Cross-client system review · July 2026

Inside CiteDelta’s PR alert network.

A fleet-wide audit of 13 recurring client digests and 39 recent runs, showing how each alert is scheduled, researched, verified, filtered and delivered across travel, pets, outdoors, automotive, hair, ecommerce, sexual health and electronics.

Open the full reviewAudit window: 10 to 16 July 2026

13active digest jobs

39recent runs reviewed

99links in delivered reports

20intentional silent outcomes

Review scope

Across the whole client fleet.

This is no longer a Lordhair run report. It compares the latest three runs for every comparable active client trend or opportunity digest.

Fleet inventory
Shared architecture
Scheduling model
Client-specific briefs
Research tools
Verification logic
Freshness rules
Silence and delivery gates
Output contracts
Observed recent behaviour
System-wide gaps
Recommended standard

Thirteen recurring editorial agents.

Each job has its own brand territory, cadence, delivery channel and tool allowance. The shared purpose is to turn current information into usable digital PR opportunities.

DigestPrimary territoryCadenceLast 3

Daily PR briefingGeneral cross-sector PRWeekdays3 reports

WaggelUK pets and veterinary newsWeekdays3 silent

My BaggageUS and UK travel, luggage, relocationWeekdays3 reports

OutforiaOutdoors, parks, wildlife, safetyWeekdays2 reports · 1 silent

OvokoDrivers, repairs, used parts, motoringWeekdays3 silent

StasherTravel disruption and luggage storageDaily3 reports

LordhairHair loss, systems and male groomingWeekdays3 reports

BoatBookerBoat travel and lifestyleMon · Wed · Fri1 report · 2 silent

Boxie24Decluttering, cleaning and storageWeekdays3 reports

SleekFlowAI commerce, travel and beautyWeekdays3 silent

Say It With A CondomUS and UK sexual healthMon · Wed · Fri1 report · 2 silent

WellPCBConsumer electronics and PCB anglesWeekdays3 silent

OurPCBPCB and electronics manufacturingWeekdays3 silent

Audit basis

The review uses the three most recent cron sessions for each job, 39 runs in total, between 10 July at 08:00 and 16 July at 08:08 Philippine time.

Every digest follows the same seven-stage shape.

These are agent-driven scheduled jobs. They are not one central crawler distributing pre-scored alerts.

Trigger

Cron starts a fresh session at the client’s scheduled time.

Brief

The agent receives the brand, market, topics and editorial rules.

Discover

Web queries, trends and sometimes browser or terminal tools gather candidates.

Verify

The agent tries to confirm dates, claims and source pages within the run.

Filter

Recency, client fit, news value and pitchability narrow the list.

Gate

The output is either a digest, a watchlist where allowed, or silence.

Deliver

The scheduler posts the final response into the designated Slack channel.

Fresh-session consequence

Each run begins without the previous run’s conversational context. Cross-run memory only happens where the prompt explicitly calls session history or another persistent source.

Most of the fleet wakes up at 08:00.

08:00

Ten jobs share this minute on at least some days.

07 jobsWeekdays at 08:00

02 jobsMon, Wed, Fri at 08:00

01 jobDaily at 08:00

03 jobsStaggered at 08:02, 08:04 and 08:08

Ten jobs are scheduled exactly at 08:00 on shared operating days.
Waggel, My Baggage and Outforia are staggered by two, four and eight minutes.
Stasher runs seven days a week, unlike the other client jobs.
BoatBooker and Say It With A Condom run Monday, Wednesday and Friday.
Most weekday prompts roll useful weekend news into Monday instead of posting at weekends.

Operational concentration

The 08:00 cluster creates avoidable concurrency. Staggering every job would reduce simultaneous browser and search pressure and make failures easier to isolate.

The architecture is shared. The editorial contract is not.

Each prompt defines a different opportunity surface, source hierarchy, silence threshold and format.

Travel cluster

My Baggage, Stasher, BoatBooker

Travel disruption, visas, baggage, moving abroad, airports, consumer rights, sports equipment and boating. My Baggage has the most explicit browser fallback. BoatBooker requires exactly five verified items.

Lifestyle cluster

Waggel, Outforia, Lordhair, Boxie24

Pets, outdoors, grooming, cleaning and storage. The prompts turn seasonal problems, data and consumer behaviour into advice-led commentary.

Technology cluster

SleekFlow, WellPCB, OurPCB

AI commerce, consumer electronics, repairability, recalls, PCB production and manufacturing. These prompts use strict silence gates when evidence is sparse.

Specialist cluster

Ovoko, Say It With A Condom

Drivers and sexual health. Both emphasise fresh, source-grounded material. The sexual-health digest uniquely checks recent session history for duplicate angles.

General briefing

Daily digital PR trendspotting

A cross-sector discovery layer selecting five broadly interesting opportunities, with a Monday “This week” planning component.

Common prompt ingredients

1,019 calls across 39 runs.

Tool access varies by job, but actual sessions show a wide research mix.

493Web searches

269Browser operations

138Web page extractions

71JournoFinder trend searches

35Terminal calls

8File reads and searches

4Session-history searches

1JournoFinder client-list call

Configured tool allowances

7 jobsWeb only

3 jobsWeb and browser

1 jobWeb, browser and file

1 jobWeb and terminal

1 jobWeb and session history

What this means

Two clients can receive superficially similar digests while using materially different evidence paths. Browser-enabled jobs can open difficult pages. Web-only jobs depend more heavily on search and extraction availability.

Prompts request verification, but enforcement differs.

All 13 jobs require source links. Ten explicitly use verification, credible-source or source-check language. The strictest prompts define what to do when a source cannot be confirmed.

Discovery

Search results and trend feeds identify possible stories.

Retrieval

Web extraction, browser navigation or terminal-based fetching attempts to open the source.

Claim check

The agent checks dates, figures, geography and whether the story is materially new.

Editorial check

The story must create a credible route for the specific client, not merely mention the wider sector.

Stronger prompt examples

My Baggage forbids inclusion unless the claim, date and link are verified in that run.
Ovoko requires reliable source links and rejects search headlines alone.
WellPCB and OurPCB require multiple verified, timely items before posting.
Say It With A Condom checks dates and recent session history.

Remaining weakness

Verification is written in natural language rather than enforced by a shared schema.
A job can finish “ok” even when it returns silence or a weakly verified digest.
Source links are not centrally validated after generation.
There is no universal fact-versus-angle labelling rule.

Source mix observed in delivered reports

Recent outputs linked to official bodies and regulators alongside mainstream and trade media. Frequently used domains included BBC, GOV.UK, The Guardian, ITV, the Met Office, the EU Council, TSA, Skift and Condé Nast Traveler. Seven final links still used Google News rather than resolved original sources.

Freshness ranges from 24 hours to seven days.

24 hours

Lordhair

Tuesday to Friday priority, with weekend catch-up on Monday.

24 to 48 hours

My Baggage

Verified weekend material allowed on Mondays.

24 to 72 hours

Stasher, BoatBooker, SleekFlow, Boxie24, WellPCB, OurPCB

Recent context can fill gaps when the newest cycle is thin.

Up to 7 days

Say It With A Condom

Older than 14 days is barred unless an upcoming event has a new reason to pitch.

24 hours, then 2 to 3 days

Ovoko

Older recent items must be labelled.

Current and useful

General, Waggel, Outforia

Freshness is described more qualitatively.

No single fleet-wide freshness definition

This is reasonable for different sectors, but the page date, publication date and “why now” should be captured consistently for every chosen item.

Silence is a core quality-control mechanism.

20silent runs

19reporting runs

More than half of the audited sessions returned [SILENT]. This intentionally prevents weak, stale or unverifiable content from reaching Slack.

All three recent runs reported

Daily PR briefing
My Baggage
Stasher
Lordhair
Boxie24

All three recent runs stayed silent

Waggel
Ovoko
SleekFlow
WellPCB
OurPCB

Mixed outcomes

Outforia, 2 reports
BoatBooker, 1 report
Say It With A Condom, 1 report

371 calls

were made by the 20 silent runs, averaging 18.6 calls each.

648 calls

were made by the 19 reporting runs, averaging 34.1 calls each.

Important distinction

A silent result is not necessarily a failed run. It can indicate that the quality gate worked correctly. Scheduler status and editorial outcome need separate reporting.

The final format is similar, but thresholds vary.

01Current headline or hook

02Short factual summary

03Why it matters for the client

04PR, commentary or data angle

05Visible source link

06Optional market or timing label

Different inclusion thresholds

Exactly 5BoatBooker posts only when five usable items are verified.

At least 2WellPCB and OurPCB require at least two strong stories.

3 to 5Lordhair requests three to five and allows a watchlist.

Up to 5Stasher and Boxie24 cap the normal digest at five.

FlexibleOther prompts favour fewer strong items rather than padding.

Recent delivered-output profile

The 19 reports contained 99 visible links, an average of 5.2 links per delivered report. Response length ranged up to 4,210 characters.

What the latest 39 sessions reveal.

Strong

Quality gates are active

Twenty runs suppressed delivery instead of padding channels with generic stories.

Strong

Research is substantial

The sessions made 1,019 calls across search, extraction, browser, trends and supporting tools.

Strong

Links are normally visible

Delivered reports averaged 5.2 source links, supporting quick editorial review.

Watch

Exact sources repeated

Recent My Baggage, Outforia and Stasher reports repeated at least one exact URL across runs.

Watch

Original sources are not universal

Seven final links used Google News rather than the original publisher or primary source.

Fix

Internal process text leaked

One BoatBooker and one sexual-health report began with “I now have enough...” instead of starting cleanly with the digest.

Fix

Deduplication is inconsistent

Only the sexual-health prompt explicitly searches recent digest history. Other jobs mostly rely on the fresh model’s judgement.

Fix

Tool access is uneven

Seven jobs are web-only while others can browse, use terminal retrieval or check session history.

Repeated exact links found

My Baggage repeated an Australia visa-price article from The Guardian.
Outforia repeated a Met Office heatwave update.
Stasher repeated EU passenger-rights sources from the EU Council and DW.

A repeated source is not automatically wrong if the story materially develops. The current system does not centrally record or explain that distinction.

The fleet works, but it has grown through prompt-by-prompt variation.

No shared story ledger

There is no central first-seen date, story fingerprint, prior client use, verdict or material-update field.

No uniform evidence schema

Source verification is an instruction, not a required machine-checked set of fields.

Uneven silence semantics

Some prompts demand exact [SILENT], one asks for a completely blank response, and Lordhair can produce a watchlist.

Prompt drift

Freshness, item thresholds, source language and output structure differ more than client needs alone explain.

Scheduler concentration

Ten jobs can start at 08:00, increasing concurrent dependency load.

No fleet-level quality telemetry

Run status records execution success, not source quality, duplication, confidence or editorial usefulness.

Source resolution gaps

Google News and secondary pages can remain in final outputs when original URLs are not resolved.

Generated angles are not labelled

Confirmed facts and model-created PR recommendations are visually blended in many digests.

A common core with client-specific modules.

Keep each client’s editorial territory, but standardise the evidence, deduplication, output and monitoring layers.

01 · Shared evidence record

Require headline, canonical URL, publisher, publication date, update date, geography, source type, extracted claim and verification timestamp.

02 · Central story ledger

Store a content fingerprint, first-seen date, clients used for, previous angle, verdict and reason for reuse.

03 · Source hierarchy

Primary or official source first, original publisher second, reputable secondary reporting third. Resolve Google News before delivery.

04 · Explicit scores

Record recency, authority, client fit, novelty, journalist appeal and confidence separately. Use the score to support, not replace, editorial judgement.

05 · Unified gate

Return one machine-readable outcome: SEND, WATCHLIST or SILENT, with clear per-client thresholds.

06 · Clean rendering

Generate the Slack message from structured fields so internal reasoning or “I now have enough” process text cannot leak.

07 · Fleet telemetry

Track runs, send rate, silent rate, average research calls, source mix, unresolved links, repeated stories and human usefulness feedback.

08 · Staggered execution

Spread starts across the hour and give browser-heavy jobs their own slots.

Bottom line

The fleet is best understood as thirteen scheduled AI editorial researchers sharing a broad operating pattern. It is already effective at suppressing weak news and producing client-specific ideas, but it is not yet one centrally governed alerting platform. The next improvement is standardisation, not more prompts.