AI plan dependency audit: trace seven tasks before trusting a schedule

What must finish before a task can begin?

For this exercise, a dependency means that a named earlier task must finish before the dependent task can start. Microsoft describes this as a finish-to-start relationship and distinguishes it from other relationship types. We deliberately use only that rule here. Do not generalise the findings to a real plan containing overlapping work, different relationship types or other conditions.

Write a boundary card before reviewing the records: “Practice version V1; supplied tasks E1–E7 only; every listed predecessor must finish; all tasks initially incomplete; no dates, durations or missing links may be invented.” A predecessor is simply a task that must finish first under this rule. The task identifier, such as E2, is the stable reference you will inspect.

If a row lists two predecessors, both must finish. It is not an either-or choice. Likewise, a task without a listed predecessor is unblocked by these written dependencies, not automatically authorised for real execution. Staff, equipment, access and permission are outside the fictional record. Keep those limits beside the table so a planning observation cannot turn into an operational instruction.

The practical artificial intelligence guide covers other first tasks. This walkthrough is narrower: produce a saved evidence ledger showing an ordinary wait, a circular wait and an unresolved reference in one supplied plan. It does not schedule a real project or change anyone’s task board.

Copy the seven-task fictional practice card

Imagine a small exhibition team preparing an internal sample display. The following record is intentionally flawed teaching material, not the team’s real operating procedure. Task names help you remember the rows, but do not use their apparent business logic to rewrite the supplied relationships. You are checking what the card actually says, not designing what it should have said.

Check that E1 through E7 each appear once. The reference E8 does not have a row in this supplied card. That is a missing reference in the material you have, not proof that no such task exists anywhere. Ask for the source or a corrected identifier. Do not create a plausible eighth task to make the table look complete.

The E4 and E5 rows point at each other. Their names may tempt you to remove the approval dependency from preparation. Resist that shortcut. Choosing which link is wrong requires a source-owner decision. The audit should preserve both original rows and make the contradiction visible, rather than silently changing the problem into an easier one.

Which task is unblocked at the starting snapshot?

Create a separate ledger with these headings: task ID, supplied predecessors, initial review status, evidence and question requiring clarification. Preserve the source card above it. A review status is your annotation; it is not a command to mark a task complete in an application. At this snapshot, no task has actually been performed.

Asana’s dependency guidance distinguishes a task that blocks another from a task that is blocked by another. Preserve that direction in plain sentences. “E2 waits for E1” is not equivalent to “E1 waits for E2.” Reading the relationship aloud is often enough to catch a reversed arrow without installing any planning software.

Only E1 lacks a written predecessor at the initial snapshot. E7 has two requirements, even though E3 itself follows E2 and E1. Retain the explicit E1 entry when copying the card; do not simplify source relationships during an audit. In a separate hypothetical trace, completing E1 would free E2, not E7. E3 would still be incomplete.

How do you demonstrate a circular wait?

Start at E4 and follow what it waits for: E5 must finish first. Move to E5 and follow its requirement: E4 must finish first. You have returned to the starting task. Because both are incomplete and the only allowed relationship requires the earlier task to finish, neither can be the first task to start in that pair under the supplied rule.

Record the evidence as two sentences: “E4 waits for E5. E5 waits for E4.” Then record the question: “Which supplied relationship should the source owner confirm or correct?” Assigning a higher priority to E4 does not satisfy E5’s completion requirement. Priority and prerequisite are different concepts; changing the display order does not remove the written dependency.

E6 has a different problem. You cannot follow E8 because its row is absent. There is no evidence here of a circular return through E8. Label the reference unresolved and ask whether the card omitted a task, used the wrong identifier or relies on a separate source. Do not describe every blocked item as a cycle.

A visual connection is also not sufficient evidence of work order. The mind-map source review distinguishes a topic relationship from an operating sequence. Here the actual predecessor column supplies the relationship. An attractive diagram is optional; it cannot replace the two source rows proving the circular wait.

Why does a valid partial order not repair the whole plan?

Run a thought experiment, keeping it separate from the initial snapshot. If E1 finished, E2 could satisfy its written prerequisite. If E2 then finished, E3 could do the same. If E3 finished, E7 would have both E1 and E3 complete. Thus E1, E2, E3, E7 is one feasible order for those four tasks under the supplied dependency rule.

None of those completions actually happened. Write hypothetical trace, not completed work. That trace leaves the E4/E5 circular wait and the E6/E8 reference unresolved. The correct conclusion is a usable partial sequence plus explicit open issues, not “the project plan is fixed.” A complete schedule would need additional information that this teaching card does not provide.

Review three intentionally flawed claims. “E7 is ready when E1 finishes” ignores E3. “Remove E6 and all remaining tasks are ready to schedule” ignores E4 and E5. “Put E4 above E5 and the circle disappears” confuses list position with a source relationship. Mark each revise and name the exact row supporting your correction.

The infographic review checks preserved order and a stated condition in a visual output. This task instead follows explicit predecessor identifiers and detects a closed wait. If you later write a real procedure, the standard operating procedure guide covers a broader deliverable. Keep these outputs distinct.

What can you save and reopen within fifteen minutes?

The core exercise requires only the supplied fictional text and a note editor. On Windows, open Start, type Notepad and open a new document. Add “Dependency practice V1: every task initially incomplete” as the heading. Copy the boundary card, all seven source rows and the separate review ledger. Include both cycle sentences and the unresolved E8 question.

Add the four-task sequence under a heading saying hypothetical trace, not actual completion. Press Ctrl+S, choose Documents and save as dependency-review-V1.txt. Close the editor, open the saved file from that folder and reread E7’s two prerequisites, the E4/E5 circular wait and the missing E8 reference. If any disappeared during copying, repair the record before handing it over.

For an optional second opinion, give the fictional source and boundary card to an assistant you already use. Copy this request: “Every listed predecessor must finish before its task starts. All tasks are initially incomplete. Separate initially unblocked tasks, ordinary waits, a circular wait proved by identifiers, and references without source rows. Invent or delete no IDs or links. A feasible partial sequence does not validate the complete plan. Show evidence for each judgement.” No particular product account or model response was tested for this walkthrough.

Compare the proposed result with the human key above. If the assistant returns a complete seven-task schedule, check where it silently removed a relationship or invented E8. Ask it to preserve the source and identify the unresolved issue. If access is unavailable, record optional AI comparison not performed. The manual source-and-decision artifact can still be complete; do not claim that an unperformed assistant stage passed.

Allow three minutes for the source, four for initial statuses, four for tracing and questions, and four for saving and reopening. This is a suggested work allocation, not measured completion performance. My earlier writing about perfection and effectiveness described work delayed by pursuing every detail. Applying that principle here means finishing one traceable review before redesigning the whole plan, not claiming personal results from a dependency-management trial. The freelancer prompt collection offers separate writing tasks for later.