A single source of truth in manufacturing means every report, screen, and meeting pulls the same number from the same live record, captured once at the source. When the ERP says 4,180, the spreadsheet says 4,312, and the whiteboard says 4,250, you do not have one; you have three.
Every plant has lived this meeting. Friday's output comes up. The ERP report says one number. The supervisor's spreadsheet says a second. The whiteboard by Line 2, photographed on someone's phone, says a third. Twenty minutes go to arguing about which number is real, nobody fully wins, and the actual question, why was Friday short, never gets asked. A single source of truth is the fix for that meeting, and this post covers where the three numbers come from, why copying data into one place does not solve it, and how plants actually get to one number.
What is a single source of truth in manufacturing?
It is an operational state where each fact about the plant, a count, a stop, a test result, an inventory level, has exactly one authoritative record, and everything else references that record instead of maintaining its own copy. The output of Line 2 for first shift is captured once, at the line, when it happens. The floor display shows that record. The daily report aggregates that record. The ERP transaction posts from that record. Change it in one place, for a good reason, with a trace, and every view changes together.
The test is simple: pick a number that matters, output, downtime, scrap, and ask three different roles for it. If you get one answer, you have a single source of truth for that number. If you get a negotiation, you do not. Most plants pass for financials and fail for operations, because operational data is the data that gets captured twice, retyped, and summarized by hand on its way through the classic manufacturing data silos.
Why do the ERP, the spreadsheet, and the whiteboard disagree?
Because they are not three views of one record. They are three separate records of one event, and they forked at four predictable points.
Capture forks. The operator tallies on the whiteboard through the shift, the machine's counter counts every cycle including the rejects, and the ERP gets whatever quantity is typed in at the end. Three measurements of the same production run, none wrong by its own definition.
Retype forks. Every manual transfer, paper to spreadsheet, spreadsheet to ERP, is a chance for transposition, omission, and rounding. A plant that retypes data has chosen to have multiple sources of truth; the errors are just the interest on that choice.
Definition forks. Does "output" mean gross or net of scrap? Does the shift end at 6:00 or when the last pallet is wrapped? Does a changeover count as downtime? The ERP, the spreadsheet, and the whiteboard each answer these questions slightly differently, so even perfect data entry produces different numbers. This is why standard KPI definitions like OEE calculation matter more than they look.
Timing forks. The whiteboard is current to the hour, the spreadsheet to the shift, the ERP to whenever transactions post. Compare them at any given moment and they disagree even when they will eventually agree.
What does the disagreement actually cost?
The visible cost is reconciliation labor: supervisors and clerks spending the first hour of every day retyping and cross-checking, the swivel-chair work described in manufacturing data silos. The larger costs are quieter.
Meetings argue about numbers instead of causes. Every minute spent litigating which figure is real is a minute not spent on why the line was short. In plants with chronic disagreement, the argument becomes the meeting.
Decisions wait for reconciliation. When leadership does not trust the operational numbers, decisions get deferred until someone "checks the real number", which usually means another day. The plant runs on a lag not because data is missing but because it is contested.
Trust drains downward. Operators watch their carefully recorded tallies get overridden by an ERP figure they know is wrong, and conclude that accuracy is optional. The quality of source data falls to meet the respect it receives, which is the cultural failure covered in plant floor transparency.
Is a single source of truth just a central database?
No, and this is where many projects go wrong. Copying everyone's numbers into a warehouse gives you a single location for your disagreements, not a single truth. If the whiteboard, spreadsheet, and ERP still capture independently, the warehouse just stores all three forks side by side, hours later.
The real fix is to reconcile at the source: capture each event once, digitally, where it happens, with one agreed definition, and let every downstream system read or post from that record. The whiteboard becomes a display of the record rather than a competing record. The spreadsheet dies because there is nothing left for it to do. The ERP receives transactions from the source instead of a typed summary. In ANSI/ISA-95 terms, the fix lives at Level 3, the operations layer between the machines and the business systems, which is precisely the ground an MES occupies; see what is an MES for that map. And for the definitions themselves, ISO 22400-2 provides 34 standard manufacturing KPI definitions so that "availability" or "OEE" is not something each department invents; adopting them removes the definition fork in one move.
How do you build a single source of truth?
- Pick one number and follow it backward. Take last Friday's output and trace every place it is recorded, whiteboard, spreadsheet, ERP, report deck. Each copy you find is a fork to eliminate.
- Agree the definitions in writing. Gross or net, shift boundaries, what counts as downtime. One page, signed by production, quality, and finance. Steal from ISO 22400 rather than inventing.
- Capture once, at the source, digitally. Machine signal or station tablet, at the moment of the event. This is the step that kills the retype fork, and it is the heart of a paperless factory.
- Make every display a view. Floor screens, daily digital production reporting, and management rollups all read the same records. If a screen needs its own spreadsheet to exist, it is not done.
- Post to the ERP from the record. The ERP stays the system of record for orders and financials; it just stops being fed by hand. No rip-and-replace, different wiring.
- Retire the shadow copies deliberately. The parallel spreadsheet must be turned off, publicly, once the layer earns trust. Every surviving shadow copy is a standing vote of no confidence.
How does Harmony AI deliver one source of truth?
This is one of Harmony AI's founding commitments, stated on our homepage in six words: one source of truth, the same number in every report. Harmony AI is an AI-native MES, and the mechanism is exactly the source layer described above. We digitize capture at the station, connect the machines, and ingest the ERP, QMS, and SOPs you already run, along with the tribal knowledge your senior people carry, into one live model. Every app, screen, and report is a view of that model, and when the AI answers a question it cites the record it drew from, so a disputed number can be traced to its source in seconds instead of relitigated in meetings.
Honest scope note: getting there is partly organizational. Definitions have to be agreed, and shadow spreadsheets have to actually be retired, and no vendor can sign that page for you. What we can say is that the wiring works and the meetings change, which is what we watched happen while unifying live plant data at CLS. From there, one truth in one plant becomes the foundation for comparing many, the subject of real-time visibility across multiple plants, and for trusting the live numbers day to day, covered in what real-time manufacturing visibility means.