A research plan might list cleaning data, estimating models, and interpreting results. Work does not necessarily finish in that order. Estimation can reveal an unsuitable definition; interpretation can reopen sample questions; a new paper can change how earlier work is understood. A completed action is real, but its checkmark may no longer explain the next step.
Many next steps become clear only after the preceding work produces information. Today’s analysis can be arranged without predicting every question it will raise. A plan that allows tasks only to move from pending to complete can then diverge from the investigation: the checklist advances while the research needs to return.
Planning the unplannable means organizing work within that uncertainty. A full route can emerge progressively, while current questions, available outputs, and decisions are organized now. We are integrating Planar into Marx Max around that idea, including feedback relationships and human–AI roles in a shared plan.
Why Research Plans Change
A question does not prescribe every operation. Comparing groups requires finding whether the data contain the relevant information. Testing a mechanism requires assessing its measures. Interpreting an estimate can call for another comparison. Execution supplies information that revises subsequent execution.
Understanding Science describes investigation as more varied than a single linear recipe, with conclusions revisable in light of evidence. Our implication for task management is to record conditions for continuing and findings that reopen earlier questions alongside the actions themselves.
Executing a specified regression can be clearly bounded: load data, calculate, save output. Whether the model answers the research question involves additional discussion of samples, measures, and methods. Finishing the first activity does not establish completion of the second. A single completion status can obscure that distinction.
What is difficult to predetermine is where exploration will lead. Defined work can still be scheduled and stable procedures automated. A research plan helps established and unresolved parts advance together without requiring every exploratory turn to be fixed on day one.
Including the Return in the Plan
Consider a hypothetical analysis divided into defining variables, constructing a sample, comparing models, and interpreting results. This illustrates task relationships rather than an observed user project:
During model comparison, observation counts differ. The issue may originate in sample processing as well as adding controls. Continuing directly to interpretation could turn an unresolved difference into a conclusion. Returning to sample construction allows missing-value and exclusion rules to be examined.
The plan now contains a feedback relationship:
More generally, a current task sends a question back to a relevant predecessor:
The relevant task need not be the immediately preceding one. A definition problem can reopen an earlier node. Feedback identifies where the problem arose, why a return is needed, and what work remains, making it more specific than a generic instruction to redo everything.
Planar’s feedback design can return a question to a designated task, reopen selected checks, or add an unresolved item. Valid work is retained while rework follows the issue. Returning can be an ordinary way of advancing a plan, without resetting every completed part.
Changed Inputs and Existing Outputs
A return raises another question: which outputs are affected? Revising a sample can require reviewing tables, figures, and interpretations calculated from its earlier version. Reopening the sample task alone may not reveal those relationships.
Plans can therefore represent how outputs are used, alongside task order. A dataset supplies a model; model results support an interpretation. These links help locate what needs review when an input changes and make the next affected work easier to identify.
Planar’s design distinguishes conditions for proceeding, artifact-use relationships, and feedback to previous tasks. They answer different questions: when work can proceed, what it relies on, and where an unresolved issue returns. Keeping them together is one reason we consider the design relevant to research.
A change does not necessarily invalidate every later output. Renaming a variable and revising a sample definition have different consequences. Flagging related work for review leaves room to assess that impact rather than declaring everything valid or invalid in advance.
Assigning People and AI within the Loop
A plan involving AI also needs to identify who handles the current work. Data inspection, programming, and comparisons under specified conditions can receive AI assistance. The meaning of a measure, relevance of a comparison, and interpretation supported by evidence involve researcher judgment. Roles can change as the task develops.
For the sample discrepancy, AI can list observation counts, trace exclusions, and produce a common-sample comparison. A researcher considers which sample fits the question. Missing information can become another task or feedback item; once conditions are clarified, AI can continue computation on that basis.
Human participation need not be confined to a final confirmation. A consequential decision can be represented as a task with required materials, a decision-maker, and work that can continue afterward. If AI cannot resolve it, progress and the obstacle can remain visible for a person to consider, instead of inventing a condition merely to continue execution.
The Planar integration represents lead, collaborator, and review responsibilities. AI can report progress, identify blocks, and submit results, with submission distinguished from human acceptance. The plan records responsibility and progress; assigning a task does not automatically start every activity in the background.
People and AI can then discuss the same task rather than maintain disconnected lists. Revised requirements, outputs, and feedback remain associated with it. We aim to reduce repeated reconstruction of where work stands and why it returned, leaving more attention for the unresolved question.
More Than an Extra Arrow
Complex planning is not exclusive to research. Asana supports dependencies, and ClickUp’s AI tools manage tasks, comments, and dependencies. A case for Planar cannot depend on claiming that alternatives only offer checkmarks. An existing system that represents the work adequately may not need another layer of maintenance.
Our focus is a specific mismatch: organizing research as predetermined actions that end once performed, while findings continually revise questions and arrangements. Planar brings rework, affected-output review, and human–AI responsibility into the same plan. Its usefulness belongs to that workflow, rather than a feature-count comparison.
A simple list can be sufficient for a stable routine. Investigations requiring repeated decisions, returns, and handoffs benefit from expressing more of the process: why work proceeds, pauses, or returns after a step finishes. That motivates the design.
Keeping the Plan Responsive to Discovery
Planning the unplannable does not require forecasting every turn. It asks whether a discovery can reorganize the work: return a question to its source, identify affected outputs, locate a human decision, and continue AI assistance under clarified conditions. The next step can change without losing accumulated understanding and outputs.
Planar’s integration into Marx Max is still in progress. We intend feedback and shared roles to make the plan reflect how research unfolds. Whether this reduces repeated explanation, omissions, or unnecessary rework remains a question for actual use. The design choice is to treat a return prompted by a new finding as part of moving research forward.