Editorial desk guide

The Editorial desk is where editors decide what readers of the Arcel Konnect app see. New stories arrive from publishers' news automatically; an editor reviews each one, shapes its title, summary, key facts, category and image, and approves or rejects it. This guide follows the staff app as it is today.

Getting to the desk

Open the staff app (/admin/ on the same address as these pages) and sign in with your work email: a one-time code arrives by email. Only staff members with the editor or admin role can sign in; editors land on Editorial. The Editorial area has four tabs: Editorial Workspace (the desk), Article Status (the history of decisions), Moderation (readers' comments and profiles) and Quotes (the quote of the day). The bar under the tabs shows when the sources were last refreshed ("Sources synced · …").

On a wide screen the desk shows the queue and the selected story side by side; on a phone or a narrow window the queue comes first and a story opens on its own page, with ← Back to the queue.

The queue

The Priority queue lists the stories waiting for a decision, ordered by urgency and impact; the header shows how many are pending review. Each story has a priority from 0 to 100, computed from its relevance to the industry, how recent it is, the publishers' credibility and how new the news is, and shown as a band:

Band	Priority
P0	80–100
P1	60–79
P2	40–59
P3	below 40

All shows every waiting story and High only the P0 band; while High is on, the queue says "Showing High priority only" and its count and pages are those of the P0 band. The queue is ordered by priority, so its first page is often the same either way. The filter and the page stay as you open stories, and are part of the page's address, so a reload or a shared link keeps them. Each row shows the title, the publishers (several when more than one publisher reported the same news), how long ago it arrived and its category. The queue is paged; choose a page number to move through it. A story leaves the queue once it is approved or rejected.

The story page

Choosing a story opens it. The Preview has three modes:

• Edit: the editing card (below).
• Consumer: the story as readers will see it in the app, with its key facts under a collapsible "Key facts" heading when it has any.
• Article: the publisher's text the story was built from, with its length ("Source text · N words") and Open the original article. When a source's licence does not allow showing its text, the desk says so and links the original instead.

Two labels tell you that the text is not the full article:

• Excerpt (publisher paywall) — open the original for the full article: the publisher's page showed only a teaser behind a paywall, so the story was built from that excerpt.
• Excerpt (publisher blocks automated access) — open the original for the full article: the publisher refuses automated access to its article pages, so the story was built from the description in its news feed.

In both cases, read the original before approving; the summary can only be as complete as the text it was written from.

A third label marks licensed content, not a partial text: Licensed feed — publisher. The publisher licensed its news feed to ARCEL (headline, short description and photo, with a link back), so the text is the publisher's own description, shown with its photo credit ("Photo: publisher"), and Read the full article on publisher opens the article on the publisher's site.

Editing

• Title: at most 80 characters (the counter shows how many are used).
• 60-word summary: between 55 and 62 words (the counter shows the count and the range).
• Save changes saves both. If someone else saved the story meanwhile, the desk keeps your text so you can copy it, and reloads the story.
• Key facts: up to five short facts readers see under the story. Add, reorder and remove them, then Save key facts. The card says whether an editor or AI wrote them.
• Category: one of the five pillars (Development, Design & Engineering, Construction, Operations, Practice Intelligence). Choose it and Save category. The card says whether the category is rule-based, an AI draft or chosen by an editor.
• Article image: Upload / replace takes a JPEG, PNG or WebP image of at least 1200 × 675 pixels and at most 10 MB, with its credit (who took or owns it, for example "Photo: Publisher Name").

AI actions

Where AI is switched on, the desk offers these actions; otherwise their buttons say "AI is off in this environment". AI writes only from the publisher's text of the story, and every result is a draft for you to check.

• Rewrite with AI (or Regenerate with AI) writes a new title and summary. While it works the button says "Writing with AI…" and cannot be pressed twice.
• Key facts with AI writes the key facts. It is offered only while no editor-written facts exist ("An editor wrote these key facts. Remove them to ask AI.").
• Suggest with AI proposes a category with its confidence ("AI suggests Construction · 82%", and any close alternatives); Use suggestion applies it.
• Generate with AI under the image draws a realistic news photograph from the title and summary, with no text, logos or watermarks; each image is checked for text and, if some shows, one more is drawn. It becomes the article image, credited "AI-generated image". If both images showed text, the desk says so: try again or upload an image.
• Restore original next to the title, summary, key facts, image and category puts that field back to its original value: the rule-based title, summary and key facts from the article (the summary and key facts are usually empty until AI or you write them), the article's rule-based category and the publisher's image. You confirm first; the field's AI flags are closed and the change is in the story's audit. The button is disabled ("Already the original") when the field has not changed.

When the AI service is busy the desk says "AI is busy right now. Try again in a minute or edit manually.", and when the environment's AI limit for the day or month is reached it says so; nothing is changed, and you can always edit by hand.

Checks and flags

Pre-publish checks show what approval requires:

Check	Kind	Passes when
Title	Blocking	the title has 1 to 80 characters
Summary words	Blocking	the summary has 55 to 62 words
Grounding	Warning (an amber "Grounding — review" chip; it does not stop approval)	the AI draft has no open high-severity flags
AI checks	Blocking while running (a grey "Checking AI text…" chip, not a failure)	the checks of AI-written text have finished (see below)
Image	Warning	the story has an image
Key facts	Warning	the story has at least one key fact

Flags list what automatic reviews noticed, from high to low severity: for example an unsupported claim (a figure or name not found in the publisher's text), a length problem, a classification to review, a limited source, or a possibly related story. A flag about one field (the title, the summary, the key facts, the image or the category) appears under that field; the checks list says which fields have flags, and keeps the flags about the whole story. A flag belongs to the text it checked: when that text changes (you edit it, or ask AI again), its flags are closed. If you have checked a flag yourself, for example the claim is in the article in other words, choose Dismiss under it and say why (at least 3 characters); the flag leaves the desk and your reason is kept in the story's audit.

Grounding is a warning. When AI flags something it could not find in the publisher's text, the Grounding chip turns amber, "Grounding — review", and the desk says "AI flagged possible unsupported facts — review before publishing". Read each flag next to its field and compare it with the article: edit the text, Dismiss the flag with your reason, or leave it if you are sure. Approve & publish stays available; the story's audit records that it was approved with open AI flags. These flags can occasionally be wrong (for example a claim the article makes in other words), which is why they no longer stop publishing.

Approving and rejecting

• Approve & publish publishes the story to readers. Add a reviewer note first if you want one kept with the decision. After a decision the desk opens the next story and says "Story approved." (or "Story rejected."); the message goes away when you choose another story.
• Reject asks for a reason, which is kept in the audit log and shown in Article Status.

Approval is refused while a blocking check fails: the desk says "Approval is blocked: …" with the checks to fix.

The AI-check hold. After an AI action, the AI-written text is checked in the background against the publisher's text (unsupported claims, an independent review). Until those checks finish, Approve & publish waits and the desk says "Checking AI text… AI checks are still running for this story — try again in a minute, or edit the text yourself." The Pre-publish checks show a grey "Checking AI text…" chip while they run: it is a short wait, not a failure. The desk reloads the story by itself; when the checks arrive the chip turns green, the flags appear, Approve & publish is enabled and the desk says "AI checks finished", without reloading the page. Text an editor saved is never held. If the checks have not finished after 30 minutes, approval is allowed and the AI checks line becomes a warning: review the text against the source yourself before approving.

Refresh feed

New stories arrive on their own as the platform reads the publishers' feeds during working hours. Refresh feed asks for a fetch now. The desk then shows its progress: "Fetching sources: 4 of 12…", then "Preparing stories: 5 articles left…", and finally "3 new stories." (or "No new stories."), and the queue reloads. If some sources could not be read, it adds "Some sources could not be read."

If the refresh takes longer than about three minutes, the button is free again and the desk says "Still running — new stories will appear automatically.": keep working, and the new stories join the queue as they arrive.

To be polite to publishers, a refresh does not ask a source again within a few minutes of its last fetch, and leaves alone a publisher that asked to slow down or refused access. When every source was fetched recently, the desk says "Sources were fetched in the last few minutes or are waiting for their publishers — nothing new to fetch."

Duplicates

The same news often arrives from several publishers, or twice from one. The platform removes repeats before they reach the queue:

• an article a publisher's feed sends again is ignored;
• an exact duplicate (the same address, or exactly the same text, as an article already collected) is not a new story;
• a near-duplicate (nearly the same text as an article collected in the last three days) is not a new story either;
• when the repeat comes from another publisher, it is merged into the existing story as another source: the queue shows several publishers on one story, and the story's priority takes the extra coverage into account.

Admins see these numbers on Sources (Content → Sources & Rights) under Duplicates: exact duplicates, near-duplicates and articles merged into stories, in total and for each source, over the period chosen at the top of the page.

Article Status

Article Status lists every reviewed story with its title, category, primary source, review time and decision. Filter by decision (All, Approved, Rejected), search by article, source or topic, and use Export audit log for a CSV of the decisions.

Moderation

Readers can comment on stories and report a comment or a profile. Moderation lists the comments with open reports, oldest report first, with how many reports each has and why; the views Held, Removed and All show the rest, and a reason filter and a search narrow the list. Open a comment to see it in its thread, its earlier versions, every report with its note and its author's history, then:

• Hold hides it from everyone but its author while you check something;
• Remove takes it down for breaking the community rules and, unless you untick Tell the author, tells its author;
• Approve makes a held or removed comment visible again, or keeps a reported one.

Each action asks for a reason (optional for an approval), which is kept in the audit log with your name. Profile reports lists reported profiles: dismiss a report, or make the profile private, clear its bio or remove its avatar.

Admins also see Block from commenting on a comment (1, 7, 30 or 365 days, or permanently; reading, liking and saving are not affected), Blocked readers to lift a block, and Screening terms, the words and phrases that stop a comment from being posted or hold it for review.

Quotes

Readers see one quote a day. Quotes lists them by date with their share images; Add quote takes the quote (up to 200 characters), who said it, an optional source link and the date, one quote per date. Saving draws the share image, which opens as a preview. A quote can be edited at any time and deleted only before its date; from its date the delete action says Already shown.