| Document | CS-04 · acceptance protocol |
| Revision | A · 2026-10 |
| Sheets | 7 |
Test summary · one reference project · Modelled, inputs on sheet 7
| Item | Result | Criterion | Per project | Verdict | |
|---|---|---|---|---|---|
| T-01 | −60% | Engineer-days spent collecting performance-run evidence from six makers' screens | €14,400 | Modelled | |
| T-02 | −1 run | Repeated performance runs per project, from 1.5 on average to 0.5 | €9,600 | Modelled | |
| T-03 | −3 wks | Performance run to signed handover, from ten weeks to seven | €24,000 | Modelled | |
| T-04 | −40% | Engineer site visits during the 12-month warranty period | €9,200 | Modelled | |
| Total | €57,200 | Per reference project · €686,400 a year at 12 projects of this size | Modelled |
Each maker's machine passes its own factory test. At the site's performance run, the integrator proves the line from six exports that disagree about the time, and the items between two makers hold the handover open for weeks.
A data clause in every purchase order, a baseline at every factory test, and one read-only record on site from the first site test to the last day of warranty. A named engineer approves every punch attribution.
The customer class is the automation business of a global industrial-automation group. It sells projects: it designs a production line, specifies and buys the machines, writes the line control, commissions the site, hands it over to the plant owner and stands behind it through the warranty period.
On most of those projects, most of the machinery is not its own. Machining centres, robot cells, washers, leak-test benches, press-fit and marking stations come from specialist makers, each with its own control, operator screen, clock and export.
The integrator's own control and software run its own scope of supply, and nothing in this case study changes that. What a mixed site also needs is a neutral record of the machines the integrator did not build: one that reads every maker's machine the same way, takes no side when a line stops, and stays with the site after handover.
That is the job Pragmatos 3.0 does here. It reads the other makers' machines read-only, over OPC UA or MTConnect, into one namespace on the plant owner's own servers, from the first factory acceptance test to the last day of warranty.
Reference project · hall layout, one line shown of two identical lines
Eighteen machines per line, two lines, six makers. Letters A–F are the six other makers; each machine keeps its own control and operator screen. The integrator's conveyor and line control join them. Everything lettered is another company's machine, specified in the integrator's purchase order and accepted at that maker's works before it ships to site.
On a mixed site, each machine passes its own maker's test. The trouble starts between machines: a line stops, and six screens each show their own machine as ready. Below is the kind of punch list such a site carries after its performance run.
| Item | Location · machine | Maker | Observation | Cat | Status |
|---|---|---|---|---|---|
| P-017 | Line 1 · washer to robot cell | C / B | Line stops six times a shift at the transfer. Both screens show their own machine ready. | A | Disputed |
| P-023 | Line 2 · machining centre 4 | A | Cycle 8% over contract in the performance run. Maker reports starvation from upstream. | A | Disputed |
| P-031 | Line 1 · leak-test bench 2 | D | Rejects counted twice: the bench log and the line counter differ by 4%. | B | Disputed |
| P-038 | Line 2 · press-fit station 3 | F / B | Interlock wait after robot unload. Neither side logs which machine held the other. | A | Disputed |
| P-044 | Line 2 · robot cell 1 | B | Fault history lost at power-down. No record left to support a warranty claim. | B | Open |
| P-052 | Line 1 · marking station | E | Station clock eleven minutes off the line clock. Events cannot be lined up. | B | Open |
| P-060 | Handover file | all | Six exports in six formats. No single record of the performance run exists. | A | Open |
Items between two makers. Open items at the performance run that sit at a transfer, an interlock or a shared count.
To agree who owns one. Working days, on average, before a disputed item reaches the maker who will fix it.
Screens, clocks, exports. One of each per maker, and none of them was built to agree with the others.
What the dispute costs. Every disputed Category A item holds the handover. The commissioning team stays on site, the plant owner holds back its acceptance, and the warranty period begins with arguments that nobody can settle from the data, because the data sits on six screens that disagree about the time.
Each operator screen reports its own machine. None of them shows the machine next to it, so a stop at a transfer is always the other maker's.
Each control keeps its own time. Without one clock, nobody can say which machine stopped first, and the order of events is what decides the owner.
Fault histories are kept by the engineers who were on site. When they leave, the warranty period runs on memory and on the maker's own account of what happened.
The project Gantt, with Pragmatos 3.0 under it. Grey bars are the project as it runs today. The green rows are what Pragmatos adds: a data clause from the first purchase order, a baseline at every factory acceptance test, and one record from the site acceptance test onward. The dashed red bar is punch closure as it runs today.
| Step · when | What happens | Hold point |
|---|---|---|
| 1Purchase ordermonths 1–3 | Every machine order carries one data clause: the machine exposes cycle, state, stop reason, part count and reject count over OPC UA or MTConnect, read-only. The interface survey lists what each maker confirms, and machines without a confirmed read path are named as out of scope on day one. | The integrator's project engineer accepts the survey, maker by maker. |
| 2Factory acceptancemonths 4–5 | At each maker's works, a read-only edge gateway reads the machine through its factory acceptance test. Pulse records its cycle and stop baseline, and that baseline travels with the machine to site. | Maker and integrator sign the baseline with the factory test. |
| 3Site acceptancemonth 8 | On site, read-only edge gateways bring all 36 machines into one namespace, with one tag dictionary and one clock. The performance run is measured from that record, and Pulse sets each machine beside its own factory baseline. | The plant owner's representative accepts the performance-run record. |
| 4Handover · warrantymonth 9 onward | Orbis rules flag a machine that drifts from its baseline and route the punch item to a named commissioning engineer. The instance and its record are handed over with the site, on the plant owner's servers, for the 12-month warranty. | The commissioning engineer approves every attribution. The plant owner signs handover. |
A read path specified in the purchase order costs a line in the order. The same read path asked for at the performance run costs a site visit from the maker. Writing it in first is what makes six makers' machines readable on the same day.
At the factory test, the machine runs alone and the maker is present. That is the one moment its own performance is undisputed. Taken there, the baseline is the reference every later argument is settled against.
In a control cabinet, a marshalling strip brings field wiring from many sources onto one row of terminals. Pragmatos 3.0 does the same for data: six makers' machines land on one namespace, read-only, and every recommendation that comes out of it ends with a named engineer.
The marshalling strip. Data flows left to right only; there is no write path back into any machine control. Each maker's machine keeps its own control and screen. Read-only edge gateways land all of them on one namespace with one clock. Pulse compares each machine with the baseline from its own factory test. Orbis proposes a punch item and its likely owner as a rule-based recommendation, and a named commissioning engineer approves or rejects it. Both are kept on record.
Who does what, and what is kept
| Action | Done by | Kept on record |
|---|---|---|
| Read cycle, state, stops and counts | Pragmatos, read-only | Raw events, on one clock |
| Set each machine beside its factory baseline | Pulse | Deviation per machine and shift |
| Propose a punch item and its likely owner | Orbis rule | Rule, evidence, proposed owner |
| Approve, reject or reassign the item | Commissioning engineer | Name, time, decision |
| Close the item on the machine | The maker's engineer | Closure time, run after the fix |
| Accept the site | Plant owner | Handover against the full record |
No write path into any machine control, so a maker can admit a gateway without a debate. An outage costs continuity, not data: the edge buffers roughly 72 hours.
Cycle, stop, reject and ready mean the same on every machine, on one time base. That turns six screens into one order of events.
Each project's instance runs on the plant owner's servers. At handover the record stays with the machines, under the owner's control.
Punch closure on the reference project, modelled. Both curves start from the same 120 items at the performance run. Today, the curve flattens from week 4: the easy items are gone and the 38 that sit between two makers wait about nine working days each for an owner. With one record, an Orbis rule proposes the owner from the order of events and the engineer approves it within a day, so the curve keeps falling and the site is handed over at week 7.
| Criterion | Today | With Pragmatos 3.0 | Verdict |
|---|---|---|---|
| Performance-run evidence | Two engineers spend about 15 days each copying six makers' screens into one workbook. | Read from one record. 60% fewer engineer-days, €14,400 per project. | Modelled |
| Repeated performance runs | 1.5 per project on average, often because a stop could not be traced. | 0.5 per project, with each stop traced to a machine. €9,600 per project. | Modelled |
| Run to handover | Ten weeks, held open by items between two makers. | Seven weeks, with owners proposed from the data. €24,000 per project. | Modelled |
| Warranty period | Ten engineer site visits in 12 months, most of them to find out what happened. | Six visits. The record answers the rest. €9,200 per project. | Modelled |
| Write access to machines | Each maker's remote access, each on its own terms. | None. Pragmatos reads and never writes. | By design |
| Where the record lives | On six screens and in the commissioning team's notes. | On the plant owner's servers, handed over with the site. | By design |
| Row | Figure | Type | Basis |
|---|---|---|---|
| C-01 | €800 / day | ASSUMED | Internal cost of one commissioning engineer-day, used in every row below |
| C-02 | 30 days | ASSUMED | Engineer-days per project collecting performance-run evidence: 2 engineers × 15 days |
| C-03 | −60% | ASSUMED | Share of that time removed when the evidence is read from one record |
| C-04 | €14,400 | DERIVED | 30 × 60% = 18 engineer-days × €800 (T-01) |
| C-05 | 1.5 → 0.5 | ASSUMED | Performance runs repeated per project, today and with one record |
| C-06 | 12 days | ASSUMED | Cost of one repeat: 4 engineers × 3 days on site |
| C-07 | €9,600 | DERIVED | 1 repeat avoided × 12 engineer-days × €800 (T-02) |
| C-08 | 10 → 7 wks | ASSUMED | Performance run to signed handover, today and with one record |
| C-09 | 2 engineers | ASSUMED | Team held on site for punch closure |
| C-10 | €24,000 | DERIVED | 3 weeks × 5 days × 2 engineers = 30 engineer-days × €800 (T-03) |
| C-11 | 10 → 6 visits | ASSUMED | Engineer site visits in the 12-month warranty period |
| C-12 | €2,300 / visit | ASSUMED | 2 engineer-days × €800 + €700 travel |
| C-13 | €9,200 | DERIVED | 4 visits avoided × €2,300 (T-04) |
| C-14 | €57,200 | DERIVED | €14,400 + €9,600 + €24,000 + €9,200 per reference project |
| C-15 | 12 / year | ASSUMED | Projects of this size the integrator delivers in a year |
| C-16 | €686,400 | DERIVED | €57,200 × 12 projects a year |
| C-17 | 120 · 38 · 9 → 1 d | ASSUMED | Open items at the performance run · items between two makers · working days to agree an owner, today and with one record |
| C-18 | Chart values | ASSUMED | Sheet 6 curves: a typical pattern of punch closure on a mixed site, consistent with C‑08 and C‑17 |
Assumed figures are rates chosen for the model and can be changed. Derived figures are arithmetic on them and move with them. The four results do not overlap: evidence collection sits inside the site acceptance test, repeats are separate runs, punch closure follows the accepted run, and warranty visits follow handover.
An integrator is judged on sites it handed over years ago. A dashboard on handover day shows a line once. A data clause in every purchase order, a baseline taken at every factory test, and a read-only record that stays with the site make every later project start from the same method, whichever makers the plant owner chose. The neutral record across makers is the point: it is what lets the integrator settle a fault on a machine it did not build, from data both sides accept.