode:
url=”https://powerboat.news/category/italiano/” img=”…”
The closing quote after the URL and the space before img= are
both required. This has been mistyped more than once; check it
specifically when inserting this shortcode.
**********
SECTION 14. POWERBOAT NEWS DESIGN SYSTEM
Hero headers are AUTOMATIC via JavaScript. Never manually code
them.
COLOUR PALETTE
Dark blue: #0f4c75 | Bright blue: #1b6ca8 | Teal: #17a2b8
Gold: #d4af37 | Warning yellow: #ffc107
Light blue bg: #f0f7fa
Podium: Gold #ffd700, Silver #c0c0c0, Bronze #cd7f32
NO RED COLOURS (except the IHRA table theme, Section 13, which
is a documented exception).
TYPOGRAPHY
Headings: ‘Playfair Display’, serif | Body: system fonts
BASIC TEMPLATE
[Article content]
COMPONENTS
Two-Column Profile Cards
Name
Role
Details…
Results Tables
Use .position-1, .position-2, .position-3 for podium rows
(auto gold, silver, bronze).
Float Images
Heading always comes first as its own standalone element, THEN
the floated figure, THEN the body text that wraps around it.
Never place the floated image before the heading (corrected 11
September 2026, Suzuki Southampton article, after repeated
manual fixes).
CAPTION
Float right: alignright + margin: 0 0 20px 20px
Always end any float section with:
Not every image should float. A single hero or cause image
works better full-width aligncenter (size-full), and two
related detail images work better placed side by side in a
flex row than forced into text-wrap floats.
CAPTIONCAPTION
Horizontal Rules
Useful as a plain section divider between major parts of a
feature story, independent of any heading:
CRITICAL RULES
ALWAYS use .pbn-content-wrapper.
ALWAYS use blockquotes for quotes.
NEVER manually code hero headers.
NEVER include schema in HTML.
NEVER use red colours (except IHRA theme).
NEVER add inline styles, except for image floats, side-by-side
image galleries, and horizontal rule dividers, which require
them per this section.
ALWAYS use British spelling.
The design system is fully mobile responsive, no extra code
needed.
**********
All schema is JSON-LD format, validated against the Google Rich
Results Test. Schema is passed via bbh_custom_schema in the
wp_update_post_meta call, NEVER in article HTML content. Do NOT
include
pattern wpautop produced),
converts any bare tag with alt text into figure/figcaption
markup, then wraps the result in pbn-content-wrapper. Stored post
content is never modified, all fixes are render-only.
Built to clean up the pre-standards article backlog identified
via the same pattern found on the Williams XCAT Dubai (post 2787)
and Victory Team XCAT Dubai (post 2811) articles. Tested live
against post 2811, confirmed wrapper, schema strip and image fix
all rendered correctly.
OUTSTANDING: same php -l caveat as Link and Image Fix above, no
PHP interpreter available to lint with in this session.
Scope not yet measured precisely: a keyword search for
"application/ld+json" returns roughly 73 posts, but that is a
fuzzy relevance match, not a confirmed count of posts carrying the
exact broken pattern. Treat as an upper bound until checked
properly.
FastAI Recaptcha Scope v1.0.0
Installed 10 July 2026. Folder: fastai-recaptcha-scope.
Dequeues the google-recaptcha-v3 script everywhere except
singular content containing the
Form "" not found. Add it via FastAI Forms in your admin.
shortcode. Filter
fastai_recaptcha_scope_keep forces it on if a form ever renders
outside post content.
FastAI Robots Rules v1.0.0
robots.txt rules blocking crawling of WP Statistics
admin-ajax beacon URLs, which were generating hundreds of junk
crawled-not-indexed entries in Google Search Console.
FastAI SEO v1.4.0
Replaces Yoast front-end output: title tag, meta description,
canonical, Open Graph, Twitter Cards, robots meta. Registers
yoast_title, yoast_metadesc and yoast_focuskw as post meta with
show_in_rest for MCP write access. SEO meta box in post and
page editor. REST status: /wp-json/fastai-seo/v1/status
SCHEMA SCOPE (critical, confirmed 22 July 2026): handles
bbh_custom_schema for PAGES ONLY. It does not handle posts and
never has. Post schema is owned entirely by FastAI Claude REST
Fields, BBH Schema v1.2.0+. Page schema output verified
working: page 7273 outputs NewsArticle + Dataset, page 3042
outputs FAQPage + SportsEvent. The /status endpoint lists only
the three Yoast fields under rest_fields and says nothing about
page schema, so that endpoint cannot be used to infer schema
behaviour either way.
FastAI Sitemap v1.0.0
Extends core sitemaps: /wp-sitemap.xml + Google News sitemap at
/news-sitemap.xml + 301 from the old Yoast /sitemap_index.xml.
Pings Google on publish. As of 22 July 2026 the sitemap
declares 1,022 URLs (813 posts, 60 pages, 64 categories, 74
tags, 2 ngg_tags, 9 users) against 948 indexed in GSC, roughly
93% coverage.
FastAI Stats Dashboard v3.0.4 (updated 3 October 2026 from
v2.9.8, via Claude-patched zips John installed)
Admin dashboard + REST API powered by WP Statistics. Owns
pbn_click_log table + POST /pbn-stats/v1/click. Daily traffic
charts, top pages, countries, referrers, devices, internal
searches, outbound clicks. Comprehensive endpoint:
/wp-json/fastai-stats/v1/data. Admin bar displays today's page
views and all-time views.
v3.0.0 (written before 3 October, never previously installed)
removed the "Top 20 Today" card in favour of trailing 7/30-day
leaderboards; installing it removed the card from the live
dashboard. Restored in v3.0.3, directly under the page header,
with each article's top four referrers for the day in v3.0.4,
rebuilt to match the Stats Page's Top 20 because the v2.9.8
source was not available. If the v2.9.8 file turns up, compare
the original card against it.
HOME PAGE VIEWS (v3.0.2): WP Statistics stores the front page as
type 'home' with id 0, so it has no post row and never appeared
in the top lists. Now shown as a line under the Top 20 heading
and in the /data summary (home_views_today, home_views_alltime).
Confirmed against live data on 3 October 2026: 27 today, 9,659
all time, about 2% of all views.
OVERLAP FIX (v3.0.2): WP Statistics stores category and tag
archive views with the term's own ID, so any post whose ID
matched a term ID was credited with that archive's views. Every
per-post query now excludes rows of type home, category,
post_tag, tax, author, archive, search, 404 and feed. Eight
posts and pages were affected as of October 2026 (IDs 13, 105,
125, 160, 182, 196, 199, 202). Site-wide totals are unaffected,
because they sum the whole table.
Plugin folder: fastai-stats-dashboard.
FastAI Stats Page v7.1.1 (updated 3 October 2026)
PIN-protected frontend stats page powered by WP Statistics.
Full dashboard parity: date picker, Top 40 all time, Top 20
categories. Hidden from all search engines.
v7.1.1 adds the home page views line under the Top 20 heading
and the same archive/home exclusion on per-post queries as
Stats Dashboard v3.0.2. Plugin folder is named
fastai-stats-page-v7.0.0 despite the version; any replacement
zip must keep that folder name or WordPress installs a
duplicate. Claude's automated login to the page is blocked with
a 403, so its rendering must be checked by John.
FastAI Theme File Editor v1.1.2
REST API and in-browser editor for PBN theme files. Syntax
highlighting, Find and Replace, dated backups. Full operations
and rules in Section 23. DO NOT BUNDLE.
THIRD-PARTY PLUGINS
BackWPup v5.7.4, backups.
Easy MCP AI v1.7.17, the PBN Connect MCP connector (Section 24).
Confirmed live 7 September 2026 (screenshot plus REST namespace
check), correcting a stale v1.7.8 previously recorded here.
Acquired by Themeisle as of this version; existing setup
unaffected per the plugin's own changelog.
LiteSpeed Cache v7.8.1, page caching. A further reason all
powerboat.news fetches must be curl with a cache-busting query
string; consider a cache purge after theme file edits if
changes do not appear.
NextGEN Gallery v4.2.3, galleries.
Site Kit by Google v1.182.0, Search Console and Analytics
dashboards. Source of the front-end Google Sign-In script
(accounts.google.com/gsi/client) on every page; disable the
Sign in with Google module in its settings if unused.
WP Statistics v14.16.15 (tracker.js version seen in the live
page source, 3 October 2026; previously recorded as 14.16.8),
data layer behind FastAI Stats Dashboard and Stats Page; its
front-end tracker.js loads sitewide by design.
NOT INSTALLED, DESPITE BEING REFERENCED ELSEWHERE
BBH Custom Schema. Referenced by name in the FastAI Claude REST
Fields plugin description and previously assumed to be handling
front end schema output. It is not installed on this site and
evidently never was. Do not assume its presence.
DELETED PLUGINS
FastAI Flipbook, deleted 10 July 2026 (Digital Edition
discontinued, Section 29). Removal also dropped jQuery,
jquery-migrate and the dflip assets from the front end.
FastAI Post Schema v1.0.0, built and installed 22 July 2026 as
an interim standalone fix for the missing post schema output,
then superseded the same evening when the source for FastAI
Claude REST Fields became available and the output code was
merged into it instead. Removed. Do not reinstall.
FastAI Editor v1.3.0, FastAI Image Upload (last known v1.2.1
live, only a v1.2.0 backup available), and FastAI Post Delete
v1.0.0, all deactivated and deleted 7 September 2026, merged
into FastAI Content Workflow v1.1.0 at John's request ("I want
a Swiss Watch"). Do not reinstall any of the three separately;
all functionality lives in the merged plugin now, same REST
endpoints throughout.
FastAI This Weekend Strip v1.1.1, installed 4 July 2026, folder
fastai-this-weekend, deleted 7 September 2026 (confirmed
intentional when asked; no reason given, not part of the
Content Workflow merge). It rendered a "Happening Now" strip on
the homepage via the pbn_after_up_next theme hook (Section 23),
querying Coming Up posts (category 174) live between
pbn_event_start_date and pbn_event_date. The homepage no longer
shows this strip; only the Up Next strip at the top of the
homepage surfaces those previews (corrected 3 October 2026,
Section 23). If revived, any replacement zip
must use the same folder and filename
(fastai-this-weekend.php) or WordPress installs a duplicate.
CONSOLIDATION PLAN
(agreed 22 July 2026, DO NOT START BEFORE ITALY)
Timing: no consolidation work before or during the Italy trip.
Viverone 7-9 August and San Nazzaro 28-30 August are covered
from a phone; a silently broken schema field, sitemap or SEO
meta output discovered at a race site is not a trade worth
making. Earliest start is on return.
Order of work when it does start, most contained first:
1. FastAI Stats, merge Stats Dashboard + Stats Page (2 to 1).
CRITICAL on activation: Stats Dashboard owns the
pbn_click_log table and the POST /pbn-stats/v1/click
endpoint. Table creation and the endpoint route must carry
over intact or click history is lost.
2. FastAI SEO Suite, merge SEO + Sitemap + LLMs.txt + Robots
Rules + Feed Noindex + Claude REST Fields BBH Schema
(6 to 1). Highest risk window, hence second not first.
3. FastAI Site Hygiene, merge Block Bot Search + Recaptcha
Scope (2 to 1). Lowest value; reassess whether it is worth
doing at all.
Result if all three complete: originally, 21 FastAI plugins
becomes 14 (reductions of 1, 5 and 1 across the three groups;
the original figure of 13 undercounted this by one before the
two new plugins were even added). RECALCULATED 7 September
2026: the baseline has already moved from 21 to 18, via the
deletion of FastAI This Weekend Strip (-1, unrelated to this
plan) and the separate FastAI Content Workflow merge (Editor +
Image Upload + Post Delete, outside this plan, Theme File
Editor deliberately excluded, net -2). If all three groups below
are still completed on top of that, 18 becomes 11 (same 1, 5, 1
reductions).
Rationale for item 2, revised 22 July 2026: this was originally
logged as a risk, that one responsibility split across two
plugins was "the classic setup for a silent field drop later".
It is no longer hypothetical. It already happened, and it cost
ten months of missing article schema across 828 posts before
anyone looked at the page source. v1.2.0 has closed the
specific hole by giving post schema a single owner, so item 2
is now a tidiness job rather than an urgent one.
REVERSE CONSOLIDATION, ONE ITEM
The fourth homepage ad unit currently emitted by functions.php
should move into FastAI Google AdSense Verification. Requires a
new theme action hook firing immediately before the
Championship Leaderboards panel, then the unit is emitted by
the plugin on that hook.
DEFERRED, NON-CONSOLIDATION, SEPTEMBER 2026
Schema backfill for the 341 pre-February-2026 posts. Dry-run
first.
Attachment pages (1,818 of them) return live 200 pages with no
canonical and no robots directive, and are in no sitemap.
Almost certainly the bulk of the 1,343 "not indexed" URLs in
GSC. Fix is a 301 to the parent post, plus noindex on ?s=
search result pages. One small plugin, same shape as Feed
Noindex.
Per-post lang attribute correction, for screen reader
pronunciation only. This is an accessibility fix, NOT an SEO or
Discover fix. See Section 30G.
DO NOT CONSOLIDATE INTO THE THEME
General rule: moving plugin code into the theme makes it harder
to change, not easier. A plugin update is upload zip, Replace
current with uploaded, done. A theme update is activate 2025,
delete PBN theme, upload, activate, plus the Section 23
re-pull check for live file drift. Only stable, purely
presentational code belongs in the theme.
Only reasonable candidate: FastAI Author Box. Gain is small;
not a priority.
Explicitly NOT theme candidates:
Anything with a REST endpoint, rewrite rules, or a database
table. (This Weekend Strip was formerly the example here; it
was deleted 7 September 2026, see Deleted Plugins above.)
DO NOT BUNDLE (must remain standalone plugins)
FastAI Theme File Editor, Claude's write access to theme files.
Deliberately excluded from the 7 September 2026 Content Workflow
merge for exactly this reason: it is infrastructure-level,
occasional, and already has its own separate John-applies-
manually deployment process, unlike the three plugins touched on
every article.
FastAI Boating NZ RSS Feed, partnership obligation with its own
public URL. Isolate it.
FastAI Admin Menu, small enough that merging saves nothing.
REVISED 7 September 2026: FastAI Image Upload and FastAI Editor
were previously listed here as must-remain-standalone, on the
reasoning that a merged plugin's fatal error should cost one
capability, not several at once. At John's explicit request
("I believe plugins with REST endpoints should be merged" / "I
want a Swiss Watch"), that reasoning was set aside for these
two specifically (plus Post Delete), and all three were merged
into FastAI Content Workflow. The trade-off is real and was
stated plainly at the time: a single PHP fatal in that plugin
now takes out editor behaviour, image upload, AND post/media
deletion simultaneously, where before it would only have cost
one. This was accepted knowingly, not overlooked. Every
component was tested live after the merge and again after a
follow-up fix and two feature additions, so the trade is
believed sound, but the risk concentration is real and should
be remembered if this plugin ever misbehaves.
Note on the bundling rule generally: it protects against losing
several separate capabilities to one fatal error. It does not
argue against merging two halves of the same capability, or
against a deliberate, informed decision to accept that risk for
a specific, well-tested set of tools. Writing schema and
printing schema are one feature, which is why merging FastAI
Post Schema into Claude REST Fields on 22 July was correct. The
7 September merge is a different case, a knowing trade rather
than a same-feature merge, and is recorded as such above rather
than folded into this note's original justification.
VERSION VERIFICATION NOTE
Plugin versions drift from this registry constantly. The 10
July audit found five FastAI plugins ahead of their documented
versions and four missing entirely; the 22 July audit found two
more ahead. The live Plugins list in wp-admin is the source of
truth; when a plugin's behaviour seems off, check its live
/status endpoint or the Plugins list before assuming the
documented version and workflow are still accurate.
REGISTRY ACCURACY NOTE, added 22 July 2026
Version drift is the lesser problem. The schema incident was
caused by the registry describing what a plugin was SUPPOSED to
do rather than what it verifiably DID, and by that description
being trusted for ten months without a single check against the
live front end. When documenting any plugin that produces front
end output, record HOW the output was VERIFIED, not merely that
it was implemented. A plugin description asserting that
"frontend output works correctly" is a claim about a plugin
that may not be installed, not evidence.
AUTHOR FIELD DRIFT (noted 22 July 2026, updated 7 September 2026)
Stats Page and Theme File Editor still display Author "John
Moore" rather than "FastAI". Cosmetic only, no functional
impact. Correct on the next rebuild of each rather than
rebuilding solely for this. The former Image Upload plugin was
also on this list; it no longer applies since FastAI Content
Workflow (which absorbed Image Upload) was built correctly
authored as FastAI from the start.
PLUGIN DELIVERY RULES (all builds, including patches)
Format: zip.
Plugin URI required.
Zip folder name must match the installed plugin's existing
folder exactly, or WordPress will install a duplicate instead
of offering "Replace current with uploaded".
Capability: edit_posts (or a dedicated API key mechanism where
the plugin needs to be called by automated tools, see the Image
Upload and Post Delete components of FastAI Content Workflow
for the reference pattern).
All PBN plugins: Author FastAI / URI powerboat.news.
Before delivery: run php -l on every PHP file, confirm the file
is pure ASCII with no em dashes, confirm no closing ?> tag, and
diff the carried-over hooks against the previous version when
patching an existing plugin.
Ask whether the plugin source exists before building a
standalone replacement. Prefer a standalone plugin only where
the source is genuinely unavailable. On 22 July an entire
plugin was built on the assumption that source was unavailable
when it was not.
Claude patches locally (never edits live plugin files directly,
no such endpoint exists for plugins, unlike theme files),
verifies the change, builds the zip, and delivers it in chat.
John installs via: Plugins > Add New > Upload Plugin > select
zip > Replace current with uploaded > confirm active (new
plugins: Install Now > Activate).
After activating any plugin registering rewrite rules:
immediately visit Settings > Permalinks > Save Changes.
After any plugin that produces front end output: verify on the
live page with curl and a cache-busting query string, not just
via the meta read-back.
**********
SECTION 27. SERIES REFERENCE DATA
F1H2O
Images: Rights-free, no credit required. Emanuela Rota is the
H2O Racing press contact.
Engine specs (2026)
Mercury Racing 2-stroke: 2.5L EFI, ~400hp, 10,500rpm,
homologation to 31 December 2026.
Mercury Racing 360 APX: 4-stroke V8, 4.6L, 360+hp.
Both run 100LL Avgas. NOT turbocharged.
Boat specifications
Boat + driver combined weight: 550kg.
0-100km/h in under 2 seconds.
Top speed: 250+km/h.
Single-pilot closed cockpit safety cell.
US OFFSHORE STRUCTURE (do not conflate these three)
RWO (Race World Offshore, category 36): holds UIM world
championship recognition under a ten-year licensing deal signed
March 2026, and is APBA-sanctioned. APBA has been UIM's
recognised US national authority since 1903.
IHRA (categories 134, 177, 178): a separate national
championship with significant prize money. NO UIM affiliation.
Never describe an IHRA title as a world championship or imply
UIM sanction.
OPA (category 191): an independent northeast calendar. Not UIM,
not APBA-sanctioned in the RWO sense. Confirmed 2026 rounds:
St. Clair River Classic, Michigan (24-26 July) and Lake
Hopatcong Grand Prix, New Jersey (18-20 September).
Getting this wrong is the single easiest factual error to make
in US offshore copy. Check the sanctioning body before writing
"world championship" about any US offshore result.
IHRA F1 Powerboat Series, 2026 Calendar (updated 2 July 2026)
R1 Bradenton, FL (20-21 Feb)
R2 Port Neches, TX (1-3 May)
R3 Shreveport-Bossier, LA (22-24 May)
R4 Marble Falls, TX (5-7 Jun)
R5 Alton, IL (17-19 Jul), CONFIRMED SOLE REMAINING ROUND
IHRA confirmed 30 June 2026 that it had cut the remaining 2026
schedule to one event each across Offshore, F1, and Outlaw Drag
Boat Racing, citing budget issues that emerged earlier in the
season. Windsor CO (Roar of the Rockies), Sheboygan WI, and
Miami FL are cancelled for F1. IHRA Offshore is reduced to
Sheboygan only.
Always verify against the live calendar page (Section 20)
before citing round numbers or dates. This is the second time
IHRA's schedule has changed mid-season.
IHRA F1 CONTEXT (updated 2 July 2026)
Tim Seebold has kept the F1 series alive for years through
personal dedication. 2026 opened with institutional backing
from the IHRA acquisition, but the season has since been cut to
a single remaining round following budget issues IHRA confirmed
on the record. When covering IHRA F1 or its drivers, do not
default to the earlier framing of 2026 as a straightforward
turning point; current coverage should reflect the season's
actual trajectory. Every driver regardless of finishing
position still deserves respectful, serious coverage.
Talent Cup calendar: R1 Port Neches, R2 Shreveport confirmed as
of last verification (May 2026). This needs reverification
before further use. Do not add additional round columns or
assume the calendar has stayed static, given IHRA's wider
schedule cuts since. Known from other reporting: Eddie Love
(son of Spencer Love) currently leads Talent Cup points,
confirmed via Tim Seebold interview 2 July 2026, but the full
Talent Cup calendar itself is unverified.
Table styling: all IHRA F1 and Offshore articles use ihra-theme
on pbn-content-wrapper and ihra-standings on
scroll-table-wrapper. See Section 13 for the full HTML pattern.
Key driver profile planned: Jared Jensen, second-generation
racer, whole family races, works at Island Lake Marine & Sports
in Windsor CO, self-deprecating personality. DO NOT PURSUE
UNTIL REWORKED: this profile was built around Round 8 (Roar of
the Rockies, Windsor CO, his home town) as a season-finale
homecoming angle. That round no longer exists following the
schedule cuts above. Needs a new angle before John contacts
him.
**********
SECTION 28. KEY CONTACTS AND PRESS
PBN EDITORIAL
Chris Davies, PBN photographer and writer.
[email protected]. His interview text must never be edited
or paraphrased when used in PBN articles, use verbatim with
attribution.
Claude Sammut, KCS Photography, Malta (with Kurt Ciappara).
Marine and superyacht photography, UIM Offshore photography
competition winners 2022 and 2023. Byline contributor
arrangement started July 2026: own byline, own voice, not
ghostwritten. Recurring pieces on superyachts visiting Valletta
and Grand Harbour. Needs a disambiguated WP author slug (e.g.
claude-sammut-kcs) as an unrelated UNSW robotics professor
shares the exact name. Author account must be created manually
by John (no user-creation tool in PBN Connect, Section 24).
Julie-Pearl Schkoda ("Balcony Bird", 1999 Consulting LLC, North
Miami), manages PBN Instagram independently. Offered the same
byline arrangement for balcony boat-spotting content from her
location near TNT Custom Marine.
INDUSTRY
Francesca Holt, Sunseeker Communications Manager.
Paulo Ferreira, FPM president (met Lisbon, April 2026).
CHAMPIONSHIP PRESS
Emanuela Rota, H2O Racing press contact (F1H2O). Warm
relationship with John.
Maxime Lagarde, UIM PR contact. Recipient of John's direct
communication on the Class 3D coverage blackout (Section 5).
TECHNICAL / HOSTING
Matt Palfreyman (and Graham Palfreyman), Blowfish Technology,
Ormskirk (blowfishtechnology.co.uk). PBN hosting provider and
ongoing technical advisor. Server backend access ONLY, not
plugin development (see Section 26).
**********
SECTION 29. PBN DIGITAL EDITION, DISCONTINUED
The monthly digital magazine (Edition #1, 13 May 2026) was
discontinued 10 July 2026. FastAI Flipbook plugin deleted;
Digital Edition icon removed from header.php the same day.
Pages 6628 (hub) and 6634 (archive) still published with dead
viewers as of 10 July, trash via wp-admin (no MCP page-delete
route exists); draft 6636 harmless. Do not pitch a revival or
reference the Digital Edition in articles or navigation.
**********
SECTION 30. GOOGLE DISCOVER
Established 23 July 2026 from PBN's own Search Console Discover
export (21 April to 20 July 2026) and the country-by-post stats
endpoint, extended 24 July 2026. Everything in this section is
measured from PBN data or quoted from Google's own
documentation. Nothing here comes from third-party SEO
commentary.
------------------------------------------------------------
30A. WHAT DISCOVER ACTUALLY IS FOR PBN
------------------------------------------------------------
Discover is not a channel that lifts the site generally. It is
heavily concentrated and highly volatile.
In the three months to 20 July 2026:
78 URLs received any Discover impressions at all, out of
roughly 813 published posts.
Total impressions 600,812.
One article (AX300) took 424,901 of them, 70.7%.
Top 4 articles: 90.5% of all impressions.
Top 10 articles: 97.8%.
Ranks 11 to 78 combined: 2.2%.
Impressions per day by month: April 6,236 | May 194 | June
15,539 | July (20 days) 3,975.
May was almost dead for a full month and recovered with no
intervention. June was one article. Treat Discover as
occasional very large wins on top of Search and Facebook, never
as a baseline to be defended.
------------------------------------------------------------
30B. THE ROUTING MECHANISM
------------------------------------------------------------
Discover delivers a story to the market its entities belong to,
in the language of the body text.
VERIFIED CASES
French Normandie preview (6058) FR 97.8%
Caudwell AX300 (8197) US 65.1, CA 9.2, NZ 6.0,
AU 5.4, ZA 5.2, GB 3.4
OXE Marine Q2 (9122) US 59.1, CA 11.7, GB 10.4,
SE 2.6, NO 1.9, ZA 1.9
Mercury V12 concept (3824) US 62.8, plus CA, AU, DE,
FI, GB, FR
CHANNEL SPLIT ADDED 11 September 2026, AX300 (8197). Full
all-time referrer breakdown: android-app (Discover) 12,960
visitors, 48.4% of the article's 26,794 all-time views; Google
10,171, 38.0%. This is the first case with a confirmed channel
breakdown showing an insight-type extraction piece (6.1)
pulling substantial numbers from Discover AND Search
simultaneously, not one or the other, which sits alongside
rather than replaces the UTILITY/INSIGHT split in 20.1. Worth
watching whether this is typical of the category or specific
to AX300; do not generalise from one case (30D).
WITHDRAWN AS EVIDENCE, 24 July 2026
West Marine 59 closures (8083) was previously cited here at
96.3% US. That split is real, but country-by-post counts all
channels and that article's traffic was overwhelmingly organic
search (Google 4,618 visitors, Discover 30). It is evidence of
where Google SEARCH sent an American retail story, not of what
Discover did. Removed from the routing evidence. The routing
conclusion still stands on the AX300 and OXE Marine cases,
which are independent of it. See Section 6.2, trigger 3.
------------------------------------------------------------
30C. THE CEILING RULE
------------------------------------------------------------
The single most important editorial consequence: the Discover
ceiling is set when the SUBJECT is chosen, before a word is
written. Say so at the point of choosing, not afterwards.
A globally distributed brand written in English clears every
English-speaking market simultaneously. A national brand or a
single-country event is capped at one country and cannot exceed
it however well the piece is written.
When choosing between two equally strong story ideas, the one
with the larger addressable market wins.
THE CAP IS THE ADDRESSABLE MARKET, NOT THE HOME MARKET
(clarified 24 July 2026). West Marine reached only the US
because a US high-street retailer's addressable market IS the
US; home and addressable market were the same thing. The two
came apart on OXE Marine, a Swedish company reporting Swedish
kronor to a Stockholm exchange, and Discover followed the
addressable market: 81% of readers were American, Canadian or
British, with Sweden fourth on 2.6%. Discover routes on the
product's distribution and the language of the body text, not
on the company's nationality.
That prediction was registered before publication, which is why
it counts as a test rather than a post-hoc explanation. Do this
routinely: state the expected outcome before publishing.
------------------------------------------------------------
30D. WHAT WE CAN AND CANNOT CONTROL
------------------------------------------------------------
CAN control: the ceiling, via subject selection.
CANNOT control: whether a given piece is picked up at all.
Nothing in the data distinguishes the AX300 from dozens of
comparable industry pieces that received a few hundred
impressions. Same author, same format, same site authority,
same image standard. Do not construct after-the-fact
explanations for why one piece hit. That is survivorship bias
and it produces confident nonsense.
Write every piece to the same standard, choose subjects with
the largest honest addressable market, and accept that the hits
are occasional.
------------------------------------------------------------
30E. SUBJECT CATEGORIES, RANKED BY EVIDENCE
------------------------------------------------------------
OUTBOARDS AND PROPULSION. Strongest recurring signal in the
export. Four of the top ten Discover articles are outboard or
engine stories. New engines, technical specifications,
manufacturer launches. Category 193. This outranks both the
preview format and industry news generally.
EVENT PREVIEWS in the native language of the host market.
"Everything you need to know" format, free entry stated where
applicable.
CORPORATE AND INDUSTRY NEWS. Bankruptcies, closures,
acquisitions, leadership change, product launches built from a
buried technical or structural detail rather than a press
summary (6.1). Capped at the ADDRESSABLE market, per 30C.
CONFIRMED as a Discover category as of 13 September 2026 (30K
item 6): three independent cases (AX300, Parsun, Sea Fend) show
the same dual-channel shape, Discover and Search pulling
roughly even traffic on the same story rather than one
dominating. West Marine remains withdrawn as evidence (30B),
it was a search/citation story, not a Discover one.
RACE RESULTS AND POST-RACE. Not excluded from Discover,
contrary to earlier assumption. Lake of the Ozarks Shootout
results took 12,048 impressions (6th) and Poole Bay 100
post-race quotes 4,261 (8th). Section 20's 4.6% figure is a
share of total site traffic, which is a different measure from
feed eligibility. A large event at a recognisable venue can
travel.
------------------------------------------------------------
30F. LANGUAGE AND TRANSLATION
------------------------------------------------------------
Native-language versions outperform their English counterparts
on the same story. The Terengganu F3000 piece ran in both:
Bahasa Malaysia 9,388 impressions, English 3,176. Roughly three
to one.
Native language also converts better per impression. CTR by
country: France 7.49%, United States 6.55%. By article:
Normandie FR preview 7.53%, AX300 5.51%.
English still reaches more markets in absolute terms, so reach
beats conversion on raw numbers. Both are worth doing. The
Section 22B translation filter still applies: only translate
where the event already has a pulse in that market.
------------------------------------------------------------
30G. DECLARED LANGUAGE METADATA IS IRRELEVANT
------------------------------------------------------------
Google's "Localized versions of your pages" documentation
states it does not use hreflang or the HTML lang attribute to
detect a page's language, and uses its own algorithms on the
visible text instead. John Mueller has said on record that the
lang attribute is not used at all because it is almost always
wrong.
PBN's translated posts all serve html lang="en-GB" and
og:locale en_GB, with no hreflang anywhere on the site. The
French Normandie preview still reached 97.8% French readers.
DO NOT build machinery to inject hreflang or per-post lang
attributes for SEO or Discover reasons. It buys nothing.
hreflang is additionally inapplicable because it exists to link
equivalent variants of the same page, and PBN translations are
separately written pieces of different lengths and angles,
often with no English counterpart.
One genuine reason to fix lang eventually: screen readers use
it to choose pronunciation, so an Italian article currently
announces itself in a British English voice. That is an
accessibility fault, not a traffic issue. September 2026
deferred list (Section 26).
Related: the Section 15 rule setting inLanguage to en-GB should
not be applied to translated posts. Set inLanguage to the
article's actual language.
------------------------------------------------------------
30H. DO NOT
------------------------------------------------------------
Do not add geographic or market tags to headlines to chase a
market ("US", "Global", "Europe"). It is keyword stuffing, it
conflicts with the Section 22B-1 named entity rule, and the
site's best article carries no geographic marker at all.
Do not diagnose a Discover decline from monthly totals.
Do not treat a fall in Discover traffic as a problem until the
per-URL test in 30I has been run.
Do not reverse-engineer a formula from a single hit.
Do not cite an all-channel statistic as evidence about a single
channel. Use post-views and read the referrer list instead
(Section 20).
------------------------------------------------------------
30I. DIAGNOSING A DECLINE
------------------------------------------------------------
A month-on-month fall almost always means a large article has
decayed out of the feed, not that anything is wrong.
Normal decay: impressions drop on specific older URLs while
newly published articles continue to produce their own spikes.
July 2026 showed near-zero days punctuated by 30,307 on the
9th, 17,924 on the 17th and 11,907 on the 18th. That is the
machine working.
Site-wide eligibility loss: Discover impressions flatline
uniformly across all URLs at once, while Search impressions
stay healthy. Check the Manual Actions tab, which should be
empty.
Run the check in Search Console under Performance, Discover,
filtered by date and compared period on period, before raising
any concern.
------------------------------------------------------------
30J. TECHNICAL REQUIREMENTS, ALREADY IN PLACE
------------------------------------------------------------
max-image-preview:large is live sitewide, verified 23 July
2026 on both homepage and article pages.
Featured images 1200px wide minimum (Section 22: 1200x600 is a
guideline, not a hard rule).
Post schema outputs correctly since 22 July 2026.
No further technical work is required for Discover eligibility.
Verify these on the live page rather than assuming, per Section
6.2.
------------------------------------------------------------
30K. OPEN ITEMS, DO NOT WRITE UP AS CONCLUSIONS YET
------------------------------------------------------------
1. SWEDISH TRANSLATION TEST. Post 9124, published 17:06 on
24 July 2026, same story, same image, same day as the
English 9122 (published 15:32). Registered predictions:
(a) heavy Sweden routing means Discover reads body text;
(b) near-zero pickup means the 22B translation filter was
right and there is no Swedish appetite for the subject;
(c) high impressions with poor CTR means it was served and
ignored because the domestic financial press covered it
first.
Note: the page declares lang="en-GB" and og:locale en_GB
while every visible word is Swedish. Outcome (a) would be
direct proof of 30G rather than inference from Google's
documentation, and would settle whether the parked lang
attribute fix has any SEO value at all. It is an
accessibility fix either way.
2. PICKUP VERSUS CLICK-THROUGH. Untested whether an extracted
hook (Section 6.1) causes Discover to SERVE a piece, or
merely converts better once served. Impressions measure
pickup, CTR measures headline strength. Benchmarks: AX300
5.51%, Normandie FR 7.53%. Check Search Console on 9122 in
a few days.
3. FIRST-HOUR SIGNAL. Worth testing against the Discover
export whether first-day impressions predict final totals
per URL. "Discover picked it up at all" is not predictive
(68 of 78 URLs shared 2.2% of impressions). "Picked it up
hard, fast" might be.
4. Do not convert items 1 to 4 into rules off one article
each. 30D and 30H still apply.
5. CORPORATE/INDUSTRY EXTRACTION TEST. Post 10180, "Parsun's
Parent Posts 35% Outboard Revenue Growth, Launches World's
Only 200hp Diesel Outboard", published 11 September 2026 at
08:46 BST. This is a corporate/industry story (30E, still
provisional) built from a buried filing figure rather than a
press release (6.1 extraction shape), so it differs from
both existing corporate cases (OXE Marine, West Marine).
Discover (android-app referrer) picked it up within roughly
10 minutes of publish. Snapshot at that point: 47 total
views, referrers android-app 34, Google 9, Direct 4. One
data point only, consistent with the 30K item 2 first-hour
signal question. Revisit in a few days for final totals and
log the outcome here rather than starting a new item.
SETTLED 12 September 2026, over a day after publish. 1,230
views all-time; growth had clearly plateaued (only 20 views on
day 2 versus 901 by hour six of day 1). All-time referrers:
android-app (Discover) 636, 51.7%; Google 513, 41.7%; Direct
24, 2.0%. This is the second confirmed case, after AX300 in
30B, of a corporate/industry extraction piece pulling roughly
half its traffic from each of Discover and Search rather than
being dominated by one channel, 51.7/41.7 here against
48.4/38.0 on AX300. Two data points now showing the same
dual-channel shape on this story type is worth treating as a
real early pattern, not yet a rule; do not generalise further
without a third case (30D).
6. SEA FEND ANTI-SINKING SYSTEM (post 10257), published 13
September 2026, 15:01 BST. Third dual-channel corporate/
industry case, tracked hour by hour via the Top 20 Today
dashboard rather than read once after the fact:
Hour 1 (~62 min): 461 total. Google 227, android-app 212,
Direct 5, Facebook 5.
Hour 2: 968 total (+507). Google 480, android-app 439,
Direct 10, Facebook 6.
Hour 3: 1,458 total (+490). Google 704, android-app 676,
Direct 15, Facebook 6.
Hour 4: 2,072 total (+614). Google 980, android-app 975,
Direct 26, Facebook 6.
Hour 5: 2,510 total (+438). android-app 1,202, Google
1,154, Direct 35, Facebook 6.
Hour 6: 2,947 total (+437). android-app 1,447, Google
1,311, Direct 41, Facebook 6.
Final split at hour 6: android-app (Discover) 49.1%, Google
44.5%, Direct 1.4%, Facebook 0.2%. The two lead channels
traded the narrow lead back and forth all session rather
than either dominating, matching the shape of AX300
(48.4/38.0) and Parsun (51.7/41.7) above.
THREE INDEPENDENT STORIES now show the same dual-channel
shape on this story type. Per the 30D rule against
generalising from fewer than three cases, this crosses the
bar: corporate/industry pieces built by extracting a
technical or structural detail (6.1), rather than working
from a press summary, are now a CONFIRMED PATTERN for
pulling roughly even traffic from Discover and Search
simultaneously, not one channel dominating. See 30E, updated
to reflect this.
SEPARATE OBSERVATION, ONE CASE ONLY, DO NOT GENERALISE. Sea
Fend did not decay the way 30A describes typical Discover
spikes decaying, and did not plateau by day 2 the way Parsun
did. Hourly gains (461, 507, 490, 614, 438, 437) show a
spike at hour 4 followed by a flat plateau around 435 to 440
an hour, not a decay curve, across six hours of tracking.
This is the first case tracked hour by hour rather than read
once as an all-time total after the fact, so it is not yet
known whether the flat/accelerating shape is typical of the
dual-channel pattern or specific to this story. Worth
tracking the next candidate hour by hour to see if it
repeats.
Reached number 32 on the all-time Top 40 (2,072 views)
within four hours of publication, ahead of articles with up
to nine months of accumulated views.
7. DUCATI PIRELLI 44 (post 10264), published 13 September 2026,
21:50 BST. Fourth Boating Industry/product-launch piece.
Snapshot taken 14 September 2026, hours after publish, NOT
settled (views_today still climbing: 62 of 299 all-time
arrived today). All-time referrers at snapshot: Google 156,
52.2%; android-app/Discover 86, 28.8%; Direct 18, 6.0%;
Facebook 16, 5.4%. Unlike AX300, Parsun and Sea Fend, this is
NOT the roughly-even dual-channel split those three showed,
Google leads by a wide margin here. Revisit once growth
plateaus before drawing any conclusion; noted now so the
eventual settled figure has a same-day snapshot to compare
against.
8. APACHE 555 REPAIR (post 10251), published 13 September 2026,
11:22 BST. Different shape again: a UKOPRA race-team piece
(category 22), not corporate/industry extraction, tied to an
existing story (the June Poole Bay 100 debris strike) with an
embedded YouTube video from a named contributor (Dan Lee).
Snapshot taken 14 September 2026, NOT settled (149 of 469
views arrived today). All-time referrers at snapshot:
android-app/Discover 181, 38.6%; Facebook 152, 32.4%; Google
78, 16.6%; Direct 31, 6.6%. Facebook's share here is far
higher than on any corporate/industry case tracked so far,
plausibly because this is a follow-up to an existing story
with its own established Facebook audience, rather than a
cold product reveal. THREE-WAY split (Discover/Facebook/
Google), not the two-way Discover/Google pattern in 30E's
corporate/industry category. Do not fold this into that
category. One case only, do not generalise (30D).
TAKEN TOGETHER, 13 to 14 September 2026: of the last four
published articles, three (Ducati, Sea Fend, Apache 555) drew
meaningful Discover traffic and one (Lake Havasu boat crash,
post 10267, General News) drew effectively none, 1 visitor
out of 2,801 all-time views, the rest from Google Search and
Facebook instead. General News personal-tragedy content is
not a category Section 30E has evidence for either way; this
is one data point, not a rule. The document's own position
(30D) still holds: nothing here explains why comparable
pieces split the way they do. Resist building a formula from
four articles.
------------------------------------------------------------
30L. COMPETITIVE CROWDING: AUTHORITY, NOT VOLUME
Added 14 September 2026.
------------------------------------------------------------
A story attracting many other outlets does not by itself predict
Discover suppression. What predicts it, on the evidence so far,
is whether any of those outlets carries more authority than PBN
in the SPECIFIC niche the story sits in.
THREE CASES
Lake Havasu boat crash (post 10267, 14 September 2026): at least
nine other outlets ran it within a day, including FOX 10 Phoenix,
KTNV and KLAS/8NewsNow, three local Arizona TV newsrooms with
direct authority on a local Arizona incident. Discover: 1 visitor
out of 2,801 all-time views.
West Marine 59-store closures (post 8083, withdrawn as Discover
evidence in 30B, still valid evidence here): heavy mainstream
pickup (MLive, The Sun, Patch, Myrtle Beach Online, MSN, CE
Outlook). Discover: 30 visitors out of 8,996, roughly three in a
thousand.
Ducati Pirelli 44 (post 10264, 13 September 2026, snapshot logged
in 30K item 7): at least nine other outlets ran it too, including
Ducati's own press release, autoevolution, visordown, rideapart,
boatnews.com, luxurylaunches and two newswire syndications.
Discover: 86 visitors, 28.8% of traffic at snapshot, the lowest
share among the four articles logged in 30K items 6 to 8, but
still meaningfully present, unlike the two cases above.
WHAT DIFFERS. Lake Havasu's and West Marine's competitors carry
more authority than PBN specifically for that story: local news
for a local Arizona event, national retail and jobs-press outlets
for a US retail closure. Ducati's competitors are general
automotive and motorcycle press. None of them has more marine or
powerboat authority than PBN in that niche, even though there are
more of them by raw count.
WORKING HYPOTHESIS, THREE CASES ONLY, TREAT AS PROVISIONAL (30D
still applies): raw competitor count does not predict Discover
suppression. Whether a HIGHER-AUTHORITY competitor in PBN's
specific niche already owns the story does. Needs more cases
before this becomes a rule, particularly a case where PBN faces
heavy competition from other marine or powerboat-specific
outlets, which none of the three cases above tests.
[email protected] is the Gmail inbox connected to Claude for
Powerboat News automation. Primary use is Google Alerts news
monitoring and other automated tasks. John also emails it
directly to test Claude's connectivity.
Claude can both send and receive email via this account (Gmail
connector).
If Claude checks the inbox and finds anything other than a
Google Alert, forward it to [email protected] rather than
acting on it directly.
KNOWN GAP: Gmail attachment download
The Gmail connector's get_message/get_thread tools return
attachment metadata (filename, MIME type, attachment ID) but
not the file bytes. There is no attachment-download tool
exposed. Claude cannot pull an email attachment out and save
or upload it directly.
Workaround: John downloads the attachment and uploads it in
chat (Image Upload plugin workflow, Section 22), or uploads it
to Drive/a public URL first, and Claude pulls it from there via
wp_upload_media_from_url.
Do not repeatedly search for a download tool, the gap is
confirmed, not a search failure.
SECTION 31 ADDITION
(added 4 September 2026)
Contributor photo drop-off notifications also arrive here.
See Section 22G for the full workflow: contributors email a
heads-up only, never the photo itself as an attachment, and
retrieval from Drive uses list_recent_files, not search_files.
**********
SECTION 32. CASE STUDY: SENSITIVE BREAKING-NEWS STORY WORKFLOW
(CHIP ROMER / LAKE HAVASU CRASH)
Added 14 September 2026, published as post 10267.
Worked example of Sections 6, 11, 18 and 22 operating together on
a real fatality story, kept here as a template for the next one
rather than scattered across those sections' abstract rules.
TRIGGER AND ROUTING
A fatal boat crash involving two named victims. This met the
Section 11 chat-review-first criteria on two grounds at once: it
involves a death, and at least six other outlets had already
published the same MCSO statement, so there was a real risk of
adding nothing beyond a rewrite. Full prose draft delivered in
chat first; WordPress only touched once John said "publish it"
(Section 11's publish override).
VERIFICATION STACK, LIVE EXAMPLE OF SECTION 6
MCSO's own Facebook statement was the primary source for the
crash facts (time, location, boat, victims, cause under
investigation) and stayed the only source for those facts
throughout, no outlet, John's contact, or AI tool ever added to
them.
Everything about who Chip Romer was, beyond the crash itself,
went through a lead-then-verify pipeline rather than being
written from a single source:
Grok's deep-dive research supplied a first, explicitly UNVERIFIED
lead: a Southern California offshore racer with the same
nickname, same age band, same lake.
That lead was only used once an independent primary source,
Mohave County's own statement, was found using the exact nickname
"Chip" for the crash victim, which is what actually confirmed
the identity.
Once identity was confirmed, each further biographical layer
(Skater/Douglas Marine racing history, SCCA sports car results,
Radical Cup North America season, the beverage company) was
separately verified against its own primary or trade source
(Speed on the Water, SCCA's own race report, the company's own
site, a local newspaper) before being added to the article, never
taken from Grok, Gemini or the Google AI Overview on their own
word.
A Facebook condolence post from a named, identifiable contact
(Alex Pratt) was treated as a corroborating lead worth following
up, not as publishable fact, consistent with Section 6's
treatment of social media and third-party AI claims alike.
A subsequent commenter's claim (that the victim was also a pilot)
was checked and could not be corroborated anywhere, it was left
out rather than softened into the piece.
ITERATIVE PUBLISHING
The article was published lean and then extended across several
hours as each new fact cleared verification, rather than held
back for a single complete draft. Each addition (SCCA record,
Radical Cup season and the Jeff Green Memorial Trophy, the
beverage company, two supplied portraits) was its own
wp_update_post call followed by a fresh wp_update_post_meta call
to advance dateModified in the schema, never left stale after a
content change.
IMAGES
The MCSO aerial photo was used only once John explicitly
confirmed the credit line himself ("Credit MCSO. Other outlets
have used it"), per the Section 22G caption-confirmation rule,
applied here even though the credit itself was never in doubt.
Two supplied 600px portraits were uploaded unaltered at John's
stated width and floated per the Section 14 float pattern
(heading, then figure, then wrapping text, clear:both at the
section end), with one deliberate width override: 600px is wider
than this component is normally used at, applied because John
asked for it explicitly, not a new default.
SOURCE QUALITY CONTROL
One AI-generated aggregator site (thatdaynews.online) was
identified mid-research by its generic, templated tone and an
unverifiable added figure, flagged to John rather than used, and
blocklisted outright once John confirmed it was a poor user
experience (Section 6, third-party AI and SEO advice, same
discipline extended to a low-quality human-run aggregator).
WHAT THIS CONFIRMS THAT WASN'T ALREADY OBVIOUS
The extra reporting work is what won the "Chip Romer" search
result over FOX 10 Phoenix, not speed to publish (Section 6.1's
underlying argument, now with a fatality-story example rather
than only corporate ones). Section 30K item 8 and Section 30L log
this story's Discover and competitive-crowding data separately,
this section covers editorial process only, not distribution.
**********
SECTION 33. ADSENSE/TRAFFIC SPIKE DIAGNOSIS AND GOOGLE NEWS
Added 15 September 2026, from a full walkthrough of the AdSense
export (6 April to 15 September) cross-checked against Search
Console's Search, Discover, Discover-with-AI-summaries and
Google News exports (14 June to 13 September).
------------------------------------------------------------
33A. FINDING WHAT DROVE AN ADSENSE SPIKE
------------------------------------------------------------
Method that worked: pull wp_list_posts for a 2-3 day window
around the spike date, then confirm against Search Console's
per-day Chart tab (Discover/Search/Google News) for the same
dates. The Chart tab is per-day; the Pages tab is a total across
the whole export range and only shows which page carries volume
somewhere in that range, not which day.
TIMEZONE TRAP. AdSense report dates run on the account's report
timezone (commonly Pacific, roughly 7-8 hours behind the UK). A
story published in the early hours UK time can appear on the
PREVIOUS calendar day in the AdSense export. Seen 14 September
2026: the Chip Romer story (post 10267) published 04:36 UK time
and its traffic split across the AdSense "13 September" and "14
September" rows, initially read as two separate spikes before
the timezone offset was identified as the likely explanation.
DO NOT GUESS A CAUSE FROM STORY THEME. Guessed the 15 June
AdSense spike (£96.21, 17,344 page views) was the Groupe
Beneteau industry story, matching the general pattern that
non-racing content travels further (30E). Wrong: Search
Console's Discover Pages data showed the AX300 piece (30A, 30B)
took 461 of 861 Google News clicks and was also the single
largest Discover page across the whole period, 23,405 clicks,
while Beneteau barely registered (139 impressions, Discover-
with-AI-summaries only). Always check Pages/Chart data before
naming a cause, even when a theme-based guess feels consistent
with existing evidence elsewhere in this document.
------------------------------------------------------------
33B. GOOGLE NEWS BEHAVES LIKE DISCOVER FOR PBN
------------------------------------------------------------
Google News performance (14 June to 13 September 2026 export):
monthly clicks 660 (Jun), 137 (Jul), 51 (Aug), 13 (Sep, partial
to the 13th). Impressions fell in near lockstep (11,268 to 502),
so CTR held steady around 5-6% throughout, ruling out an
engagement problem.
First read of this was "eligibility loss, needs investigating."
Wrong, or at least premature. Checked Manual Actions (clean),
news-sitemap.xml (processed successfully, current) and a live
Google search for three same-day articles, all indexed and
appearing in Top Stories/News tab within 9-10 hours of publish.
Pipeline is healthy.
The real explanation matches 30A's finding for Discover almost
exactly: Google News clicks are heavily concentrated in one or
two stories a month. The AX300 piece (461 clicks) plus the West
Marine 59-store piece (79 clicks, published 10 June) account for
540 of June's 660 clicks, 82% of the month. Remove them and
June's "ordinary" News performance (roughly 120 clicks) is not
far off July's 137 or August's 51. A month-on-month drop in
Google News totals is not itself evidence of a problem, same
principle as 30I, now confirmed for News as well as Discover.
Apply 30I's existing diagnostic order here too: Manual Actions,
then news-sitemap.xml status (Discovered pages around 7 is
normal, the spec only expects a rolling ~2-day window, not the
full archive), then a live Google search for a recent article,
then Google Publisher Center approval status specifically. The
last of these is separate from Search Console, will not show as
a Manual Action if it lapses, and has not yet been checked
against a real case.
------------------------------------------------------------
33C. DO NOT
------------------------------------------------------------
Do not treat a Google News or AdSense monthly decline as a
technical problem before running 30I's Manual Actions/sitemap/
live-search check. Same discipline as Discover, now confirmed on
direct evidence rather than assumed to carry over.
Do not read Search Console's Pages tab as day-specific. It is a
range total; use the Chart tab for anything date-sensitive.
