Commissioning Has a Critical Path: Planning for Turnover

The construction critical path is the one the board sees. Foundations, structure, envelope, mechanical, electrical, controls — each week’s slippage measured against a milestone the whole team can name. It is a real path and worth watching. It is not the path that decides when the facility can be handed over.

The commissioning critical path is different in shape. It runs backward from ready—for—startup and threads through documents and dependencies most weekly reviews never surface. The owner’s project requirements have to be specific enough to argue with. The basis of design has to be current. The sequence of operations — the plain-language description of how the systems are supposed to behave under normal, abnormal, and failure conditions — has to be finished before test scripts can be honestly written. Factory acceptance has to catch the faults that are cheap to catch there. Pre-functional testing has to be complete before functional performance, and both before the integrated systems test can prove the facility as a whole. Break the chain at any link and the durations downstream become an estimate of an unknown.

Most owners see the first path in every review and the second one in almost none. That is how a program with a green construction schedule at month eighteen goes red at month twenty-two on a finding that was baked in a year earlier — when someone approved a sequence of operations that was too vague to script against, or accepted an equipment substitution nobody costed against the test plan, or let two trades sign off on their own interfaces without a shared gate. None of those decisions looked expensive on the day. Each of them quietly moved days onto a path nobody was measuring.

We watched a Gulf Coast industrial client stand up a program a few years ago with a commissioning authority in the room at basis-of-design review. Not for a briefing. For an argument. The engineering lead had drafted a sequence of operations that read cleanly in prose and left three failure modes without a defined response. The commissioning authority pushed back before a single specification was released. It took most of a week to close, and it saved the test plan the program would have written to it.

From that point the commissioning path lived on the same wall as the construction path. Sequenced activities with real durations and named resources, not a bar labeled Cx floating near the finish. Factory acceptance criteria were tied to the sequence of operations, not to a vendor’s standard checklist. Pre-functional testing started on the first system that was mechanically complete, not when the whole plant was. One deficiency log ran across every discipline, with one owner, and the argument about whether a fault was electrical or controls or mechanical happened once, in the review, and then was closed. When the integrated test ran, roughly seven in ten scripts passed on the first attempt. Nobody in the room thought that was luck.

The move that made the difference was structural, not clever. The commissioning authority had authority to slow the construction schedule when the commissioning path demanded it, and used it twice in the first six months. Both times the argument was resolved by looking at the same integrated schedule. Neither slip made the board report. The date the business had committed to did.

A different program, a hyperscale build in a moderate climate, put its commissioning authority under contract eleven months in. The construction team had run a disciplined job to that point. The steel went up on rhythm. The mechanical rooms looked photograph-ready. Two floors were substantially complete when the commissioning team walked the site for the first time.

The problems were not in the field. They were in the paper. The sequence of operations had been written by three different engineers across two design revisions and never reconciled. The resilience claim in the owner’s project requirements said fault-tolerant, which scripts unannounced faults, dual-path compartment loss, and continuous-cooling ride-through — a scope materially larger, riskier, and longer than concurrent maintenance. Nobody had made that scope real in a test plan. Two vendor substitutions during construction had changed the fault behavior of specific gear in ways the design intent document did not reflect. The factory acceptance packages that came with those substitutions were the vendor’s standard, not a script written to the sequence of operations that governed the building.

The commissioning team did the work that could still be done. They rebuilt the sequence of operations against as-installed drawings, wrote functional performance scripts that matched, and ran pre-functional testing on completed systems in parallel with construction on the rest. They found what they always find in that position — two mis-wired current transformers, a controls tuning issue that had been passed downstream three times, and a cascade in the load hand-off between the building management system and the generator plant that only appeared under a specific fault. Each of those was a morning’s rework at pre-functional and a lost day at integrated test. Because pre-functional had started late, most of them landed at integrated test.

The go-live moved twice. The commissioning team was not the reason. They were the messenger, and the message was that a commissioning path picked up at mechanical completion is a path with most of its float already spent.

The two programs are not different in talent. They are different in when the commissioning path was named and whether it was allowed to influence decisions upstream of the test floor. Every commissioning path has the same links, and the ones that break are usually the same ones.

A specific sequence of operations sits at the head of the chain. Without it, the scripts written downstream are estimates of an unknown, and the durations attached to them are fiction. A resilience claim — concurrently maintainable or fault-tolerant — sits with the owner and has to be decided early, in writing. Made late, that decision either shrinks to fit the calendar, which quietly abandons the claim, or consumes the float that was protecting everything else. Factory acceptance has to catch what factory acceptance can catch, because the same fault caught during a full building test invalidates the run and costs days. Pre-functional testing has to start on the first system that is mechanically complete, not on the last, because the compounding is where the calendar gets its air back.

The seam that fails most often on facility programs is the control sequence. The generator starts correctly. The cooling plant rides through. The building management system fails to hand the load off in the right order, and the whole run is invalidated by a fault that lived in the space between two teams. Where facility commissioning and process or IT commissioning run on separate schedules under separate owners, both records can show a pass while the real defect lives in the gap. The fix is structural. One integrated schedule. Named interlock gates that neither track can pass alone. One deficiency log with one owner. The clever versions of that fix do not work; the boring version does.

Once early blocks of the facility are carrying live load, destructive testing changes character. A fault injection that was routine in an empty facility becomes a decision about protecting somebody’s production. Which faults may be injected at each stage, and how live-block redundancy is preserved throughout, has to be written down and signed by the people who will operate the facility. Not resolved in the room at midnight.

That document is also the fairest way to say no. Without it, every scripted test late in the program becomes a negotiation, and negotiations under schedule pressure tend to end the same way. The test that gets deferred is almost never the one whose deferral is safe. It is the one whose owner is tired, or whose champion is not in the room that week, or whose calendar cost is easiest to explain.

There is a related judgment on the test rig itself. A load bank proves topology and redundancy under smooth, constant current. Some modern workloads do neither — they ramp from idle to full in milliseconds, impose transients the rig cannot mimic, and dump heat into a liquid loop. Where the real load will behave differently from the test rig, the honest test before go-live is a proxy run of the workload itself. Where it will not, adding one is calendar spent for its own sake.

The list of moves is short and unfashionable. Put commissioning in the master schedule as sequenced, resourced activities with real durations. Engage the commissioning authority during design, where the cheapest defects are found. Insist on a specific sequence of operations before scripts exist. Decide the resilience claim at owner level and put it in writing. Run one deficiency log across every discipline. Protect the last four to six weeks of the calendar, because compression almost always lands on integrated testing, which is precisely the wrong place for it.

Two honest counterweights. Commissioning is not free, and gold-plating is real. The governing standards rely on sampling rather than one-hundred-percent checking, because exhaustive verification consumes calendar without proportional risk reduction. And the commissioning authority is not a substitute for anyone else’s responsibility. It verifies. Design, code compliance, construction management, and acceptance rest where they always have.

The move that separates programs that hand over on the committed date from programs that discover the commissioning path in the last eight weeks is single-named ownership. One person owns the commissioning critical path. That person has both schedules in one view and the authority to slow the construction path when the commissioning path demands it. Without the authority the role is a report. With it, the two paths converge on a date the business can actually run.

Late commissioning always feels like a surprise. It is almost never a surprise. The finding that shows up in an integrated test at month twenty-two was written into a document at month six, or accepted with a substitution at month ten, or allowed to sit in the gap between two teams for a year. Everyone signed something. Nothing looked expensive on the day.

The date the business committed to is not a construction date. It is the day the plant runs at the quality the operator can actually stand behind — the day the pharmaceutical line makes qualified product, the day the data center takes production traffic, the day the refinery hits nameplate and holds it. Between the last poured slab and that day sits a body of work that has its own logic and its own path. Whoever owns that path owns the turnover date. If nobody owns it, the date is decided by whichever pressure is loudest in the last month, and that pressure is never the one that gets the facility ready.

The operators who finish on the committed date see both paths on day one and treat them as one program. The ones who don’t are not less capable. They are looking at the wrong schedule.