← Journal Testing · · 7 min

A bench report a founder can actually read

How we structure a bench report so a founder can act on it from page one: the summary, the fix list, and the evidence behind every finding.

Milo Perović
Founder & lead engineer

  • Pre-cert
  • QA
  • Reports
A bench report a founder can actually read

Most test documentation is written for the engineer who ran the test. It is long, chronological and full of oscilloscope captures, and the person who has to decide what happens next — usually a founder or product manager — reads the first page, skims the rest and still does not know whether the product is ready. A good bench report fixes that by putting the decision first and the evidence second.

This post is for founders and PMs who commission hardware testing, and for engineers who write the reports. It describes the format we use at the end of the Exact phase, why it is shaped the way it is, and the mistakes that make test results hard to act on.

What a bench report is for

A test report has one job: to let someone who was not at the bench make a correct decision. That decision is rarely “pass or fail”. It is usually one of a handful of questions: can we build the next batch, do we need a board respin, is it safe to start pre-certification, and what will the fixes cost in time?

If a report cannot answer those questions on its first page, it has failed its reader, however rigorous the testing behind it. Rigour belongs in the appendix, where it can be checked. The front of the document belongs to the decision.

The one-page summary

Page one is the only page we can guarantee will be read, so it carries everything needed to act. We keep it to a fixed structure so readers know where to look from report to report:

  1. Verdict in one sentence. For example: “Ready for a second prototype batch after two layout changes; not yet ready for pre-certification.”
  2. What was tested. Hardware revision, firmware version (a commit hash, not “latest”), number of units and the date range.
  3. Blocking issues. Anything that stops the next step, each in one line with a reference to its fix.
  4. Open risks. Things we have not been able to rule out, stated plainly, with what it would take to close them.
  5. Next step and rough effort. What we recommend doing now, and whether that is days or weeks.

The revision and firmware details matter more than they look. A result without a traceable build is close to worthless six weeks later, when a second batch behaves differently and nobody can say which board or binary produced the original numbers.

Why the fix list matters more than the pass/fail table

The traditional centre of a test report is a table of requirements with a tick or a cross next to each. It is useful as a record, but it tells the reader very little about what to do. A cross next to “sleep current below 20 µA” says nothing about whether the cause is a pull-up resistor, a regulator’s quiescent current or firmware that never enters deep sleep — and those have very different costs.

So the second page of our reports is a fix list, ordered by impact. Each entry follows the same shape:

  • Finding: what we observed, with the measured value and the expected value.
  • Cause: confirmed, or our best hypothesis clearly labelled as such.
  • Proposed fix: the specific change — a component value, a layout note, a firmware patch.
  • Type: firmware-only, BOM change, or layout change requiring a respin.
  • Verification: how we will know the fix worked, and which test to rerun.

The “type” field is the one founders tend to value most. A firmware-only fix can often ship the same week; a BOM change touches procurement; a layout change means new boards, typically a couple of weeks for prototype quantities, depending on stack-up and assembly. Grouping fixes by type turns a list of problems into a schedule.

Hypotheses are allowed, if they are labelled

Not every root cause is found within a test window. That is normal. What is not acceptable is presenting a guess as a conclusion. We mark each cause as confirmed (reproduced and removed by the fix), likely (consistent with the evidence, not yet proven) or unknown. A reader can plan around uncertainty; they cannot plan around false confidence.

Writing measurements so they can be trusted

Behind the summary and the fix list sits the evidence. This is where most reports either bury the reader or leave out the details that make a number meaningful. For every measurement we record:

  • the instrument and its relevant setting — for current, the range and sample rate, because a meter averaging over a second hides the transmit bursts that a power analyser or source-measure unit will show;
  • the test conditions — supply voltage, ambient temperature, firmware state, what was connected;
  • the number of units and the spread, not only the best one;
  • a pass criterion that was agreed before the test, not chosen after seeing the result.

The last point is the one most often skipped. If limits are written after the data, every result becomes negotiable. Agreeing them in the Abstract phase keeps everyone honest and makes the verdict on page one defensible.

Spread across units deserves its own mention. Testing a single board tells you about that board. With even three to five units, you start to see whether a marginal result is a design problem or one bad joint. We report the worst unit alongside the typical one, because the worst unit is closer to what a customer will eventually receive.

Layout of the full bench report

Put together, the structure looks like this. Most of our reports land between a handful of pages and a few dozen, depending on scope, but the order never changes:

SectionReaderPurpose
One-page summaryFounder, PMDecide the next step
Fix listFounder, engineersPlan and schedule the work
Results tableEveryoneRecord against agreed requirements
Test detailsEngineersMethod, setup, conditions
Raw data and capturesEngineers, auditorsReproduce and verify

Environmental results — temperature cycling, humidity, drop — follow the same pattern and slot into the results and details sections. We wrote separately about running environmental testing on a small-batch budget; the reporting format does not change because the chamber is borrowed.

Common ways test reports go wrong

Most of the reports we are handed when taking over an existing project share the same problems. None of them are about the quality of the engineering; they are about communication.

  • Chronological order. The report reads like a lab diary, so the most important finding sits on page fourteen.
  • No build traceability. “Rev B” with no firmware version, or a firmware version with no record of which boards it ran on.
  • Screenshots without context. A scope capture with no timebase, probe setting or explanation of what the reader should see.
  • Problems without fixes. A list of failures that leaves the reader to work out the remedy and its cost.
  • Silent omissions. Tests that were not run simply do not appear, so the reader assumes they passed.

The last one is the most dangerous. A report should list what was not tested as clearly as what was. “ESD not tested in this round” is a perfectly good line; leaving it out invites a nasty surprise at the certification lab.

A checklist before you send a bench report

Whether you write reports or receive them, these are the questions we check before one goes out:

  1. Can someone answer “what do we do next?” from page one alone?
  2. Is every result tied to a hardware revision, firmware commit and unit count?
  3. Does every failure have a proposed fix, a fix type and a way to verify it?
  4. Are causes marked as confirmed, likely or unknown?
  5. Were pass criteria agreed before the test?
  6. Is the worst unit reported, not only the best?
  7. Are untested items listed explicitly?

If the answer to all seven is yes, the report will do its job even if the reader never gets past the second page. Testing and reporting are part of how we work across every project; you can see where they fit alongside design and firmware on our services page.

Working on something similar?Free scoping call →