The studio runs in Arabic. Screen labels are shown below in Arabic, with an English gloss in brackets, so you can find them on screen. Body names, quotes and anything you type are shown as written.
التجارب المماثلة (Similar experiences) is the second step of a project. It answers the question every client asks: how did others do it? A ministry that split regulation from delivery, a regulator carved out of a ministry, a shared-services centre. A research agent searches the web for comparable bodies and writes a short brief for each one. You read the briefs, check them against their sources, choose which ones to keep, and the chosen briefs are what the AI consultants read when they design alternatives.
The step is optional. It never blocks Generate alternatives.
Who can do what
| Action | Permission |
|---|---|
| Open the step and read every brief | View (operating_model.view) |
| Add, edit, select, check, uncheck, remove and restore a brief; make a brief from a document | Act (operating_model.act) |
| Start, estimate or cancel a research run | Run AI actions (operating_model.ai.run) |
An archived project shows its briefs read-only.
How the step is laid out
Three side tabs, in this order:
- محلية (Local): bodies inside the Kingdom.
- عالمية (Global): bodies outside the Kingdom.
- الملخص (Summary): what the consultants will read.
The step's pinned action bar at the bottom of the screen links back to السابق: المدخلات (Previous: Inputs) and on to التالي: التوجهات والمعايير (Next: Directions & criteria).
Research
Each of محلية and عالمية has an optional instruction box, for example a body name or a topic you want covered, and a research button.
- Before it starts, the step shows an estimate (التقدير) of the time and number of searches, ≈ [min]–[max] دقيقة · حتى N عملية بحث (≈ [min]–[max] min · up to N searches), next to the ابحث (Search) button. Research is an AI action and counts toward the project's monthly AI usage limit.
- While it runs you see its progress (searching, reading sources, checking quotes) and you can cancel it.
- One research run at a time per project and section. Starting a second one for the same section is refused.
- A run adds at most 4 new briefs. Run it again, with a different instruction, to get more. A body you already have is not added twice, and neither is one you removed.
- Your own organisation is never a similar experience. The agent is told your organisation's name (the workspace name and the top unit of the current structure) and never writes a brief on it, on its subsidiaries or units, or on a reform your organisation itself was part of (for example the one that created it), even when that reform is described from another body's side.
- No brief is better than a weak one. The agent only writes a brief on a body that is genuinely comparable: a similar mandate or sector, or a reorganisation of functions like yours. If it finds none, the run ends with انتهى البحث ولم يجد تجارب مماثلة جديدة ذات صلة (The search finished and found no new relevant similar experiences) and adds nothing. Try a different instruction, or write a brief by hand.
- If the AI service has no search provider configured, the button is disabled and shows البحث على الويب غير مُفعَّل في خدمة الوكلاء (Web search is not enabled in the agent service). Writing briefs by hand and making briefs from your documents still work.
What leaves the system
- Search terms go to an external search service (Tavily). They name public bodies and topics only. They never contain text from your documents: the research agent is not given your documents at all. It receives the project name, your organisation's name (so it can leave it out; it is never put in a search term), the names of the units in scope, your function catalogue, your confirmed directions and your instruction.
- Pages are captured by our server, not by the search service. For every source a brief relies on, the server fetches the page itself, extracts its text, and stores it with the address and the retrieval date. The search service never supplies the text a quote is checked against.
- The server only fetches public web pages (HTML, PDF or Word). It refuses addresses that point inside its own network, follows only a few redirects, and gives up on a page that is too large or too slow. A page that cannot be read (for example a site that is all script) is listed on the brief as a capture issue; the brief stays, its claims from that page are marked unverified, and you can attach a saved PDF of the page instead.
- Every quote in a brief is checked against the captured page text, with the same word-for-word rule used for your own documents (after normalising the text, Arabic included). A claim whose quote is not on the page stays on the brief, flagged unverified.
The brief card
Each brief shows:
- the body, country and year;
- a state chip: مُراجَع من المصدر (Checked at source) or للاستئناس (For inspiration);
- a summary and a lesson; a quote cited inside the text shows as a small icon: hover it to see the quote and its source domain, click it to open the source in a new tab;
- the moves (التحركات) the body made: فصل (separated), دمج (merged), مركزة (centralised), لا مركزية (decentralised), نقل إلى جهة أخرى (transferred out), وحدة جديدة (new unit), أخرى (other), each with the functions of your catalogue it relates to, shown as chips;
- the sources, each with its domain and retrieval date, and any attachments;
- your notes, the search terms the run used for it (collapsed) and any capture issues.
There is no relevance score and no ranking. Briefs are not sorted by how good they are; that judgement is yours.
Card actions: select or deselect, check at source, edit, remove (soft; restorable), and add a brief by hand.
Adding a brief by hand
Write the body, country, year, summary, lesson and moves (picking functions from your catalogue), add notes and attach files. A claim you write has no quote and stays marked as written by you. A brief with no captured source and no attachment cannot be checked, so it always stays inspiration.
Making a brief from a document
Documents you uploaded in Inputs and that the system recognised as describing a similar experience stay in Inputs and stay citable as before, because your client brought them. They are also listed in the Summary tab. To turn one into a brief, press إنشاء ملخص تجربة من هذا المستند (Create an experience brief from this document), choose Local or Global, and give the body and country. Only this action turns an upload into a brief; the system never does it for you. The brief starts as checked, by you, since a person chose it. Its moves start empty. Removing the brief returns the document to Inputs.
A document attached to a brief is read for evidence only. The system does not propose functions, pillars or bodies from it, so another entity's structure never ends up in your own catalogue.
Check at source
A brief can be checked when it has at least one captured source or an attachment that has finished reading. Open the check dialog: it lists each source (title, a link that opens the live page in a new tab, retrieval date) and each claim with its quote, with a toggle to show the quote in the saved text. Claims that could not be verified are marked. When you have compared them with the source, confirm قارنتُ هذه المعلومات بالمصدر (I compared this information with the source). The brief then shows who checked it and when, and its chip changes to مُراجَع من المصدر.
- Editing the summary, lesson or moves of a checked brief unchecks it, because what you checked has changed. Editing the body, country, year or notes does not. You can also uncheck by hand.
- Checking is a human act. Nothing in the system checks a brief for you, and citing a brief's text by hand does not check it either.
What the consultants read
The الملخص (Summary) tab is titled ما سيقرؤه المستشارون (What the consultants will read). It lists the selected briefs grouped by state, with counts such as "3 checked, 2 inspiration", and carries the note ملخصات الاستئناس لا يُستشهد بها (Inspiration briefs cannot be cited). When no brief is selected, the tab offers a button, لا توجد تجارب مماثلة (No similar experiences). Pressing it completes the step without briefs. The mark is saved on the project, so the whole team sees it, and selecting a brief clears it. إلغاء العلامة (Remove the mark) undoes it by hand.
Briefs are selected by default. A deselected or removed brief reaches no one.
| Checked at source | For inspiration | |
|---|---|---|
| Consultants read its summary, lesson and moves | Yes | Yes |
| A consultant may point a box at it as a precedent | Yes | Yes |
| Its captured page text can be cited as a quote | Yes | No |
| Counts toward the "sourced" rate | Only through a quote from its page | Never |
A precedent never stands in for your own sources: a legal basis still has to come from your documents.
In generated alternatives, a claim that rests on a brief carries the chip سابقة خارجية — غير مُراجَعة (External precedent, not checked) or سابقة خارجية — مُراجَعة (External precedent, checked). The grounding line counts them separately, as «92% موثّق · 2 من سوابق خارجية · 3 كتبتَها أنت» (92% sourced · 2 from external precedents · 3 written by you). See Generate alternatives.
Selecting, checking, editing or removing a brief changes the project's inputs, so existing alternative cards show «تغيّرت المدخلات بعد إنشائه» (Inputs changed since it was created).
Records
Approving an alternative freezes the briefs selected at approval in the decision record, with the moves cited as precedents. The decision PDF carries a short annex, التجارب المماثلة (Similar experiences), listing each brief with its state, URLs, retrieval dates and attached documents. Decisions made before this step existed render as before.
Every research run, check and edit is recorded in the Audit log. AI actions are recorded under your name and marked as AI.