Revisions & change control
Reopening an Eight-Year-Old PCB Design: A Forensic Order of Work
Rank your evidence before you read any of it, start from a physical board, ask who kept copies, and write down what you could not establish.
A customer wants 200 more of a product you last built in 2017. The engineer who designed it has gone. There is a folder, a few boards in a drawer, and an invoice from a fabricator who may or may not still exist.
You have to establish what this product actually is before you can build another one. That is a forensic exercise, and like any forensic exercise it has an order: start with the evidence that cannot lie, and work towards the evidence that can.
Rank your evidence before you read any of it
Not all sources are equally trustworthy, and the usual instinct — open the newest file — starts at the wrong end.
| Source | Trust | Why |
|---|---|---|
| A physical board | Highest | It cannot be edited. It is what was built |
| Fabricator's archive | High | What they received, not what you meant |
| Purchase records | High | Independent, dated, hard to fake |
| Assembler's documentation | High | Includes deviations you may not know about |
| Released output files | Medium | Trustworthy if identifiable; often several candidates |
| Design source files | Medium | May contain unreleased work in progress |
| Folder and file names | Low | final_v2_USE_THIS means nothing |
| Anyone's memory | Low | Confident and frequently wrong |
The two counter-intuitive entries are the last ones. File names and recollections feel like evidence and are mostly noise; treat them as hypotheses to be checked, not facts.
Reading a board
Start with a physical unit, because it settles questions nothing else can.
Read the silkscreen. Part number, revision, date code, manufacturer logo. This is the single most valuable piece of information in the whole exercise, and it takes thirty seconds.
Photograph both sides at high resolution, before anything gets desoldered or lost. You will refer to these images repeatedly.
Inventory the fitted parts. Reference designator, marking, package. Note which footprints are empty — the DNP set is part of the configuration and is rarely written down anywhere else.
Look for rework. Bodge wires, lifted pins, added components, scraped soldermask. Rework on a production unit means the released design was wrong in some way that somebody fixed in the field, and whatever they did is probably not in any file.
Count the layers, if you have a sacrificial unit and no stackup documentation. Cross-sectioning a board is destructive and sometimes the only way to recover a stackup.
Ask the people who kept copies
This step is frequently skipped and is often the highest return per hour spent.
The fabricator keeps the data package you sent. Ask for exactly what they hold for your part number, with their internal revision. They may also hold the stackup they actually used, which is worth more than your intent.
The assembler keeps the BOM and placement data they worked from, and frequently a list of substitutions and deviations agreed by telephone that exist nowhere in your records.
Your distributors can produce purchase history by part number, which lets you reconstruct what was bought and in what quantity, and therefore roughly what was built and when.
Expect some of these to have purged their archives. Ask anyway; the cost is an email, and one hit saves days.
The order of reconstruction
- Establish identity. Part number and revision, from the board.
- Find the evidence set for that specific revision — not the newest
files, the matching ones.
- Reconstruct the BOM from the board, and compare it with the BOM you
found. Differences are the interesting part: they are either rework, substitutions, or a file that was never the released one.
- Verify the fabrication data against the physical board. Does the
outline match? The connector positions? The layer count?
- Check the parts are still available. Usually several are not, which
turns a reproduction job into a redesign job — better to discover this now than after you have spent a week on documentation.
- Only then open the design source. By now you know which file should
match, and you can treat a mismatch as information rather than confusion.
Write down what you could not establish
This is the step that distinguishes a reconstruction you can rely on from one that quietly misleads the next person.
For every question you could not answer, record the question and what you tried. "Stackup not recovered; fabricator no longer holds records; assumed four-layer 1.6 mm from board thickness and measured impedance — not confirmed."
An honest gap can be closed later by someone with better evidence. A guess written as a fact cannot, because nobody afterwards knows it was a guess. In a reconstruction, the inventory of unknowns is part of the deliverable.
When to stop reconstructing and redesign
Reconstruction is not always the cheaper path. Consider stopping when:
several critical parts are obsolete with no drop-in equivalent; no fabrication data can be found and the board is dense; the design depends on a tool version you cannot run; compliance has moved and the product would need requalifying anyway; * the volume is large enough that a modern design pays for itself.
The honest comparison is between reconstruct and build 200 and redesign and build 200 plus everything after that. Teams routinely over-value reconstruction because the redesign cost is visible and the reconstruction cost is not.
Close the loop
Whatever you recover, the exercise is wasted if it ends in a folder on your machine. Before moving on, establish the identity properly for next time: one record of what this product is, at which revision, with the files that match it and the list of what remains unknown.
The next person to be handed this product in eight years is probably you.
A reconstruction can only recover information that still exists somewhere — in a board, an archive, a purchase record. Nothing recovers a decision that was never written down, which is the argument for recording it the first time.
A board, its BOM and its documents are stored in XKEMA as versions with an author and a date, so a past configuration is looked up rather than reassembled from evidence.
Legacy redesign Hardware project handover * Hardware revision control: a practical guide