Generate alternatives

For HR administrators

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. Your documents can be in English; quotes keep the source wording.

التوليد (Generate) is the step where a team of AI consultants designs alternative operating models for your project. You choose who is on the team. Each consultant reads the same inputs and the same locked criteria & weights, applies its own lens, and returns one complete alternative. You watch the consultants work on live desks, then get one card per alternative in the next step, Alternatives. You can ask a critic to score any card against the criteria you locked.

Who can do what

ActionPermission
Open the step, read the roster, the desks, the list of alternatives and older generationsView (operating_model.view)
Create, customize, reset and remove consultants in the workspace library; choose who is on the project's team; write project notesAct (operating_model.act)
ابدأ التوليد (Start generating), إعادة توليد الكل (Regenerate all), أعد التشغيل (Run again), قيّم (Score) and ألغِ التشغيل (Cancel run)Run AI actions (operating_model.ai.run)

Unlike other AI tasks in the studio, cancelling a generation needs Run AI actions, not Act. Without a permission the button stays disabled, and its tooltip says which permission is missing. See Overview.

An archived project shows its alternatives read-only.

Before you start

Every generation is built on one locked version of the criteria & weights, so you lock them first in Directions & criteria. Until you do, the readiness list says "Lock the criteria and weights in the Directions step first", and Start stays disabled.

Once the frame is locked, the step header (next to Project settings) shows two round buttons, مبني على (Built from, with a badge counting the documents) and ما يقرؤه كل مستشار (What every consultant reads, with a green dot when the criteria are locked); each opens its details in a dialog.

The step is laid out as three stages, with a small progress bar above them: الفريق (Team), الجاهزية (Readiness) and التوليد والمراجعة (Run & review). Only one stage is open at a time. A folded stage is a single line with its summary: who is on the team, whether everything is ready (and the time estimate), and which generation is shown with its number of alternatives. Click a stage, or its stop on the progress bar, to open it. The step opens on the first stage that still needs something: Team when no consultant is selected, Readiness when an item blocks the start, otherwise Run & review. Starting a generation brings Run & review forward. Start generating and Regenerate all stay in the step's action bar, pinned at the bottom of the screen.

The consultants

The roster asks من توظّف من المستشارين؟ (Which consultants do you want to hire?) All consultants read the same inputs and locked directions and apply the same rules; the lens is what differs.

House consultants

Each workspace starts with five house consultants for each entity kind, marked مستشار أساسي (House consultant). Each one starts from an approach: a method for grouping functions, with its own trade-offs and the criteria it tends to favour.

MinistryLens
Governance StewardBuilds the entity around the governance functions its strategy names
Systems ThinkerOrganises by the strategy's pillars
Lean OperatorThe smallest viable entity: regulate and monitor
RegionalistStrengthens the regional arm as the oversight layer
Transition ArchitectKeeps temporary units until every transfer completes
CompanyLens
Customer ChampionOrganises the company around the customer segments it serves
Business-Line BuilderGives each business line one unit accountable for its results
Lean OperatorThe smallest viable company: shared services and few layers
Market LeadOrganises the company by market and geography
Platform ArchitectBuilds shared platforms the business lines run on

The company team is a house proposal that an expert will confirm before the first company client.

Jadiroon tunes the approaches over time. Every alternative records the approach version it used.

The workspace library

Consultants are saved to your workspace library, not to one project. Every project of the same entity kind in your workspace sees the same library.

  • تخصيص (Customize) opens the consultant's dialog: اسم المستشار (Consultant name), العدسة: على ماذا يركّز عند التصميم؟ (Lens: what does it optimise for?) and Default instruction ("Every project using this consultant reads it."). The dialog notes "Saved to the workspace library; every project in it sees the change." A customized house consultant shows مستشار أساسي · معدّل (House consultant · Edited). A house consultant's approach is fixed.
  • Reset (house consultants only) puts the house name, lens and instruction back.
  • أنشئ مستشارًا جديدًا (Create a new consultant) opens the same dialog. Write a name and a lens, optionally a default instruction, and choose under يبدأ من نهج (Starts from the approach of) which approach it builds on; your text is layered on top. أضف واختره (Add and select) saves it to the library and selects it for this project. Its card shows "Created by [name]".
  • إزالة من المكتبة (Remove from library) archives a consultant after a confirmation: "Other projects in this workspace will no longer see it. Alternatives it already built stay as they are."

You write the name, the lens and the instruction in any language. They are shown as written, in their own reading direction.

The project's team

Each card on the roster has a checkbox; selected consultants take part in the next generation. In a new project the house consultants are selected. A consultant created in another project appears here unselected.

The ملاحظة لهذا المشروع (Note for this project) in the dialog belongs to this project only ("Only this project."). Use it for what a consultant should know here but not elsewhere, such as "Focus on the northern-region branches." A ملاحظة المشروع (Project note) mark on the card shows that a note exists.

What a consultant reads

Every consultant reads the same project material:

  • the locked frame: the directions that are on, the design rules, and the criteria with their weights;
  • the confirmed functions of the catalogue, where each one sits today, and the matrix axes;
  • the current state, and the design scope worked out from your applied scope;
  • when the current state came from the Organizational Structure, the activity allocation (see Inputs): the FTE each function carries today, filled and vacant seats per unit, coverage and findings. Consultants are told to consolidate a function spread across three or more units when the directions allow, and to keep a function's FTE with it when they move it. A current state from Excel has no allocation, and consultants work as before;
  • the confirmed decision rights held by units in the design scope, and your confirmed committees (see Decision rights); no person names are included;
  • the ecosystem: the bodies, the "Do not take back" list and the functions another body holds. Bodies found by web research that nobody has checked at source are given as background only: consultants never cite them and never hand work to them (a handover to one is the rule breach handover_unchecked_body, sent back for repair like any other);
  • the text of every document in the project that has not been removed (latest revision, documents of authority first, up to 8 MB). Text captured from the web for a similar experience is included only when that brief is selected and checked at source;
  • the briefs you selected in Similar experiences, checked or not, as precedents: a consultant may borrow an idea and point a box at one. Only a checked brief's saved page text can be cited as a quote.

On top of that, each consultant reads its own guidance, highest priority first:

  1. the ملاحظة لهذا التشغيل (Note for this run), if you gave one with أعد التشغيل (Run again) or إعادة توليد الكل (Regenerate all);
  2. the note for this project;
  3. the consultant's default instruction;
  4. its name and lens;
  5. its approach.

Higher items win where they conflict. Nothing overrides the design rules or the rule that every claim must cite your evidence. If a note asks for something that breaks a rule, the consultant follows the rule and says so in its assumptions.

What a consultant designs

A consultant designs the head of the entity and at most two levels below it. Units deeper than that are carried whole under the unit the consultant chooses. Every function in scope is placed in exactly one unit, or handed over to an ecosystem body. A function another body holds by statute is never brought back. Names and claims are written in Arabic only; quotes keep the wording and language of their source.

When your scope covers only part of the entity, the design starts from the lowest unit that holds all of the scope:

  • top units with nothing in scope are locked: they appear unchanged in every alternative ("Out of scope · unchanged") and count toward span;
  • a top unit with a scope unit deeper inside it is redesigned as a whole. Before you start, the readiness list warns "Design widened to [unit]".

Start a generation

The readiness card shows the readiness list (الجاهزية للتوليد (Ready to generate)) and the estimate. The start button sits in the step's pinned action bar at the bottom of the screen: ابدأ التوليد (N) (Start generating (N)), where N is the number of selected consultants. Once a generation exists, إعادة توليد الكل (Regenerate all) there starts the next one. Each item in the list either يمنع البدء (Blocks the start) or is a تنبيه (Heads-up):

ItemLevel
Lock the criteria and weights in the Directions step firstBlocks the start
Select at least one consultantBlocks the start
Add the current structure before generatingBlocks the start
No confirmed function is inside the design scopeBlocks the start
لا يتسع ما بقي من حد استخدام الذكاء الاصطناعي لهذا المشروع هذا الشهر لتوليد جديد (What is left of this project's AI usage limit this month cannot fit a new generation)Blocks the start
No scope check has been applied yetHeads-up
N documents are still being read; consultants won't see what isn't read yetHeads-up
Inputs changed after the criteria were lockedHeads-up
Design widened to [unit]Heads-up

When nothing is listed, it says "Everything is ready to generate."

Next to the button, the action bar's status line sums up the readiness: a generation in progress, the blocking items, or that everything is ready. It also holds السابق: التوجهات والمعايير (Previous: Directions & criteria) and, once alternatives exist, التالي: البدائل (N) (Next: Alternatives (N)).

The estimate is a time range, in the form ≈ [min]–[max] دقيقة (≈ [min]–[max] min). It is based on the number of consultants and the size of your evidence. Up to three consultants work at the same time, so five take about two rounds.

If the AI service is offline, the readiness card says "The AI service is offline, so there is no estimate right now.", Start is disabled with "AI service is offline; everything else works.", and the rest of the step keeps working.

The desks

While a generation runs, the step shows المستشارون يعملون (Consultants at work), with a desk for each consultant. The lines on a desk come from what the consultant actually does: the documents it opens, what it searches for, the functions in scope, the boundary, the grouping, its own checks, repairs and redesigns. They are never a canned animation. A consultant waiting for a free desk says "Waiting for a free desk (3 consultants work at once)".

Each desk shows a status (بانتظار البدء (Waiting), يعمل الآن (Working), يُصلح المسودة (Repairing), يعيد التصميم (Redesigning), أنهى العمل (Finished) or تعذّر (Failed)). When a consultant finishes, its desk lists the names of the draft's top units, marked "Awaiting the diversity check" until the whole run ends.

ألغِ التشغيل (Cancel run) stops the whole generation, after a confirmation: "Consultants that haven't finished stop. Finished drafts are kept and get their cards." The generation is recorded as cancelled. You cannot cancel one desk on its own. If every desk has finished but the cards never appear (the results couldn't be prepared), ألغِ التشغيل (Cancel run) stays available as the way out: it tries once more to prepare the cards, and whatever still fails is closed as a failed card with the reason, so you can start again. If the AI service is offline at that moment, Cancel says so and closes nothing, so no finished draft is lost; try again once the service is back.

One consultant failing, or reaching its usage limit, never stops the others.

What happens to each draft

  1. Draft. The consultant reads, designs and submits. It can run the rule check on its own draft before submitting.
  2. Check. Code checks the draft against the rules and checks every quote against your documents.
  3. Repair. If a rule is broken, the draft goes back to the same consultant with the errors, up to two rounds. Failed quotes never trigger a repair: they are counted.
  4. Diversity check. When every consultant is done, each draft is compared with the others. When two are 80% alike or more, the one that finished later goes back once for a different grouping.

Decision rights

When you have confirmed decision rights in Inputs, every alternative has one more thing to settle. A decision right is held by a unit (or a committee). If an alternative merges, splits, hands over or removes a unit that holds one, the right would be left with no holder, so the alternative must say where it goes.

  • Rights of kept units carry over. A unit the alternative keeps, including one it only moves or renames, keeps its rights untouched, and nothing is asked of the consultant. A right held by a committee never needs reassigning.
  • Each reassignment names a new holder and a reason. The new holder is a unit of the draft or one of your committees, and the reason is one sentence ("The rights move with the functions to the support agency"). Consultants read your confirmed rights as they design and do not invent new decisions.
  • Two rules check it, like the other design rules. decision_orphaned: a right whose holder disappears and has no valid new holder. approver_is_executor: after the reassignments, the same unit or committee approves and executes the same decision, and that clash is new compared with today (one your current state already has is left to the المعتمد هو المنفذ (The approver is the executor) finding of the inputs). This also applies when two units merge into one box and one of them approved a decision the other executed. Both are violations: the draft goes back to the same consultant for repair, up to two rounds, and a draft that still breaks one keeps a badge that names it. They appear in القواعد والفحص (Rules & checks) under صلاحيات القرار (Decision rights).

A generation stores the confirmed rights as they were when it started, and an alternative is always checked against the rights that existed when it was generated. Rights you confirm later apply to new generations, not to alternatives already built; confirming them counts as a change to the inputs, so an earlier generation shows "Inputs changed since" and Run again asks for Regenerate all instead. An alternative generated before any rights were confirmed has no decision-right rules at all, and when no confirmed right is held by an in-scope unit, nothing in the generation changes.

The reassignments of each alternative are in its صلاحيات القرار (Decision rights) tab, and you can change them there by hand. See Alternatives.

Reading a card

When the generation ends, the Generate step lists its alternatives briefly (letter, title, consultant, rule badge and التفاصيل (Details)), with عرض البدائل ← (See the alternatives). The full list and pane, with Score, Score all, Run again, Edit and Compare, are in the Alternatives step. Each alternative's card here shows its letter and consultant, a title, a mini chart of the head and the top units, and the lines below.

"Passes the rules"

The badge covers only what code checked ("Checked by code: …"): span, layers, boundary (nothing another body holds is brought back), no orphan functions (every function in scope is placed or handed over), locked units left unchanged, separation, grouping basis stated for every top unit, unit kinds nested in order, and internal audit and cybersecurity placed independently (for functions with a control role), that the units a statute mandates are present, that the staff moved stays under your cap when one is set (see Mandatory units and Cap on staff moved), and, when you have confirmed decision rights, that every right of a changed unit has a new holder and that nobody approves a decision they execute (see Decision rights). Design quality and control-placement warnings (duplicate units, thin units, scattered support functions, functions without a mandate document, mode mismatch, a regulator owning an operator, risk / compliance, procurement / treasury and data protection placement) are listed with the checks but never block the badge. A draft that still breaks a rule after two repairs is kept, and its badge names the rules it breaks, with the measured values where they apply, for example "Breaks: span 8 > 6".

With separation on, no top unit may hold both a regulating unit and a delivering unit in its branch. Regulating and delivering deputyships may still sit side by side under the head.

Code checks the modes the consultant declared. Whether the grouping is good, and whether the units really fit their declared mode, is the critic's call when you score. The القواعد والفحص (Rules & checks) tab of the alternative's pane in the Alternatives step says so.

Mandatory units

The Saudi reference lists units that a statute or regulation requires of a government body, each with the instrument and a quoted source (for example internal audit under Council of Ministers Resolution 129, 1428H). The reference is researched on the platform and confirmed by an expert. The rule checks every confirmed mandate of the reference revision your project uses, whether or not one of your functions is linked to it. When that reference has no mandates, the rule says nothing.

The check asks whether the draft keeps each mandated unit. A mandate counts as present when a unit of the draft (placed by the design, or carried over or locked as it is today) holds a function linked to it through the reference catalogue check, or a function of your catalogue whose name contains every distinguishing word of the mandated name, or a function with the same control role as the mandate (for example internal audit), or when the draft adds a new function whose name does. A function handed over to another body does not count. Names are compared without the generic words (administration, unit, general…) and the article, and a word joined with «و» counts as the word itself, so «الإدارة العامة للأمن السيبراني» and «إدارة الأمن السيبراني وحماية البيانات» both cover «الأمن السيبراني». When the consultant repairs a missing unit, it first places one of your functions that covers it, and adds a new function only when none does. A missing unit raises الوحدة الإلزامية «name» غير موجودة في المسودة (statute):

Entity kind and scopeSeverity
Ministry, whole entity in scopeError: the draft is repaired, and the badge names the rule if it still breaks it
Ministry, partial scope, and a function linked to the mandate is in scopeError
Ministry, partial scope, and no linked function in scopeWarning
CompanyWarning

A warning is listed with the checks and never blocks the badge. The rule appears as الوحدات الإلزامية نظاماً (Units mandated by statute) in the badge and the check list.

  • Switch it off with a reason. It has its own switch in Directions & criteria, like the other rules.
  • Older alternatives are judged by their own generation. Each generation stores the mandates it ran against, and every later check of its alternatives (live check, save, copy, the assistant) uses that list, not today's reference. Confirming a new mandate therefore does not change alternatives already generated; Regenerate all brings it in.

Cap on staff moved

When the frame sets a الحد الأعلى للقوى العاملة المنقولة (Maximum staff moved), a draft that moves a larger share of today's staff than the cap raises an error, for example نسبة القوى العاملة المنقولة 40% تتجاوز الحد 30%. The share is measured in FTE from the activity allocation, over the staff in scope. A function's FTE goes wherever the function goes: to the unit that receives it, or with the current unit that holds it when that unit is carried over. Its people in other units move with it, the same way the workforce plan counts them.

  • The rule acts only when allocation coverage is at least 60%. With less of the staff time tied to functions, the moved share is not reliable, so the rule stays silent rather than guess.
  • It needs a cap. A project with no cap, or no seat snapshot (a current state uploaded from Excel), is not checked.
  • It appears as its own group, الحد الأعلى للقوى العاملة المنقولة, in the check list; Switch off works through Directions by clearing the cap.

"Sourced"

«92% موثّق · 2 غير متحقق منها» (92% sourced · 2 unverified) is the share of claims that are backed: either by at least one quote found word for word in your documents (after normalising the text, Arabic included), or, for a claim that quotes nothing, by a reference to one of your directions, rules or ecosystem bodies. The tooltip splits the two and counts the quotes that weren't found. A claim that quoted your documents and had every quote fail is غير متحقق منه (unverified) even if it also carries a reference, so a reference can't stand in for a quote that doesn't exist. A claim with neither is غير متحقق منه (unverified) too: it stays in the draft, flagged, and counts against the rate.

When a claim rests on a similar experience, the line reads like «92% موثّق · 2 من سوابق خارجية · 3 كتبتَها أنت» (92% sourced · 2 from external precedents · 3 written by you). Precedents are counted separately and never raise the sourced rate, whether the brief is checked or only inspiration. A claim that quotes a checked brief's captured page counts as sourced like any other quote. A claim that points at a brief shows سابقة خارجية — غير مُراجَعة (External precedent, not checked) or سابقة خارجية — مُراجَعة (External precedent, checked). Unverified claims never cause a repair and never block a card.

The change line

A line such as "Keeps 3 · merges 4 · new 2 · hands over 9 functions" is computed by the engine from where each function sits today and where the draft puts it. The AI does not write it.

"Similar to"

"Similar to B (86%)" means this draft still groups the functions much like alternative B, even after one redesign. It is kept and labelled, so you can decide whether it adds anything. A draft that was sent back and came out different says "Redesigned to differ from B".

Details

التفاصيل (Details) opens the alternative in the Alternatives step: the full structure, every claim with numbered footnotes to its sources, the change from today with a flow map, the rule check with its repair log, the score and the version history. From there you can compare it with another alternative, or edit it by hand or with the AI assistant. See Alternatives.

A consultant that failed gets a card that says "This consultant produced no alternative", with the reason and أعد التشغيل (Run again).

Run again or Regenerate all

  • أعد التشغيل (Run again) on a card asks that consultant for one more draft in the same generation. It reads the consultant's current text and project note, so you can customize a lens and run it again. It uses the generation's criteria version and inputs. The new draft is checked against the drafts already there so it doesn't repeat them, and gets the consultant's letter with a counter: "B2". Run again works only on the latest generation, after its run ended, while the frame is still locked on the same version and your inputs have not changed. Otherwise the button explains why, for example "Criteria or inputs changed — Regenerate all". The consultant need not be selected any more, but it must not be removed from the library.
  • إعادة توليد الكل (Regenerate all) starts a new generation with every selected consultant. Earlier generations stay readable.

Both open a dialog with Note for this run (optional): "The consultant puts it first, but it never overrides the rules or the grounding contract." The note belongs to that run only: it is stored with the alternative or the generation, never in the library. Each dialog shows its estimate (Run again: for one consultant).

Score a card

Generating never scores. Click قيّم (Score) on an alternative in the Alternatives step, or use قيّم الكل (Score all) there, whether it passes the rules or breaks them, to ask a critic, a separate AI agent, to judge it:

  • each criterion scored by the critic: a score from 0 to 100, its band on a fixed rubric, and a justification with citations;
  • for each direction: يحققه (Meets), يحققه جزئيًا (Partly meets) or لا يحققه (Misses), with the reason;
  • challenges to weak claims: غير مدعوم بمصدر (Unsupported), مصدر ضعيف (Weak citation), يخالف المصادر (Contradicts the evidence) or يتجاوز ما تقوله المصادر (Overreach).

Criteria measured by the engine are computed by code. سهولة الانتقال (Ease of transition) measures disruption. When the allocation covers at least 60% of filled FTE, staff are filled and FTE is counted, it uses the share of FTE that changes unit, including functions handed over; otherwise it uses the share of functions that move, and the card does not claim a staff-based figure. The allocation is stored with each generation, so an alternative always reproduces from the numbers it was built on, not from today's snapshot. التكلفة (Cost) is measured in money when the current state has enough cost data: the cost coverage of the design scope (see Workforce cost) is at least 60% of its filled FTE (staff outside the scope do not count) and today's leaders have a known median cost. The card then shows the design's monthly cost in ر.س/شهر (SAR per month) with تغطية التكلفة (Cost coverage). The design's monthly cost is today's cost, plus the change in leadership boxes (proposed boxes minus today's boxes in the top three tiers) at today's median leader cost, minus the cost of functions handed over to other bodies. The score is 60 + (1 − design cost ÷ today's cost) × 100: 60 means the same monthly cost as today, a cheaper design scores above 60 and a costlier one below. Today's and the proposed leader boxes are priced the same way, at today's median leader cost, so a design that keeps today's leader boxes and hands nothing over scores exactly 60, even when some of today's leader seats are vacant or have no cost.

This is an approximation: where a leader seat is also in a handed-over function, the overlap can be subtracted twice. Otherwise (low coverage, no leaders with a cost, or a current state from Excel) the criterion stays تقدير تقريبي (عدد المناصب القيادية) (Rough estimate (number of leadership positions)). Only in that proxy are leadership boxes compared with today (fewer boxes scores above 60, as many scores 60), exactly as before, and the card is labelled as an estimate. Neither form uses the figures of the Workforce requirements step. Cost is a position cost, never a person's pay, and the consultants only see unit and function totals. A current state imported before cost existed has no cost until you re-import it. The card then shows the الدرجة الموزونة (Weighted score) with one bar per criterion. If a criterion has no score, the card says "Some criteria have no score" instead of a weighted score.

The critic scores against a fixed rubric, not against the other cards, so a score does not change when another alternative is added. A consultant never grades its own work. While a card is being scored it shows "Scoring…" and Score is disabled. Scoring needs a locked frame ("Lock the criteria & weights to score alternatives").

Each score keeps the criteria version it used: "Scored on criteria version N". If you lock a newer version later, the card adds "The criteria changed since this score" and offers أعد التقييم (Score again); the old score stays as it was. A score also keeps the version of the alternative it scored: after you edit an alternative, its card says "Scored on version N" until you score again (see Alternatives). Generate and Alternatives show a weighted score on each card; you choose and approve an alternative in Approve.

Generations history

Each generation gets a number. The switcher reads like التوليد 3 · نسخة المعايير 2 · 5 بدائل (Generation 3 · criteria version 2 · 5 alternatives), and the newest is marked الأحدث (Latest). Older generations are read-only, with "Built on criteria version N" and, when it applies, "Criteria changed since (now version N)" or "Inputs changed since".

Every alternative records the consultant text it was built with. Customizing or removing a consultant in the library later never changes what an old alternative says about it.

Usage limits

  • Each consultant draft has its own AI usage limit, and a generation's limit grows with the number of consultants.
  • Each project has a monthly AI usage limit. Start is blocked when what is left cannot fit a new generation, and the readiness list says so.
  • A consultant that reaches its limit stops alone («بلغ هذا المستشار حد الاستخدام قبل أن يُنهي» (This consultant reached its usage limit before it finished)). If it already had a draft, the draft is kept and badged as breaking the rules.
  • As a guide, five consultants take about 10 to 16 minutes. Your estimate depends on your evidence.

Jadiroon sets the limits for your workspace. Contact your Jadiroon representative to change them.

Where your data goes

  • A generation sends the text of your documents (up to 8 MB per consultant), your selected similar-experience briefs, the locked frame, the catalogue, the current state, the scope, the ecosystem and your confirmed decision rights and committees (decisions, holders as units or committees, thresholds; no person names) to the AI model configured for the studio, once per consultant. A score sends the alternative and the same evidence once more. The allocation sent with a generation carries FTE per unit and function, never names of employees. See AI transparency.
  • Everything is recorded in the Audit log. Starting a generation, Run again and Score are marked as AI actions. Library edits, roster choices and cancelling are recorded as your actions.