Originally published at bughuntlab.hashnode.dev

A defect report is not a complaint. It is a reproducible argument that a specific behaviour is wrong. Here is the structure that survives triage.

Most bug reports die in triage. Not because the bug was fake, but because the report gave the developer nothing to act on. "Checkout is broken" is not a bug report. "Clicking Apply on a discount code silently fails when the cart contains a gift card" is.

The difference is not writing talent. It is structure.

The six fields that matter

Every defect report needs these, in this order. If you can only fill in four, fill in these four.

1. Title — the observable wrong behaviour, not the suspected cause

Bad: Cart issue. Better: Discount code rejected when a gift card is in the cart.

Never write Fix the reducer in a title. You do not know that yet, and if you are wrong the developer stops trusting your reports.

2. Component / area — where in the product

Cart, checkout, search, payments, delivery. Be specific enough that the ticket routes itself.

3. Classification — what kind of defect it is

This is the field most juniors skip, and it is the one that speeds up triage the most.

Class What it means Example
UI Renders wrong, but behaviour is correct Price shows £12.500 instead of £12.50
Functional Correct-looking UI, wrong behaviour Discount applies but total does not recalculate
Validation Accepts or rejects input it shouldn't Quantity field accepts -3
Network Request or response is wrong POST /cart returns 200, but the item is not persisted

A UI bug and a network bug can look identical on screen. Naming the class tells the developer whether to open the component or the API.

4. Severity — you assign it, and you defend it

Not everything is critical. A misspelled button on the marketing page is not blocking a release; a total that miscounts by a penny is, because it is a money bug and money bugs are legal bugs.

5. Steps to reproduce — numbered, from a known starting state

Start from a state someone else can reach. "Logged in as a returning customer with one item in the cart" beats "somewhere in checkout".

6. Expected vs actual — two sentences, no more

Expected: applying code SAVE10 to a £40 cart yields £36. Actual: the total stays at £40 and no error is displayed.

The fields that make it good rather than acceptable

  • Evidence. A screenshot is fine. A HAR file or a request/response pair is better, because it tells the developer whether the bug is in the browser or the server.
  • Environment. Browser, version, OS, and build. A bug that only reproduces in Safari 17 on mobile is a different bug from one that reproduces everywhere.
  • Frequency. Always / intermittent / once. "Intermittent" is information, not a cop-out — but say how many attempts you made.

A template you can copy

Title: Discount code rejected when a gift card is in the cart Component: Cart / promotions Class: Functional Severity: High (money incorrect, no workaround) Environment: Chrome 131, macOS 15.1, build 2026.11.18 Preconditions: Logged in. Cart contains 1x "Northstar Mug" (£40.00) and 1x gift card (£10.00). Steps: 1. Open /cart 2. Enter discount code SAVE10 3. Press Apply Expected: Total becomes £26.00 (10% off £40, gift card applied separately) Actual: Total stays at £30.00. No error shown. Network tab shows 200 on POST /cart/discount. Evidence: [screenshot] [HAR] Frequency: Always (5/5 attempts) 

That report takes about four minutes to write and saves a developer twenty.

The failure mode nobody warns you about: false positives

Here is the uncomfortable part. A report that describes a behaviour that turns out to be intended is worse than no report at all. It costs the team a triage cycle, it costs you credibility, and it teaches reviewers to skim your tickets.

Before you file, ask one question: could this be working as designed? If there is a real chance, write it as a question in the team channel first, not as a defect. Then, if it turns out to be a defect, file it properly — with the extra context you just gained.

Serious test teams score this explicitly. Treating a false positive as a real cost, rather than a harmless over-report, is what separates a tester who gets trusted with the important areas from one who gets their tickets deprioritised.

Practising the skill

Reading about report structure does not make you good at it, any more than reading about swimming does. You need reps against something with a known answer.

That is the specific problem BughuntLab was built to solve: four working sandbox products (a storefront, a bank, a crypto wallet, a trading desk), each with five deliberately seeded faults, and your written report scored against a hidden answer key. The scoring is deterministic — component, classification and severity either match or they do not — and a false positive costs points, which is exactly the pressure a real triage meeting applies.

There is a free walkthrough if you want to see the report format before committing to a challenge.

Disclosure: I build BughuntLab. The template above is the one it scores against, which is why I am confident it is the right shape — but the advice stands on its own whether you use the tool or not.