1. LandOffer
  2. Blog
  3. All articles
  4. Manual QA Portfolio: Turn an Exploratory Session into Reproducible Bug Reports

Job search

Manual QA Portfolio: Turn an Exploratory Session into Reproducible Bug Reports

Turn exploratory observations into reproducible bug reports, then document the correction, retest and remaining coverage gaps.

9 min readLandOffer team

A manual QA portfolio should let another person reproduce a problem and understand how you checked the correction. Keep the exploration goal, starting state, exact actions, expected behavior, actual observation and retest in the same case record. A collection of screenshots with “broken” captions gives a reviewer much less to work with.

We created and exercised an owned local teaching sample on October 22, 2026. Its two defects were deliberately seeded; this is a directed exploratory walkthrough, not a blind discovery study or client engagement. Browser actions, rendered text and screenshots were recorded. LandOffer publishes this guide and offers recent technical roles; choose evidence that matches the role’s actual testing responsibilities.

A hand-operated printing press, proof sheets, magnifying glass and notebook illustrate repeating an action and recording its observed result.

Give the exploration a bounded mission

A useful charter identifies the area, risk and observations you want to gather. For our sample: “Explore the quantity estimate and catalog reset to find disagreement between the stated rules, selected controls and displayed results.” The teaching page states its requirements directly, so expected behavior has an explicit source.

The sample has no account, persistence, external resources, real orders or payment. It contains a quantity estimate priced at $12 per unit and a three-item catalog. Requirement Q1 permits whole-number quantities from 1 to 10 and requires invalid input to preserve the previous total while showing an error. Requirement F1 says Reset filter must select All and display all three items.

Probe a boundary quantity and a valid quantity, then compare the filtered view with its reset state. Record questions encountered during exploration, then isolate each reproducible issue into its own report. Do not treat a charter as proof that every possible state was covered. A follow-up investigation can address different boundaries or sequences without enlarging the original session’s claims.

Mozilla’s bug-writing guidance recommends separate reports for separate issues, precise reproduction steps and distinct expected and actual results. It also distinguishes observations from speculation. We adapt that reporting discipline to our own fixture; we did not test Mozilla’s software or file a bug with any vendor.

Keep the starting state and environment reproducible

Preserve the original sample before changing it. Our files are fixture-a.html, containing the seeded defects, and fixture-b.html, containing the corrections. Both are self-contained HTML with embedded JavaScript. Reloading either restores its starting values because the sample does not save state.

The browser walkthrough used Chrome through native browser-control tools at a recorded content viewport of 1920 × 873. We captured full accessibility snapshots, rendered body text, selected filter values and four screenshots. Exact Chrome build and operating-system versions were not collected, so exact-version reproduction remains a documentation gap. No private profile, extensions or browser settings were changed.

MDN’s minimal-test-case guidance recommends a small standalone reproduction without unrelated dependencies. It also warns that unexpected behavior can originate in application code, documentation or the browser. These defects are intentional application-logic faults. They are not evidence of a Chrome bug.

Use fresh loading when that is part of the reproduction. If a report depends on an earlier selection, make that selection explicit rather than describing a vague “reset problem.” Preserve the exact fixture version and evidence alongside the report, so an accidental fix does not erase the original behavior.

Completed report one: zero quantity replaces the valid estimate

Issue MQ-01 — “Zero quantity produces a successful estimate instead of an error.” Component: quantity estimate. Version: owned fixture A. Requirement reference: Q1, displayed on the page. The impact is an invalid estimate in a teaching widget; no money was charged or order created.

Starting state: freshly load fixture A. Quantity is 1, Total is $12 and status is Ready. This starting total matters because the invalid-input rule requires preserving it. A report that omits the previous value cannot fully describe that failure.

Exact reproduction:

  1. Replace the Quantity field’s value with 0.
  2. Activate Calculate estimate.
  3. Read the total and status without reloading.

Expected: Q1 rejects zero, displays an error and leaves Total at $12. Actual: the page displays Total: $0 and Calculated 0 units, treating the invalid value as a successful calculation. We repeated the same sequence after a fresh reload and observed the same result on both attempts.

The retained screenshot shows the entered zero and the resulting text together. The rendered-text log and full accessibility snapshot preserve the observation without relying on the screenshot alone. The two successful reproductions describe this tiny sample and route, not a defect rate among real users or a reliability estimate.

Investigation note: fixture A converts the input to a number and updates the result without checking the stated range. That code inspection explains the seeded mechanism. The initial report remains behavior-focused; “add validation” is a proposed remedy, not the observed symptom or the entire expected result.

Correction in fixture B: reject values that are not integers or fall outside 1–10 before updating the total. We retained the error message and returned without overwriting the previous result. This is a change to our owned teaching file, not a patch submitted to a third-party repository.

Actual retest: load B fresh, enter 0 and activate Calculate estimate. The page displays Quantity must be a whole number from 1 to 10, while Total remains $12. Then enter 3 and calculate: Total becomes $36, with Calculated 3 units. These observations support the corrected zero-input path and one nearby valid-input check.

Finished disposition: The observed zero-input sequence in B meets Q1’s error-and-preserved-total expectation, and the valid-3 control still calculates correctly. Blank input, decimals, negative numbers, values above 10 and combinations after previous edits were not exercised in this browser session. The presence of a broader code guard does not convert those unexecuted paths into passed tests.

Convert an observation into a report someone can use

Write the smallest complete sequence that preserves the failure. Keep a fresh start, necessary data and the exact control names. These details let a developer attempt the issue directly, without guessing which earlier selection or input made the failure possible.

Keep evidence and interpretation separate. “The selector shows All while only Pencil remains” is an observation. “The reset handler probably forgot to render” is an explanation to investigate. A report can include both, but the explanation should not replace the observable mismatch.

Microsoft’s exploratory testing guide describes capturing notes, screenshots and actions, then creating bugs with that context. Its connected extension also requires project access and permissions. We used no extension or Azure account here; the relevant principle is preserving the session evidence with the finding, not adopting a particular paid workflow.

Do not invent a company’s severity or priority decision. State the concrete affected task and whether a workaround was actually observed. In a real team, discuss ordering with the people who own the product and release. An intentionally defective local sample has no measured customer impact, revenue loss or production release risk.

Completed report two: reset changes the selector but leaves the list filtered

Issue MQ-02 — “Reset filter selects All but leaves only the Tool result visible.” Component: catalog filtering. Version: owned fixture A. Requirement reference: F1. The three teaching items are Notebook and Folder in Paper, and Pencil in Tool.

Starting state: reload A. Category displays All; the page shows Showing 3 of 3 and all three item names. The quantity section is unrelated to this reproduction and need not be edited.

Exact reproduction:

  1. Select Tool in Category.
  2. Confirm Showing 1 of 3 and Pencil.
  3. Activate Reset filter.
  4. Compare the selected category, count and item list.

Expected: Category returns to All; count returns to Showing 3 of 3; Notebook, Pencil and Folder are visible. Actual: Category returns to All, but the count remains Showing 1 of 3, with only Pencil visible. The selected DOM value is also all, so this is not merely an unclear label. A second fresh-load reproduction showed the same mismatch.

Two small wooden latch models, a key and a notebook illustrate comparing a control’s visible position with the mechanism it actually operates.

Investigation note: the seeded reset handler changes the selector’s value but does not call the function that redraws the list and count. The reset handler in A lacks the render call used after a category change, which explains why the selector updates while the retained list does not. It does not establish that every reset defect in another application has the same cause.

Correction in B: call the existing render function after setting Category to All. Keep the original item data unchanged. This targeted change allows the same selection, count and list comparison to establish whether the stated requirement is restored.

Actual retest: in B, select Tool and observe Pencil with Showing 1 of 3. Activate Reset filter. The recorded page shows Category All, Showing 3 of 3, and Notebook, Pencil, Folder. The selected DOM value is all. The report therefore passes the specific reset-path retest, with the preceding filter action also observed as a control.

Finished disposition: MQ-02’s reported path passes in B. We did not test Paper-to-All, repeated resets, keyboard-only operation, other browsers, smaller viewports or assistive technology. We also did not establish that changing the filter preserves focus. Those are possible follow-up charters, not hidden completed checks.

Separate confirmation from wider regression coverage

The two reports show a full chain: stated rule, reproducible mismatch, retained evidence, owned correction and observed retest. The chain lets the reviewer inspect each reported path; the screenshot count and two findings do not establish coverage beyond those paths. Because the faults were seeded, the exercise demonstrates reporting and verification, not independent defect discovery on an unfamiliar product.

A retest asks whether the reported failure is corrected. Wider regression work asks whether the change harms other behavior. We checked one valid quantity and the Tool-filter action near the repaired paths. That small scope cannot establish complete regression coverage, accessibility compliance or readiness for production.

Record Observation retained B retest boundary
MQ-01 A accepts zero and replaces $12 with $0 on two fresh-load attempts Zero rejected with prior total retained; valid 3 produces $36
MQ-02 A selects All but retains Pencil and count 1 of 3 on two fresh-load attempts Tool selection and reset restore the expected count and three items
Environment Chrome, 1920 × 873 content viewport; version and OS details missing Same browser surface; no cross-browser claim

A portfolio can later add automated checks, but that is another evidence layer. An assertion script inspecting saved output verifies the record’s consistency; it does not repeat the browser session or become a production test suite. Label the actual method rather than calling every script a manual test or every browser action a full automation framework.

Package the finding and its remaining questions

Keep source versions, a short charter, original observations, separate bug reports and retest notes together. Remove private URLs, credentials and customer content before sharing permitted material. Use synthetic data or an owned fixture when a real employer’s system cannot be disclosed.

Start the portfolio case with its tested version and the shortest complete sequence another person can reproduce. Explain the correction only as far as your evidence supports, and retain the gaps that shape the next session. The next useful question might concern boundary inputs, reset sequences or focus behavior; it need not become an invented claim that everything now works.

As you review recent LandOffer roles, choose a case that shows careful observation, clear communication and an honest retest. A small, inspectable report is useful when its reader can follow the evidence and see where coverage ends.

Sources and Further Reading

Public primary sources checked October 22, 2026. Only the owned local teaching fixtures were exercised; no third-party software was tested or issue submitted.