Explore a situation through goals and dependencies

Use this account when a proposed change looks useful from one participant’s position, but you cannot explain what it would require or change for the others. It develops an early inquiry using participants’ intentions, provision and alternatives. The result may be a comparison, a newly discovered difficulty or a precise question for investigation. A diagram is useful when it makes those relationships easier to inspect; a list of goals alone supplies no analysis.

The connected contributions are PSD.2 for whose concerns enter, PSD.5 for what the model can answer, PSD.8 for alternatives, and PSD.11:4.2.1 for qualitative analysis. Enter at the unresolved question and retain usable answers. If a current direct calculation or an already qualified comparison settles it, use that result.

Let the question determine what to model

Begin with an episode, proposal or unexplained result. Ask who needs an answer, what they could do with it and which difference would change that action. Obtain relevant participants’ accounts where possible; otherwise use attributable interviews, documents and observations, retaining unanswered questions for those who know the work. Participation in constructing the model can help correct it, but attendance does not establish agreement with a proposal.

Consider a car park whose unreliable sensors sometimes send drivers towards occupied floors. Attendants walk the floors and correct the displayed counts. Management proposes replacing the entire system, since the sensors cannot simply be exchanged in the current design. The initial inquiry is: does replacement address the service difficulty better than continuing manual correction? This is an adaptation of the parking exercise in the iStar tutorial, slide 57. The following participants, conditions and judgements are constructed for this walkthrough.

Include drivers because the guidance affects finding a place; attendants because correction requires their work; the operations manager because staffing and service arrangements are theirs to decide; and the technical supplier because replacement feasibility and installation depend on its contribution. Management cares about operating cost and reputation. Drivers care about usable guidance and protection of their cars. An attendant’s account of an assignment is evidence about that work, not a statement of every driver’s preferences.

Keep the questions distinct. What would make the displayed count trustworthy? Can the work supply it when needed? How would the alternatives affect cost and the concerns behind reputation? A model of intentions can expose these questions before it has the technical estimates needed to answer all of them.

Connect provision to the reasons on both sides

Separate a desired state, an action, a required entity and a quality. “Current guidance is available” states a desired result. “Walk each floor and enter a corrected count” prescribes work. The count is information used in producing the display. Accuracy concerns whether that information represents the occupied and available spaces well enough for the intended guidance. A completed round does not by itself establish that accuracy.

For each important provision, connect five positions: the receiving participant, why that participant needs it, what must be provided, the provider, and how the provider proposes to obtain it. For example:

The operations manager needs a checked floor count from the attendant so that the service can issue current guidance; the attendant obtains the count by inspecting the floor and entering the result before the next update.

Keep the attendant’s checking task, the supplied count and the manager’s need as distinct parts of this statement. If the checking method or update deadline is unknown, leave that question visible. Describing a dependency does not establish that the provider has accepted the work or can perform it.

Requiring a count with stated fitness conditions leaves room to choose its production method. Requiring a specified inspection task restricts that choice. State which relationship the work needs. Similarly, two recipients needing information called “the floor count” may need different intervals or accuracy; matching names do not establish interchangeable provision.

Now recover the internal relationships that explain the alternatives. Current guidance needs a suitable count and a working way to publish it. A count might come from supported automatic sensing or from a sufficiently recent manual inspection. These are candidate means; neither is adequate merely because it appears under an “or”. Manual correction also needs staff time and access. Replacing the system needs a qualified design, installation and an operating arrangement. A technical supplier’s future delivery is an external dependency, not a task the attendant can perform.

For each arrow, state whether it means a necessary condition, a sufficient alternative, a favorable or adverse contribution, or a quality of something. Accuracy can help reputation without guaranteeing it: the manager also relates reputation to protection of parked cars. An arrow saying that protection matters does not establish that it is provided.

Show the part of the model needed for the current question

A view showing only participants and external provision helps locate a missing supplier. Opening the manager’s and attendant’s internal reasons exposes why the count is needed and what work supports it. Opening the supplier’s proposed arrangement becomes useful when installation or later maintenance can change the answer.

In iStar these are the strategic dependency view, strategic rationale view and selective combinations of them. They show different amounts of one model. They are not required stages, and a hidden detail is not an absent condition. Preserve the same identified dependency when changing views; otherwise a convenient simplification can silently replace the thing being examined.

For this car park, an overview can show “attendant supplies count”. To examine availability during a shift, open the checking task and its staffing conditions. To compare whole cost, obtain the relevant cost account; opening more circles does not calculate it. The iStar 2.0 Guide, §§2–8 supplies the notation, distinctions and integrity rules when an actual iStar model is wanted. This prose walkthrough uses those distinctions without claiming to be a solver-ready model.

Turn a working question into an analysis, then recover its meaning

Define one alternative precisely enough to tell which work and assumptions change. Alternative A retains the existing displays and uses manual correction. Alternative B installs the proposed replacement and uses its automatic counts. For the initial comparison, suppose staffing for A is available and the supplier has described a technically plausible B, while B’s installation interval and whole cost remain unknown. Do not compare A in its troublesome current state with B in an unexplained ideal state.

An initial analysis value expresses an observation or an assumption in this particular question. To ask what B would change if installed and operating as proposed, provisionally assume that installation and specified sensing performance hold. Mark both as assumptions. Keep the unresolved installation and cost questions attached to the investment decision. A conditional service comparison can proceed without pretending those inputs have been obtained.

Likewise, leaving manual rounds out of B means they are not part of that proposed arrangement. It is not evidence that attendants cannot inspect a floor. An uninvestigated element stays uninvestigated; an investigated effect may remain unknown; opposed grounds may remain in conflict. Neither omission nor uncertainty means a negative observation.

Carry each alternative through the relationships using PSD.11’s qualitative branch. Keep the initial assumptions and intermediate judgements with its result. For A, a sufficiently recent and correctly performed round supports a corrected count; publication still needs the working display. For B, the stipulated sensing performance supports the count conditionally; it does not establish the installation date or remove all future maintenance work.

Contributions require the relevant person’s or specialist’s judgement when the incoming grounds do not settle the answer. Here management judges that fewer misleading indications would help reputation. It cannot infer satisfactory reputation from that improvement alone while the protection concern remains unanswered. The same extra staff work that supports accurate guidance also adds operating burden. Do not cancel these effects by counting favorable and adverse arrows.

Return an answer in the language of the original inquiry:

This comparison does not yet justify the investment. It identifies what the proposal can improve and which missing facts can change the choice. Obtain a bounded installation and whole-cost answer if its expected contribution warrants the work; otherwise retain the conditional comparison or defer the choice. The answer is not “B has more satisfied goals”.

Use surprising results to discover a different problem

Suppose the first model gave A an unqualified favorable result because it represented the round but omitted other assignments. Ask what must be true for that result to hold. Inspecting the rota now reveals that the same attendant must be at the security desk from 10:00 to 10:15, while the modeled correction requires leaving it for a round from 10:00 to 10:10. No substitute is assigned. These stipulated facts defeat the simultaneous arrangement.

The discovery changes the model and the problem. Add the required desk coverage and its relation to the attendant’s available work; then revise A’s count availability for this interval. The earlier claim that a completed, timely inspection can supply a count still holds. B’s separate technical hypothesis also survives. What fails is the premise that A’s work can coexist with the retained service obligations.

The next question is now: how can current guidance and desk coverage both be obtained in this interval? Another discussion asking the attendant to agree with the plan supplies no missing time or replacement. The manager who controls assignments must examine a different allocation or contributor. OCE.11 develops that coordination when change work competes with continuing service.

Work backward from the joint requirement. Current guidance needs an adequate count and publication; desk coverage needs a person at the desk. One candidate is a separately allocated competent checker while the attendant remains at the desk. Test it forward: can that person obtain the count, update the display in time and do so without abandoning another necessary assignment? Naming a replacement is not that test. Until these conditions are established, the candidate is a question for the manager, not an available solution.

For a bounded continuation, suppose the manager assigns a competent relief worker for this interval, verifies that no other required assignment overlaps, and provides the permitted access and existing checking procedure. In the stipulated trial the relief worker obtains and publishes a count meeting the agreed freshness and accuracy conditions, while the attendant remains at the desk. Both required contributions are supplied in this interval. The added staffing cost belongs in A’s cost account. If relief is available only today, retain today’s result and reopen the staffing question before promising tomorrow’s service; the earlier investment question remains open.

A different candidate might change the update interval or the information promised to drivers. That changes the service requirement, so return to whether the altered guidance remains useful and acceptable. Do not weaken the criterion silently to make the diagram succeed. An incompatible candidate can be rejected without proving that all possible arrangements fail; an unsupported candidate stays conditional.

Keep the unsuccessful combination and the reason it failed so that the next attempt does not repeat it. In a formal backward analysis, ask which supported conditions and judgements produce the incompatibility. A solver’s failure can concern its encoding, restrictions or current assumptions; it is not a proof that the real organization has no remedy.

Stop with a usable answer and preserve the right return

The manager now has a useful result: replacement is not yet justified by the conditional service comparison; the original staffing proposal cannot provide both services; and the revised allocation has supplied them in the stipulated interval. The immediate allocation question and the longer investment question have different owners and inputs. A repaired rota can settle the first without settling the second. A completed cost comparison does not itself allocate a checker.

Stop when the inquiry’s purpose has been met at its declared strength: an adequate comparison is available, the decision-changing incompatibility is explained, or the next useful investigation is precise. Do not enlarge the model simply to fill every possible actor boundary. Conversely, a final label is not sufficient when the recipient cannot understand which action or missing premise it implies.

When a new fact arrives, identify which assumption, relation or judgement changes. Reconsider the alternatives and consequences that depend on it, including their shared conditions, while retaining independent results. If a technical question needs dynamics, accuracy bounds or failure probabilities that this model cannot establish, use an appropriate technical model and bring its qualified answer back. More qualitative arrows do not supply those quantities.

When the analysis itself is difficult, try a small interpretable question whose expected result can be explained from the modeled relationships. Inspect unreached intentions and omitted providers before taking coverage for granted. The authors’ later suggestions include extreme hypothetical settings and realistic alternatives as ways to question a model. Such settings are diagnostic questions, not feasible plans or a compulsory checklist. Preparation in the chosen language and help interpreting a model may be needed; neither a completed diagram nor this walkthrough establishes independent modeling capability.

Language, analysis and evidence have different limits

The iStar Guide defines a language; the PoEM 2009 treatment develops interactive modeling and forward evaluation. The ER 2010 procedure adds formal backward search with human judgements and backtracking under stated structural restrictions. The manual backward question used here is an adaptation, not an implementation of that algorithm. If a named solver is used, select compatible language and analysis versions and preserve their restrictions, rather than applying older propagation tables to iStar 2.0 without a justified mapping.

The subsequent systematic-versus-ad-hoc study gives reasons to attend to interpretation, coverage and training. It does not establish that a systematic procedure uniquely produces better understanding or more useful new questions. Use the model and its analysis for the contribution they can actually supply, then establish the domain evidence and decisions needed for reliance.