Data center commissioning levels and process field guide

Why a data center commissions harder than other buildings

Data center commissioning is the staged quality process that verifies the facility is built as designed and survives the failures it was designed to ride, before any IT load is on the floor. A data center is not allowed to fail in production. A live floor carries other people's revenue and other people's data, so the failures get found on a scheduled afternoon with load banks, not at 2 a.m. with customers on the line. Everything in the process exists to move the moment of failure earlier, into a test, where it is cheap.

That is why data centers commission harder than almost any other building. An office can have an air handler trip and nobody calls it an outage. A data center that drops its critical bus for a half second has dropped the whole reason it exists. So the plant gets tested in stages, each stage proving more than the last, until the building has ridden a simulated utility loss at full load and held. The program side of all this, who runs it and how the record is built, is covered in the data center commissioning operations overview. This guide walks the levels themselves, what each one proves and in what order.

The short version: commissioning is not a final inspection. It is a chain of witnessed checkpoints that starts at the factory before the gear ships and ends with the whole plant proven together under a fault.

What are the data center commissioning levels?

Most data center programs organize commissioning into levels, commonly numbered Level 1 through Level 5, that climb from the factory bench to the fully integrated plant. Level 1 is factory testing, Level 2 is site receiving, Level 3 is pre-functional and static installation checks, Level 4 is functional performance testing of each system, and Level 5 is the integrated systems test. Each level proves more than the one below it, and each is signed off before the next starts.

The numbering is a convention, not a law, and it shifts from one program to the next. Some owners add a Level 0 for design review and commissioning planning, and a Level 6 for post-occupancy and seasonal testing, so a program can run Level 0 through Level 6 or further. Some split the static installation check and the first energization into separate levels, some fold them together. The project commissioning plan is the authority for what each level includes and who signs it, not a number you carried from the last job.

Many sites tag the gear to show what level it has passed, and a common scheme runs red for Level 1, yellow for Level 2, green for Level 3, blue for Level 4, and white for Level 5. The colors are a visual control, not a standard, and they vary by program the same way the numbers do. Do not assume a green tag means the same thing on two jobs. Confirm the tag legend against the plan.

The levels are gates, not labels. You do not start a level until the one below it is signed off, because a piece dropped into a higher test unproven is a finding you cannot isolate. Skip a level and the deficiency does not disappear. It hides until the most expensive moment to find it.

Level 1: factory witness testing (FAT)

Level 1 is the factory acceptance test, the witnessed test of the gear on the manufacturer's floor before it ships. The commissioning team and often the owner travel to the plant and watch the generator, the UPS, the switchgear, the chillers, and the PDUs run against the approved submittal and the bill of materials. A UPS gets driven through its operating modes, static and dynamic, with simulated load steps, mains failure, and battery transfer. Switchgear gets its protection settings and control logic checked. A generator package gets run and load-tested. The point is to catch the wrong firmware, the missed option, or the failed component while the gear is still in the factory's hands.

The economics are the whole argument. A faulty UPS module or a firmware mismatch found at the factory is a line on a punch list. The same defect found after the unit is set, terminated, and energized on site is a crane, a shutdown, and weeks. Found on a live floor it is an outage. FAT is the cheapest place in the entire program to find a problem, which is why owners pay to send people to witness it instead of accepting a paper certificate.

When Level 1 passes, many programs attach a physical tag, commonly red, that clears the gear to ship. No tag, no shipment. The trap at this level is treating FAT as a formality and signing the manufacturer's own report without watching the failure simulations run. The mains-failure and battery-transfer tests are the ones that matter, and they are exactly the ones a rushed witness lets slide.

Level 2: site acceptance, delivery, and receiving

Level 2 is the inspection of the gear when it lands on site, confirming it is what was ordered, it matches the approved submittal, and it arrived undamaged. Before anything gets set or energized, the commissioning team and the installing trade check the nameplate against the submittal, look for shipping damage, confirm the options and accessories came with it, and verify the storage conditions are right for gear that may sit for months before it is installed.

This level looks dull and gets treated as a formality, which is exactly why it bites. A transformer that took a hard hit in transit, a UPS cabinet that sat in the rain on a laydown yard, a switchboard delivered with the wrong bus rating, all of these are cheaper to reject at the dock than to discover at energization. Concealed shipping damage is the classic one. It does not show until the gear is energized and something inside that cracked in transit lets go. The freight claim is also far easier to win at delivery than three months later when the carrier is long gone.

Receiving and storage is its own discipline, covered by topic in the receiving guidance, and on a data center it carries more weight because the lead times are long and a damaged piece of switchgear can blow a schedule that has no slack. Many programs tag the gear at this level, commonly yellow, to show it cleared receiving.

Level 3: pre-functional, component, and static checks

Level 3 is the de-energized verification that each component is installed correctly and ready to energize, plus the first individual energization and startup. This is where the pre-functional checklist lives. Connections landed and torqued to spec, with the torque witnessed or torque-striped. Phasing and point-to-point wiring verified from the device back to the panel. Insulation resistance tested with a megger before anything is energized. Grounding and bonding confirmed. Settings and labels staged. Then the gear is energized individually and the vendor runs startup on its own equipment.

The discipline that separates a real Level 3 from a checkbox is the megger and the point-to-point. A new technician energizes a board trusting the wiring is right because it looks right. The pro meggers the bus and the feeders cold first, because an insulation fault found with a meter is a repair and an insulation fault found by energizing is an arc. Point-to-point catches the control wire landed on the wrong terminal, which will not show up until a functional test fails in a way nobody can explain.

Level 3 is also where the vendors' startup reports get captured, because the warranty and the functional test both depend on the gear having been started correctly. A signed pre-functional checklist is the permission slip to energize, and the functional test at Level 4 has no meaning if the static checks under it were never really done. Many programs tag this level green.

What is functional performance testing (Level 4)?

Level 4 is functional performance testing, the energized demonstration that each system does what its sequence of operations says, on its own, under normal and fault conditions. Every system gets run through its modes against a written script. The UPS is driven from normal to battery to bypass and back, and its transfer timing is measured. The generator is started, brought up to voltage and frequency, and load-tested, usually on a load bank because there is no IT load yet. The chillers and the cooling plant are run through their staging and setpoints. The controls and the BMS are exercised through their sequences. Fire detection and suppression and the security systems get their functional checks.

The key word is alone. Level 4 proves each system in isolation, not the seams between them. A generator can pass its own load bank test cleanly and the UPS can hold its own runtime, and the plant can still fail when they have to cooperate, because that failure lives in the handoff. Catching the single-system problems here is what makes the integrated test at Level 5 readable, because when the whole plant runs together you want the only new variable to be the interaction.

The technical content of the power-side functional tests, the protection, the transfer timing, the power quality, lives in the data center electrical commissioning by topic, and the airside and waterside functional tests live in the cooling commissioning by topic. The level itself is the gate: every system proven against its own sequence before any of them are asked to work as one. Many programs tag this level blue.

Level 5: the integrated systems test

Level 5 is the integrated systems test, the whole plant run together at design load and then put through a real failure to prove it holds the critical load. The utility gets dropped on purpose, the UPS carries the gap, the generators start and take block load, the cooling restarts on generator, and through all of it the critical bus never drops and the room stays in band. Then components get failed on purpose, a generator, a UPS module, a chiller, a feeder, to prove the redundancy on the one-line is real and not just drawn.

This is the keystone of the whole program, and it is the level schedule pressure attacks first because it needs the entire plant, full load banks, and a long uninterrupted block of time. It gets its own guide. The pull-the-plug, the failure-scenario script, the cooling ride-through, and the go-live gate are covered in depth in the integrated systems test guide, so this is the short version. Everything in Levels 1 through 4 proves a piece. Level 5 proves the pieces work as one machine under a fault.

A plant that never had a complete, continuous integrated test has never actually been proven, no matter how many lower-level scripts are signed. Many programs tag this final level white.

Why load banks carry the test

There are no servers in the building during commissioning, so the load has to be manufactured, and that is what load banks do. They let the plant be tested at the real design load it was built to carry instead of at no-load, which proves almost nothing. On the electrical side, resistive and reactive load banks pull the design kilowatts and kilovolt-amperes from the power chain. On the mechanical side, floor-standing heater load banks placed in the white space dump the same heat the IT racks would, so the cooling plant has a real thermal load to reject and the room has a real temperature to hold.

Testing at design load matters because most failure modes only show up under load. A generator picking up a small block comes up easy. Hand it the full design step and the alternator sags while the regulator catches up and the engine bogs while the governor recovers, and that is where you learn whether the machine can actually take the building. A lightly loaded room rides a cooling gap with room to spare. Load it to design and the temperature climbs fast the instant cooling drops. The quiet version of the test is not the one that fails in production.

How load banks are sized, the difference between resistive and reactive, the staging and connection, and the wet-stacking concern on diesels are covered in the load bank testing by topic. The rule for the levels is blunt. If a functional or integrated test was not run at the design load, it sampled a quieter building than the one that goes live.

What does a commissioning agent do?

The commissioning authority, the CxA, is the owner's independent agent who plans the verification, writes the scripts, witnesses the tests, keeps the deficiency log, and signs off whether the building meets the owner's requirements. The word that carries the weight is independent. On a data center the CxA is almost always a third party hired directly by the owner, not a subconsultant to the general contractor, because the entire value of the check is that it is not grading its own homework.

The rest of the team has fixed roles. The owner sits at the top and inherits the building and the record at the end. The engineer of record owns the design intent and answers when a test result does not match the sequence of operations. The general contractor or construction manager holds the schedule and coordinates access and energization. The trade contractors run their own pre-functional checks and staff the functional and integrated tests. The equipment vendors run startup on their gear and stand behind the warranty. Across all of it the CxA witnesses and accepts rather than performs, because the agent who runs a test cannot also be the one who certifies it.

When the roles blur, the process loses its worth quietly. The most common and most damaging failure is the commissioning agent reporting to the general contractor instead of the owner, so the party being checked controls the checker's schedule and paycheck. The team and the roles are covered in more depth in the operations overview. The one line to read on any contract is who the CxA actually works for.

The MEP trades and disciplines in commissioning

Commissioning on a data center crosses every MEP discipline, and each one carries part of the proof. The electrical scope is the power chain: the utility service, the switchgear, the UPS, the generators, the transfer scheme, and the distribution down to the rack, checked for protection coordination, transfer timing, and power quality. The mechanical scope is the cooling: the chillers, the pumps, the air handlers or CRAH units, the piping, and the controls that hold the room temperature. Controls and the building management system tie them together, and a surprising share of integrated-test failures actually live there, because the sequence logic is the part written last and tested least.

Two disciplines get treated as afterthoughts and should not. Fire detection and suppression has to be commissioned and coordinated with the electrical and mechanical shutdowns, because a clean-agent release or a fan shutdown on alarm interacts with the systems keeping the room alive. Physical security and access control are part of the commissioned facility too. Each discipline runs its own functional tests at Level 4, and the integrated test at Level 5 is where they have to behave as one, which is why the controls integrator is in the room for the whole thing.

The commissioning documents: OPR, BOD, plan, and report

The whole process hangs off a short stack of documents, and the order matters. The owner's project requirements, the OPR, states in measurable terms what the owner needs the building to do: the redundancy they are paying for, the temperature and humidity envelope, the acceptable downtime, the concurrent-maintenance expectations. The basis of design, the BOD, is the design team's documented answer to the OPR. Commissioning checks the design against the OPR, then the build against the BOD. Every test script has to trace back to one of those requirements, because a test with no requirement behind it has no defensible pass or fail.

The commissioning plan is the operating manual for the program. It names the systems in scope, the levels and what each includes, the team and who witnesses what, the hold points, the schedule, and the deficiency process. The scripts and test procedures are the per-system and per-scenario instructions with their expected results. The issues or deficiency log tracks every finding to closure. The commissioning report rolls it all up into the owner's professional record of what was proven and what remains open.

These documents and how the program runs from them are covered in depth in the operations overview, so this guide does not repeat that. The point for the levels is that each level produces its piece of this record, and a level signed off without its document behind it is a level nobody can defend later.

Script-based testing: every test is written before it runs

Every functional and integrated test runs from a written script, not from a technician's judgment in the moment. The script states the step, the expected result, and a pass or fail box, and it is written before the test from the sequence of operations, not filled in afterward to match what happened. The commissioning authority builds it, the owner often reviews it ahead of the test, and the witness signs each step as it runs.

The discipline that separates a real script from a checkbox is that the witness signs each step, not the page. A script watched once in a hurry and ticked off as a block proves almost nothing, because the value is in recording the actual measured result against the expected one, step by step. The transfer that completed but took longer than the sequence allowed is a finding, not a pass, and you only catch that if the script demanded the actual time be written down and compared.

Write the scripts so a stranger could run them and so the results stand on their own when the owner reads them two years later. A script that only its author can interpret is a script that fails the moment that person rolls off the job. The expected result is the part people skip. A pass or fail box with no expected value next to it is an opinion, not a test.

The issues and deficiency log

The deficiency log, also called the issues log, is the running list of every problem commissioning finds, tracked from the moment it is identified to the moment it is verified fixed and closed. It is the most important operational artifact of the program, because the tests prove the plant in the moment but the log carries the findings until they are actually resolved. On any given day it is the honest measure of how far from done the building really is.

Every finding moves through a fixed lifecycle: identify, classify by severity, assign to a responsible party with a due date, resolve, verify, close. The verify step is the one that gets skipped under schedule pressure, and skipping it is the failure that defines a weak program. A deficiency closed on paper without a re-test is a deficiency still in the building. A relay setting noted to be corrected but never re-read in the gear, a failed transfer marked resolved on a controls change nobody re-ran, a leak fixed and never re-pressurized. The log looks clean and the problem is still there, waiting for the night the utility actually drops.

Keeping that log honest across a long job with many parties is the hard part, because the finding, the photo, the assignment, and the closure evidence usually scatter across different inboxes and spreadsheets. This is the field tracking the tradeos workflow is built to hold, so the finding, its evidence, and the sign-off stay attached to the same record. The full lifecycle is covered in the operations overview.

Failure-scenario testing is the heart of it

Failure-scenario testing is what makes data center commissioning different from commissioning any other building. You do not just prove the systems run. You prove they keep running when something breaks, by breaking it on purpose. Utility loss is the headline scenario. On top of it the script layers the compound failures that prove fault tolerance and concurrent maintainability: a generator that fails to start during the utility event, a UPS module dropped while on generator, a chiller lost mid-ride, a pump gone, a feeder or breaker opened to force the alternate path to pick up.

Each scenario asks the same question in a different place. When this piece dies, does the redundant piece carry the load without the critical bus or the room temperature leaving its band? That is the difference between a redundancy claim drawn on a one-line and a redundancy that actually works. The Tier topology and the concurrent-maintenance and fault-tolerance language behind these scenarios are covered in the Tier and redundancy by topic. Commissioning is where the claim gets demonstrated instead of asserted.

The scenarios build through the levels. A single system's fault response is checked at Level 4, on that system alone. The compound failures, where one fault happens during another, only get proven at the integrated test, because they need the whole plant running together. A program that tests every component's own fault response and never runs the compound failures has proven the parts and not the building.

Witnessing, hold points, and sign-off

A hold point is a step the work cannot pass until the required witness has seen it and signed. First energization of switchgear, a functional test, dropping the utility for the integrated test. The commissioning plan lists the hold points and names who has to be present: the CxA always, often the owner, sometimes the engineer of record, the equipment vendor, and on a Tier job the Uptime Institute witness. Miss the witness and the test does not count, no matter how well it went.

Witnessing only works on notice. The plan sets a notification period, often several days, so the witnesses can travel and the test runs with the right people in the room. The classic expensive failure is a contractor running a witnessed test without proper notice to hit a schedule date, then being told it has to run again because the people who had to see it were not there. That is rework with no defect, pure schedule and cost burned because the notification step got skipped.

Document what was witnessed at the moment it happens. A witness signature against a script step, dated, beats a recollection a month later when the question is whether the generator actually accepted block load or just started. Sign at the test, not at the closeout meeting, because the closeout signature is reconstructing a memory and the at-test signature is recording a fact.

What is recommissioning and ongoing commissioning?

Recommissioning is repeating the commissioning process on a building that was already commissioned, to confirm it still performs after time, modifications, and drift. Ongoing commissioning, sometimes called monitoring-based or continuous commissioning, does the same job continuously using the building's own metering and trends. Retro-commissioning is the related case for a building that was never formally commissioned the first time. New-construction commissioning is the process this guide walks. The existing-building track follows the same logic on a plant that is already running.

Data centers need this more than most buildings because they never stop changing. Racks get added, loads grow, cooling setpoints get nudged, firmware gets pushed, breakers and timers get touched during maintenance. The redundancy proven at turnover slowly stops matching the building as it actually runs. A redundancy claim that has not been demonstrated under a real failure in a few years is a claim, not a proven capability.

The strong programs treat the acceptance commissioning record as the baseline and re-run a version of the integrated failure test on a defined interval, often annually, to confirm the plant still rides through after a year of changes. The handoff from new-construction commissioning to ongoing commissioning is the hinge, and the operations overview covers the ongoing side and the systems-manual maintenance program that drives it.

Why does commissioning get squeezed, and why you do not skip it?

Commissioning is the last activity before turnover, so it absorbs every slip ahead of it. When the gear ships late, the install runs over, or the controls integration drags, the owner's move-in date does not move, and commissioning gets compressed into whatever window is left. A program planned for twelve weeks gets eight, and the eight get eaten by rework nobody scheduled. That is the structural reason the temptation to skip levels shows up, and it always shows up late, when the people deciding are tired and behind.

The temptation is real and the answer is no. Skipping a level does not remove the defect. It moves the discovery to a more expensive place. A FAT skipped to save a factory trip becomes a UPS fault found after the unit is set and terminated. A pre-functional megger skipped becomes an insulation fault found by energizing. An integrated test cut short becomes a transfer failure found at 2 a.m. with customers on the floor. The cost of a defect roughly multiplies at each level you let it pass unchecked.

Protecting commissioning is a scheduling decision made at the start, not a recovery at the end. Build the commissioning sequence into the master schedule as its own track tied to construction milestones, front-load the levels that can run early, and fence off the integrated test window so the only thing left at the end is the test that genuinely needs the whole plant. The honest driver of the schedule is not the number of tests. It is how much rework the testing uncovers, and you cannot schedule your way out of a building that was not ready to be tested.

Turnover and as-builts

Commissioning feeds the turnover package, the record the owner inherits and operates the building from. The commissioning portion is the plan, the test reports at every level, the signed scripts, the integrated test results, and the deficiency log closed out with evidence. Around it sits the systems manual: the OPR and BOD so the operator knows what the building was meant to do, the final sequences of operations, the as-left settings, the operation and maintenance information, the recommended testing intervals, and the training records.

The as-builts have to reflect what was actually installed, including every field change, because the operations team will trust those drawings the day something fails at 2 a.m. A feeder that was re-routed and never redlined, a setting changed in the field and never captured, a device relocated off the plan, each becomes a trap for the operator who pulls up a drawing that no longer matches the building. The turnover and as-built discipline is covered by topic in the turnover guidance.

Two parts carry the most weight later. Training is the one that gets rushed and the one the operators remember, because a building handed to a team that was never trained on its sequences runs on guesswork until something breaks. And the dated commissioning records become the proof of when the gear was accepted and in what condition, which is what the warranty clock and any later dispute reference. A turnover package missing the as-lefts, the closed log, or the training is missing its spine.

What to document at each level

The record is judged by whether someone two years out can reconstruct what was proven at each level and check it against what the script required. Each level produces its own document, captured at the level and not reconstructed at the end. The table is the spine of a level-by-level commissioning record, and a level closed without its piece of it is a level that cannot be defended.

Build the record from the field as the levels are completed, not in the last two weeks before turnover. A factory report filed against the gear it belongs to, a pre-functional checklist signed at the board, a functional script signed step by step at the test, an integrated scenario logged against its branch with the data behind it. Captured that way, the owner report assembles itself from the record instead of being stitched together from memory.

Common mistakes

• Skipping a level to recover schedule, which moves the defect to a more expensive place instead of removing it.
• Running with no independent CxA, or letting the CxA report to the general contractor instead of the owner.
• Accepting a factory test on paper without witnessing the mains-failure and battery-transfer simulations.
• Energizing at Level 3 without the megger and the point-to-point, so an insulation or wiring fault is found by arc instead of by meter.
• Writing weak scripts with a pass or fail box and no expected result next to it.
• Closing deficiencies on paper without re-testing the change that was supposed to fix them.
• Never running a complete, continuous integrated systems test, so the seams between systems are never proven.
• Running the functional or integrated tests at no-load or partial load instead of design load.
• Running a witnessed hold point without notice, so it does not count and has to run again.
• Treating site receiving as a formality and missing concealed shipping damage until energization.

Field checklist

Standards and references

The commissioning process framework comes from ASHRAE Guideline 0, the commissioning process, which sets the role of the commissioning authority and the framework for the OPR, the BOD, the commissioning plan, the procedures, and the reports. ASHRAE Standard 202, the commissioning process for buildings and systems, carries the same process into an ANSI standard, and the HVAC technical application sits in ASHRAE Guideline 1.1. The Building Commissioning Association, the BCxA, publishes best practices that separate new-construction, existing-building, and ongoing commissioning. Confirm the current edition of any of these against the project documents, because titles and numbers shift between cycles.

The level numbering itself, Level 1 through Level 5, is an industry convention rather than a single published standard, which is why it varies from program to program. Some run Level 0 through Level 6 or further, and the tag colors and level boundaries differ by owner. The project commissioning plan and the Division 01 commissioning specification, commonly numbered around 01 91 00 in the CSI MasterFormat the project adopted, are what actually define the levels for a given job. Cite the plan, not a number you carried from the last project.

The discipline standards that govern the underlying tests sit beneath the process: NETA acceptance testing for the electrical scope, NFPA 110 for the standby power plant, the IEEE protection and grounding references, and the ASHRAE TC 9.9 thermal guidelines for the room envelope. The manufacturer's own procedures govern the factory tests. For a data center chasing an Uptime Institute Tier certification, the constructed-facility certification is a witnessed demonstration that the built plant meets its claimed Tier, so the integrated test and the redundancy demonstrations often do double duty as both commissioning and the Tier witness. The commissioning plan and the contract documents control scope. The standards give the framework.

Terms and acronyms

Commissioning carries its own vocabulary, and the same word can read differently across a commissioning plan, a spec, and a Tier document. The terms below are the ones that travel across the levels.

Cx / CxA

Commissioning, and the commissioning authority who plans, witnesses, and signs off the process, independent of the installer

L1 to L5

The common commissioning levels, from factory test through receiving, pre-functional, and functional to the integrated test, with boundaries set by the plan

FAT

Factory acceptance test, the witnessed Level 1 test of the gear at the manufacturer before it ships

SAT

Site acceptance test, the Level 2 inspection of delivered gear against the approved submittal

PFC

Pre-functional checklist, the static de-energized verification that a system is installed and ready to energize

FPT

Functional performance test, the energized Level 4 demonstration that a single system runs its sequence

IST

Integrated systems test, the Level 5 full-plant test that proves all systems ride a simulated failure at design load

OPR / BOD

Owner's project requirements and basis of design, the requirement and the design team's documented answer to it

Hold point

A step the work cannot pass until the required witness has seen it and signed

Load bank

Equipment that creates electrical or thermal load to stand in for the IT load during testing

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is commissioning in construction?
• What is an integrated systems test?
• How long does data center electrical commissioning take?
• What is the difference between a PDU and an RPP in a data center?
• What is an EPMS in a data center?