~/qa-guides/exploratory-testing
>_ Exploratory Testing: Examples, Test Charters, and Session Notes
Learn how to scope an exploratory session, choose tests from observations, and report useful evidence. Includes three charter examples and a complete session walkthrough.
Short answer
Exploratory testing combines learning about software with designing, executing, and evaluating tests. The tester uses observations from one check to decide what to investigate next, while comparing behavior with trustworthy expectations.
A useful session produces evidence about a defined area: behavior checked, defects observed, questions raised, and important work still outstanding. The ISTQB Foundation Level syllabus, section 4.4.2, describes this approach and its use with timeboxed sessions, test charters, and debriefs.
This guide uses that session-based structure to make exploration practical and reviewable. Other exploratory work can use a different structure. The application rules, observations, identifiers, and timings below are fictional teaching examples, not results from testing this website.
How exploratory testing relates to other test design work
In a scripted test, the selected actions and expected results are substantially established before execution. That is useful for repeating an agreed check and comparing results across builds. During exploration, the tester may discover a distinction that the original plan did not anticipate and design a new check immediately.
For example, a bulk action can behave correctly on the initial screen. After changing a filter, its selection counter may look inconsistent. That observation creates a reason to inspect which records are selected and which IDs the next action actually changes.
Error guessing can contribute a starting hypothesis based on experience. Exploratory testing also includes learning the workflow, discovering missing rules, and following unexpected observations that were not part of the initial hypothesis.
Formal techniques remain useful during a session. If exploration reveals that two input types behave differently, equivalence partitioning can help define those groups. If several facts determine an action, a decision table can make the newly discovered rule combinations explicit. Adaptation and systematic test design can support the same investigation.
When an exploratory session is worth planning
Choose a session when there is a question that execution and observation can help answer. Useful triggers include a changed workflow, a newly shared component, inconsistent support reports, or behavior that the existing suite observes only through one fixed path.
State the learning need before allocating time. "Understand whether selection stays aligned with the visible records after filtering" gives a direction. "Explore the application" leaves both the next action and the eventual report unclear.
Exploration can also expose gaps in a detailed specification. A rule may define what an action does but omit what should happen when the user changes context before confirming it. Keep the missing decision visible while testing expectations that are already established.
Agree on the session's priority relative to other release work. A focused investigation can complement required regression or acceptance checks; the session report should identify which of those obligations it actually addressed.
Write a useful test charter
A test charter records the purpose and boundaries of a session while leaving room to choose checks as information emerges. The ISTQB Test Analyst syllabus, section 3.4.1, discusses missions, scope, resources, constraints, and session records. The compact format below is a practical adaptation, rather than a mandatory form.
A reusable charter template
- Mission: Name the feature or journey and the question the session should investigate.
- Motivation: Record the change, uncertainty, or product risk that makes the question useful now.
- Scope: Identify the relevant actions, states, and boundaries; state important exclusions.
- Resources: Name the build, environment, account, data fixture, reset method, and observation tools.
- Expectations: Link the agreed rules and identify decisions that still need clarification.
- Budget and stopping conditions: Set an initial timebox and identify conditions requiring an earlier stop or review.
- Output: State what evidence and handoff another person should receive.
The mission should describe information to discover. Supporting fields make the work feasible and constrain conclusions. A charter can suggest initial checks without fixing the complete sequence in advance.
Improve a vague or overly prescriptive charter
"Test bulk actions thoroughly" gives no limit on the operation, record population, or questions to answer. Improve it to: "Investigate whether bulk archive stays aligned with explicit selection while the user changes region and status filters."
At the other extreme, a charter containing twenty mandatory clicks with fixed expected results has already specified a test sequence. Preserve that sequence as a repeatable case if it is valuable, and give the exploratory charter a learning objective that still permits useful follow-up decisions.
Avoid promising exhaustive coverage in the mission. A timebox allocates effort; it does not define a complete model of the feature. Choose a scope small enough that the final record can distinguish observed behavior from remaining questions.
Prepare data, expectations, and observations
Before the clock starts, agree on what the budget includes. Setup, active testing, investigating a finding, and writing notes all consume time. Teams can track them separately without treating any particular allocation as a standard.
Confirm that the test environment can reach the intended behavior. Check the required account, feature configuration, dependent workers, and fixture restoration. An unavailable dependency may prevent the mission from being exercised even when the page opens successfully.
Identify an independent observation for important effects. A toolbar message may show what the client believes happened; a request payload identifies what it asked to change; a separate record read shows what persisted. Decide which of these observations the mission needs.
Use stable identifiers in the fixture. "The two rows at the top" becomes ambiguous after sorting, filtering, or refreshing. Record IDs let another tester connect the visible rows to the action payload and stored state.
Keep assumptions separate from rules. A previous version or comparable product can suggest behavior to investigate, but an undocumented preference should remain a question until the relevant owner agrees on the expectation.
Worked example: bulk archive after changing filters
Charter C-01: investigate selection across view changes
- Mission: Investigate whether bulk archive uses the user's current selection as region and status filters change.
- Motivation: The work-item list has a new shared selection toolbar. Existing automated checks archive records from the initial unfiltered view.
- Scope: Explicit row selection, the selected-count display, filter changes, archive submission, and saved item status. Pagination, concurrent edits, permissions, and other browsers are excluded from this session.
- Resources: Fictional build
B-734, an isolated workspace, one account allowed to archive all fixture records, a resettable six-record dataset, and access to requests and an independent item lookup. All four active records fit on one page. - Budget: Forty-five minutes, including setup and notes. Stop earlier if the fixture cannot be restored or the observations needed to evaluate a finding are unavailable.
- Output: A session record, evidence for any finding, specific remaining scope, and a proposed next action for the feature owner.
The charter permits different exploratory routes. The chronology below is one invented execution, including a defect deliberately included for teaching.
Define the fixture and agreed rules
Start each independent investigation from these saved records:
| ID | Region | Status |
|---|---|---|
| W-101 | West | Active |
| W-102 | West | Active |
| W-103 | East | Active |
| W-104 | East | Active |
| W-105 | West | Archived |
| W-106 | East | Archived |
The fixture has four active records and two archived records. It stays isolated from background changes during this session.
The agreed behavior is:
- Changing either the region filter or the status filter clears the selection.
- The toolbar count equals the number of currently selected records.
- Archive is disabled when the selection is empty.
- An accepted archive changes exactly the explicitly selected IDs to Archived and leaves all other records unchanged.
- After successful archive, selection clears and the current list reflects the saved statuses.
These rules define this example. A real product may intentionally preserve selection across filters; its selection indicators and action behavior need expectations appropriate to that different policy.
Minutes 00-05: establish a trustworthy starting state
The tester records B-734, the fixture version, the workspace identifier, and the agreed rules in session S-01. An independent read confirms all six IDs and statuses. The Active view contains W-101, W-102, W-103, and W-104, with selection count 0.
This becomes evidence E-0. If the stored fixture had already been modified, a later missing row could be a setup problem rather than a filtering defect. The initial read makes that distinction possible.
Minutes 05-12: check the normal action and reset
With Active and West selected, the visible set is exactly W-101 and W-102. The tester selects both records and archives them. The request names those two IDs, the saved state marks both Archived, and the selection clears.
The independent read now contains two active records and four archived records. This normal-path observation is E-1. It establishes that archive can execute with the fixture and that its effects can be inspected.
Before exploring another history, the tester restores the original six-record fixture and confirms the four-active, two-archived baseline. Reusing the modified data would make the next filter show a different population.
Minutes 12-20: follow an unexpected counter
The next question is how the new toolbar responds when the user changes context after selecting records. Starting again in Active and West, the tester selects W-101 and W-102, then changes the region to East.
The rows change to W-103 and W-104. The toolbar still says 2 selected, and Archive remains enabled. Both disagree with the rule requiring selection to clear on a region change.
The tester records the filter, visible IDs, count, and enabled action as E-2a. The immediate next question is whether the inconsistency is limited to the display or whether the next operation still carries the old IDs. That question was chosen because of the observation, rather than being another item copied from a preset click sequence.
Minutes 20-30: connect the screen to the saved effect
The workspace is disposable and resettable, so the tester executes Archive and inspects the request. It contains W-101 and W-102, although the current view shows the East records.
An independent read confirms that W-101 and W-102 became Archived. W-103 and W-104 remain Active, and W-105 and W-106 remain Archived. The totals are again two active and four archived records. Request and saved-state evidence become E-2b.
The session now has a confirmed finding, F-01: region changes leave the old selection actionable, allowing records outside the new view to be archived. The stale counter and unintended operation belong in the same report because this execution connects them. They do not automatically establish two independent defects.
The archive service received explicit IDs and changed those IDs. This evidence does not establish a defect in that service's own contract. Keep the report focused on the view-change selection rule that failed, while leaving implementation diagnosis to further investigation.
Minutes 30-38: reduce the reproduction and record a new boundary
The tester restores the full fixture again, selects only W-101 in Active and West, and switches to East. The toolbar now incorrectly keeps a count of 1. Archive sends only W-101, which becomes Archived.
Evidence E-3 shows that the issue also occurs with a single selected row. The resulting saved totals are three active and three archived records. The smaller sequence removes a possible assumption that the defect requires selecting multiple rows.
The tester records a follow-up idea: inspect status-filter changes separately. That path is covered by the agreed rule but has not yet been exercised. A shared toolbar makes it worth checking, while the current evidence remains specific to the observed West-to-East region change.
Minutes 38-45: close the record without expanding the claim
The tester links F-01 to E-0, E-2a, E-2b, and E-3, preserves the normal-path control E-1, and records the remaining work. The next session or targeted follow-up can investigate status changes and additional view-change directions.
Session time is recorded as five minutes of setup, thirty-three minutes of testing and investigation, and seven minutes of report preparation: 45 minutes in total. These are example effort categories, not a required allocation or a coverage calculation.
The fixture is restored for the next tester, while exported evidence retains the relevant before-and-after states. The session ends with a finding and explicit gaps. It does not issue a verdict for the whole release.
A session-notes template and its completed example
Use a session sheet to retain the context and reasoning that another person will need. The format can be a document, issue attachment, or test-management record. Keeping identifiers and decisions consistent matters more than the storage tool.
Reusable session record
- Identity: Session and charter IDs, build, environment, tester, and actual start and end times.
- Data: Fixture version, starting-state reference, relevant record IDs, and reset notes.
- Investigation entries: Relative time, action or change, observation, expected behavior, evidence reference, and reason for the next check.
- Findings: Confirmed requirement violations with report IDs and affected executions.
- Questions and ideas: Unresolved expectations and possible follow-up investigations, kept separate from confirmed findings.
- Scope reached: Actions, states, histories, or data classes actually exercised, with important omissions.
- Obstacles and effort: Setup problems, blocked work, time spent, and conditions that prevented a conclusion.
- Handoff: Decisions needed, responsible roles, and the next useful investigation or retest.
Use short linked entries during execution. Expand the details that support a finding or a difficult decision; a lengthy transcript can obscure why an observation mattered.
Completed session sheet S-01
The chronology above supplies the detailed investigation entries. Its handoff can be condensed into this record without copying every action again:
- Context:
S-01, charterC-01, buildB-734, isolated workspaceqa-lab-01, fixturework-items-v1. - Tester and time: Maya; November 20, 2026, 10:00-10:45 UTC. These are fictional session details.
- Starting state: Six records, four Active and two Archived, confirmed by
E-0. - Finding:
F-01, selection survives a West-to-East region change and remains actionable. Evidence connects the visible mismatch, submitted IDs, and saved effects. - Checks completed: Normal archive of two West records; West-to-East change after selecting two records; the same history after selecting one record; independent saved-state checks for each executed archive.
- Scope not reached: Status-filter changes, East-to-West and All-region histories, pagination, concurrent changes, other accounts, and other browsers.
- Obstacles: No missing prerequisite prevented this session. The untested status-filter path remains outstanding work, rather than a blocked test.
- Effort: Setup 5 minutes; testing and investigation 33 minutes; notes 7 minutes.
- Handoff: Feature owner to review impact and priority of
F-01; developer to investigate selection reset; QA to retest the fix and schedule the remaining filter histories.
The reported checks have useful boundaries. They concern one build, fixture, account, and direction of region change. The presence of a six-row dataset does not mean that six records were each tested through every relevant history.
Two more exploratory testing charter examples
These examples produce different kinds of information. A confirmed defect is one possible outcome; a missing rule or an unavailable prerequisite also affects what the team can conclude.
Charter C-02: clarify precision in a summary report
- Mission: Investigate whether a response-duration summary communicates consistent values across its dashboard and export, and identify precision decisions that the specification leaves open.
- Resources: A fixed three-event dataset with durations of 1,000, 1,000, and 2,000 milliseconds; the aggregation response; the dashboard; and its export.
- Known expectation: All three events contribute to the arithmetic mean. Their sum is 4,000 milliseconds, so the unrounded mean is 4,000 divided by 3 milliseconds, or about 1.333333 seconds.
- Open expectation: The specification has not defined the displayed precision for each output or whether those precisions must match. That is a question for the reporting owner.
- Scope and budget: The populated summary and its two outputs, with a 25-minute allocation. Empty groups, time-zone grouping, and large datasets are excluded.
An illustrative session note might record: "The aggregation response includes all three events and the expected unrounded mean. The dashboard displays 1.3 seconds; the export displays 1.33 seconds. Both are rounded representations of the same value."
That observation suggests the next step: clarify the output-specific precision rule before reporting the difference as a calculation defect. Record question Q-01 with the source values, units, actual displays, and the decision required. Do not copy the dashboard value into the export's expected result merely to make the outputs agree.
The session still has a concrete result: checked aggregation evidence and an unresolved presentation rule. If the response had omitted an event, the known inclusion requirement could support a defect report independently of the unresolved rounding decision.
Charter C-03: investigate import preview after a mapping change
- Mission: Investigate whether a changed column mapping reaches both the CSV preview and the data ultimately committed by the import.
- Resources: A two-row CSV fixture, an empty destination, a resettable workspace, preview and job evidence, and the product's declared import rules.
- Entry conditions: The preview worker can process a normal fixture, and the tester can inspect the mapping and saved records.
- Scope and budget: One valid file and a mapping change, with a 20-minute allocation. Broader file syntax and partial-commit behavior belong in the CSV import testing checklist.
Suppose the worker is unavailable during setup. The preview job remains queued, and an environment check confirms that its required worker is stopped. After eight minutes, the tester stops because the mapping behavior cannot be reached.
Record blocker B-01, the job identifier, the worker observation, and the unmet entry condition. The session has not exercised mapping changes or verified saved data. Assign the environment repair and preserve the charter for a new session when the prerequisite works.
An additional question about how gracefully the screen handles unavailable workers can become a separate investigation. Its expectations and evidence should be identified explicitly instead of turning this setup problem into an unsupported verdict about mapping correctness.
Turn the debrief into concrete next actions
A debrief reviews what the session established, what changed in the tester's understanding, and what the team should do next. Walk through the evidence and decisions rather than reading every note aloud.
For S-01, the discussion should resolve:
- Impact: Which user work could be affected when hidden records remain selected? The feature owner determines priority from the product context.
- Confidence: Which history reproduced the finding, which saved effects were observed, and which implementation explanations remain hypotheses?
- Remaining scope: Who will check status changes and other region histories, and whether that work is needed before the release decision?
- Repair and verification: What evidence should demonstrate that the selection rule is restored and the normal action still works?
- Follow-up learning: Does the new shared toolbar affect other lists with different selection policies?
For Q-01, the next action is a product decision with a named owner. For B-01, it is restoring the worker and rescheduling execution. Giving each record the same status would hide these different dependencies.
Keep the original session record when a finding is fixed. Link the later retest to the new build so readers can distinguish the behavior observed during exploration from the behavior verified after the change.
Choose what to preserve as regression
For F-01, a focused regression can start with the restored fixture, select W-101 in Active and West, switch to East, and assert selection count 0, disabled Archive, and no archive submission caused by the context change. Then select W-103 in East and archive it. Only W-103 should change; W-101 should remain Active. This continuation checks that the old selection was cleared from the action state as well as from the display.
Choose the layer that observes the risk. A component check can exercise the reset and enabled state. An integration check can connect a permitted archive to exact requested IDs and saved effects. Keep a user-journey check when it contributes evidence across boundaries the narrower checks miss.
After the reporting owner resolves Q-01, stable precision examples can become repeatable assertions. A fixture-health check may make B-01 easier to detect before the next session. These are different follow-ups to different discoveries.
The regression testing checklist helps place those checks in the affected release scope. Preserve the original learning and its evidence as well as the resulting test; later maintainers need to know which requirement the assertion protects.
Describe coverage without inventing a percentage
Keep the unit of coverage explicit. An action reached, a record checked, a rule verified, and a transition exercised describe different aspects of the session. Several screenshots of the same action do not necessarily add a new coverage item.
If a formal model exists, map the relevant observations to its defined items and criteria. Otherwise, report the tested histories, contexts, and assertions directly. One completed charter cannot establish a percentage of all possible product behavior.
Repeated sessions are useful when they extend the evidence or revisit a changed risk. Before repeating a charter, decide whether to add another history, dataset, environment, perspective, or assertion. Repeating the same narrow path should be reported as repetition of that path.
FAQ
Yes. One practical arrangement is for one person to operate the product while the other records observations, challenges assumptions, and proposes follow-up checks. Swap roles when useful and retain one shared session record. Identify both participants so a later reviewer knows how the evidence was collected.
Automation can prepare fixtures, query saved state, capture requests, and replay stable checks. Those capabilities support exploration. When describing an automated workflow, explain how observations influence the next test choice; a fixed replay alone does not demonstrate that adaptation. Evaluate the decisions and observations it actually performs rather than classifying the work only by its tool.
Start with prioritized learning objectives and a budget for each feasible session. Include expected setup, investigation, and reporting work. Use an initial session to learn about fixture costs and obstacles, then revise the remaining allocation. Avoid estimating by a promised number of defects: a useful session may instead resolve an uncertainty, expose a missing requirement, or identify a prerequisite that blocks further testing.