1. LandOffer
  2. Blog
  3. All articles
  4. Identity and Access Management Portfolio: Demonstrate Joiner, Mover, and Leaver Controls

Job search

Identity and Access Management Portfolio: Demonstrate Joiner, Mover, and Leaver Controls

Demonstrate a bounded identity lifecycle with an original approval policy, executed joiner/mover/leaver checks, and an independent access-review packet.

10 min readLandOffer team

An identity and access management portfolio should show what access a person receives, what they lose when their role changes, and how you verify removal when they leave. A screenshot of a group membership is only one state. A stronger sample names the policy owner, states the approved rule, records the resulting state, and checks both permitted and denied behavior. The reviewer can then see whether your evidence supports the claimed transition.

This guide uses an original, entirely local Python lifecycle model and a separate fictional access-review packet. The code was executed on October 23, 2026. It does not connect to Microsoft Entra, another identity directory, or real employee accounts. Every identity, permission, approval, and business situation in the examples is invented.

Three stages of a textile workshop attach approved threads, exchange a role thread, and detach access at departure.

Use LandOffer’s recently posted jobs to identify the lifecycle, provisioning, or access-review responsibilities you want to demonstrate. Choose a small policy you can explain and test, rather than presenting a local simulation as enterprise deployment experience.

Define the authority behind an access change

Microsoft’s employee-lifecycle deployment guide describes HR-driven provisioning with the HR system as a source of authority. It also separates provisioning, lifecycle tasks, and removal of application access. That distinction matters for a portfolio: a role label in a spreadsheet is not automatically an authorized request to change every connected system.

For the authored model here, a trusted simulated HR event supplies the lifecycle category and role. A separate approved boolean represents the required grant approval. This is an input assumption, not an implemented approval system. No manager or resource owner actually approved an account change.

The completed policy matrix is deliberately small:

Event Required authority in this model Intended modeled result
Join engineering Effective HR event plus represented grant approval Active; baseline and development-repository access
Move to support Effective HR event plus represented new-role approval Active; baseline and support-queue access; development access removed
Unapproved join or move Approval absent Existing state unchanged; no new access
Leave Effective authoritative departure event Inactive; every modeled entitlement removed

The model owns exactly three entitlements: baseline, dev_repo, and support_queue. It has no shared roles, privileged administration, nested groups, or independent exceptions. Those exclusions make exact role reconciliation appropriate inside the exercise. Applying the same overwrite to an actual directory could remove unrelated permissions and would require a different design.

Microsoft’s lifecycle-workflow overview distinguishes tasks from execution conditions: who is in scope and when the workflow runs. Our local function receives an already-effective event. It does not implement an HR connector, schedule, workflow trigger, or event-authenticity check. The product documentation is context, not evidence that this sample runs in that product.

Worked asset one: an executed joiner, mover, and leaver case

The first original asset starts with a fictional principal who is inactive and has no entitlements. The trusted teaching inputs move that principal through engineering, support, and departure. A separate rejected-join branch tests the absence of approval without changing the main sequence.

The intentionally defective version has three related problems. It grants role access without consulting approval. On a move, it adds the new entitlements to the old set. On departure, it changes an active flag but leaves memberships behind, while its access evaluator ignores the flag. The finding record points to before.py: the join/move branch ignores approval and unions grants; the departure branch leaves grants behind; the evaluator checks membership alone. Each observation is inspectable in our own authored source, without alleging a vendor or employer defect.

The repaired version checks approval and a recognized role before granting joiner or mover access. It replaces the fully managed entitlement set with the approved role’s set. Departure changes the active state to false and clears the modeled grants. The evaluator also requires an active principal before allowing a resource.

The resulting states in the repaired sequence are explicit:

  • Rejected join: inactive, empty grants; development access denied.
  • Approved join: active, baseline and dev_repo; support access denied.
  • Approved move: active, baseline and support_queue; old development access denied.
  • Departure: inactive, empty grants; baseline and support access denied.

A reviewer can inspect each expected result alongside the actual value. The test runner checks thirteen assertions across both versions. The defective version passes seven; the repaired version passes thirteen. The valid-access cases show that the corrected policy still supports the intended work. The negative cases show that missing approval, an old role, or departure changes the access decision. Denying everybody would fail the approved joiner and mover requirements.

The complete comparison is:

Check group Before After
Unapproved join: denial and inactive state 0 of 2 2 of 2
Approved join: baseline, development, no support 3 of 3 3 of 3
Move: support, old development denied, baseline retained 2 of 3 3 of 3
Repeat the same active mover event: unchanged state 1 of 1 1 of 1
Leave: inactive, baseline denied, support denied, no grants 1 of 4 4 of 4

The sample includes before.py, after.py, fix.diff, regression.py, and the two saved execution outputs. From the supplied demo directory, rerun:

python3 regression.py > replay.json

The actual runs used the project’s existing Python environment and standard library only. Their outputs were byte-identical in two fresh processes. The receipt records interpreter and platform details, source and runner hashes, state snapshots, and each assertion’s expected and actual value. The retained before-version is evaluated under the same declared policy as the repaired version. No directory command, network request, package installation, or real permission change occurred.

Explain what a successful removal test does not prove

Here, “denied” means the local evaluator returned false for the named entitlement. It does not mean an employee’s browser session was terminated, a refresh token was revoked, or a downstream SaaS application rejected a request. No session, token, device, database, or application connector exists in this program.

The repeated-event check is narrow. It repeats the same approved move while the principal is active and confirms the state is unchanged. It does not demonstrate safe handling of delayed HR messages, out-of-order events, rehires, or an old joiner event arriving after departure. A later approved join or move could reactivate the principal in this toy. That limitation belongs in the report, not in a footnote concealed behind a green test count.

A real implementation would need an authoritative event sequence and a defined treatment of stale changes. It would also need evidence from each relevant target. Those are proposed next reviews; this exercise did not build them. Do not claim an offboarding service-level target from a function that runs immediately when called and contains no scheduling or timing measurement.

The same care applies to scope. Clearing this model’s three grants covers its complete invented inventory. It says nothing about direct application roles, nested membership, credentials outside the directory, or exceptions maintained elsewhere. A recorded workflow success should therefore be matched to its actual target and operation. Our exercise supplies only local state and evaluator results; it cannot fill the missing target-side evidence.

Worked asset two: an independent access-review and exception packet

The second asset uses a different fictional principal, case K, with four recorded grants. It is a completed review recommendation, not another run of the lifecycle principal or a claim that a real reviewer clicked Approve. Its invented review time is November 4, 2026 at 12:00 UTC.

The reviewer has recorded a current evidence-reconciliation assignment, an old billing project that ended October 30, no documented administrative task, and a two-day incident-index handover. Those facts are stipulated for the example; they were not collected from an employer.

Grant Completed recommendation Reason and required owner
audit_read Retain, pending confirmation Current reconciliation work; Audit owner confirms need
billing_export Remove, pending confirmation Prior project ended; Billing owner confirms closure
admin_console Remove, pending confirmation No documented admin task or sponsor; Platform owner reviews
incident_archive_read Propose a read-only exception Handover only; Archive owner must approve EX-K17 through November 6, 12:00 UTC

The packet contains one retain recommendation, two removal recommendations, and one proposed exception. The exception is not granted. Every owner confirmation is pending, and every application action is marked “not performed.” This preserves the difference between making a useful recommendation and possessing authority to implement it.

A separate local check validates the inventory counts and date arithmetic. The proposed exception window is 48 hours from the fictional review time. At the independent hypothetical check on November 7 at 12:00 UTC, that window has expired. The script only compares timestamps and checks the recommendation counts. It does not schedule a future task, change a grant, revoke a session, or prove that an owner honored the expiry.

Microsoft’s access-review overview discusses reviewing continued access with resource owners and maintaining policy exceptions. Our packet applies those ideas to an authored situation. It does not establish a review cadence required by every organization or a general rule that all temporary access should last two days.

Keep the decision, implementation, and verification records separate

The independent packet ends with this completed, unsent owner message:

Subject: Case K access-review decisions and EX-K17 expiry

Please confirm the current need for audit_read and whether the ended billing project supports removing billing_export. No administrative task or sponsor is documented for admin_console; my recommendation is removal, pending the Platform owner’s decision. The Archive owner should decide whether the read-only incident-index handover justifies EX-K17 through November 6 at 12:00 UTC. This exception has not been approved or granted. No changes have been applied. After authorized action, the operator should record the affected assignment, result, and target-side access check before marking removal complete.

The message is a finished portfolio artifact, not an instruction sent to an employer. It specifies the decision needed, the expiry, the authority gap, and the proof required after action.

Microsoft’s review-completion guidance distinguishes review completion from applying results and describes cases where applying a denial does not remove the underlying access, including certain nested or synchronized-group situations. That is a useful reason to avoid equating a recorded Deny decision with verified removal. We did not test those product behaviors in a tenant.

An orchard review separates old branch ties from one retained tie and a temporary tie beside an hourglass.

Package evidence without inflating your contribution

Give a reviewer two short routes through the repository. For the lifecycle case: policy matrix → defective transitions → patch → thirteen expected-versus-observed checks → limitations. For case K: four-grant inventory → recorded business need → recommendation and owner → pending application → required future verification. The second route must not borrow the first route’s green test result as proof that its recommendations were implemented.

A supported portfolio sentence is: “Authored a synthetic Python lifecycle model, corrected approval and stale-role removal defects, and verified thirteen local assertions; prepared an independent four-grant access-review packet with an unapproved, time-bound exception.” It does not claim Entra administration, a deployed connector, a successful audit, or production offboarding.

Before sharing, make sure the visible source matches the saved run receipt. Hashes identify which files were reviewed; they are not independent assurance. The same author created the policy, implementation, and tests, and another runtime has not been separately validated. Keep that disclosure alongside the results.

Use LandOffer’s current listings to identify relevant IAM responsibilities, then select the original policy and evidence that explain your contribution. LandOffer does not validate this lifecycle model or perform the directory changes it describes. Keep the local demonstration and any future authorized platform implementation clearly distinguishable.

Download the worked example files

Download the curated educational companion. Read PUBLICATION-README.md after extraction for the public file layout, replay commands and requirements. The listed offline Python checks were rerun from the extracted archive in a disposable directory. The earlier local report and commands describe the original evidence environment; use the companion README for the portable layout. No whole evidence directory, credentials or compiler binaries are included.

Sources and Further Reading