Data center floor flatness and levelness (FF/FL)

Flatness and levelness are two different things

Floor flatness and floor levelness measure two different defects, and confusing them is the first mistake on most data center floors. Flatness is the bumpiness of the surface over short distances, the waves and ridges you would feel pushing a cart. Levelness is whether the floor tilts, the slow fall of the whole plane away from horizontal over a long distance. A floor can be dead flat and noticeably tilted at the same time, or perfectly level and full of small humps. They do not move together.

Picture a sheet of plywood. Lay it across a hillside and it is flat but not level. Crumple it slightly and set it on a true bench and it is level but not flat. Concrete does both at once, in different amounts, and the spec usually calls a separate number for each.

On a data center this matters because the two defects break different things. Tilt across a hall throws off the plenum height and the pedestal grid under an access floor, and it lets racks lean on a hard floor. Short bumps catch caster wheels, rock a rack base, and leave gaps under leveling feet. You measure both, you accept against both, and you keep them straight in the record.

What is the difference between FF and FL?

FF is the floor flatness number and FL is the floor levelness number, and they come from two different things the surface does. FF is built from the change in slope over successive 12 inch (305 mm) intervals along a measurement line, so it captures the short-wave bumpiness. FL is built from the difference in elevation over 10 ft (3 m) along the line, so it captures the long-wave tilt and dish. Bumps drive FF down. Tilt and broad dishing drive FL down.

Both are unitless, both run on a scale where higher is better, and both are linear. An FF50 floor is twice as flat as an FF25 floor. An FL40 floor is twice as level as an FL20 floor. As a feel for the magnitude, FF25 is roughly a quarter inch of deviation in a 10 ft span, FF50 about an eighth inch, and FF100 about a sixteenth. Those are approximations to build intuition, not acceptance criteria.

The trap is treating one number as the floor's quality. A high FF with a low FL is a smooth floor that runs downhill. A high FL with a low FF is a level floor that is washboarded. The hall needs both held, and the spec should call both.

How are FF and FL measured and computed?

FF and FL are measured to ASTM E1155, the standard test method for the F-number system. The technician runs an inclinometer-type profiler, commonly a walking dipstick or an equivalent F-meter, along a pattern of straight sample measurement lines laid out across the floor. The instrument reads elevation change at a fixed spacing, usually about 12 inches, and the software turns that string of readings into the FF and FL for each line and for the test section.

The math is statistical, which is the part people miss. FF comes from the standard deviation of the slope changes between consecutive readings. FL comes from the standard deviation of the elevation differences at 10 ft. Both are reduced to a single index per line, then combined across lines into a section value with a confidence band. Because it is statistical, a few good lines do not rescue a bad floor and a few bad readings do not sink a good one. The sample has to be representative.

E1155 also pins down where you can read. Lines are kept off the very edges and away from penetrations and construction joints by a set distance, because edge effects and joints are not the field of the slab. You read the floor, not the curl at the form line. Confirm the current edition and the exact layout rules before the survey, because the method has been revised over the years.

SOV and MLV, and why an average can hide a failure

Every F-number spec carries two values, and accepting against only one is how a bad floor passes. The specified overall value (SOV) is the F-number for the whole placement, the composite across all the test sections. The minimum local value (MLV) is the floor under any single test section. The SOV says the floor is good on average. The MLV says no one region is allowed to be a disaster.

The reason both exist is that an average lies. A placement can hit an SOV of FF35 while one bay reads FF18 because the rest of the floor carried it. Under racks and access-floor pedestals, the local bad spot is exactly what bites, not the average. The MLV, commonly set somewhere around two-thirds of the SOV, is the floor under that local failure. ACI 117 frames the SOV and MLV relationship; confirm the ratio in the project spec because it is the spec, not a rule of thumb, that controls.

Read acceptance against both numbers. A floor that makes the SOV but blows the MLV in one section is a fail in that section, and that section is where the work goes. Accept on the average alone and you have signed for a floor with a hole in it that the racks will find.

Why don't F-numbers apply to every floor?

FL does not apply to every floor, and this is the limitation that gets specs in trouble. The levelness number is only valid on slabs cast on grade and on suspended slabs while they are still shored. It is not valid on unshored slabs that deflect, and it is not valid on a suspended slab after the shoring is removed. The reason is physical: once you pull the shores, the slab deflects under its own weight and any superimposed load, and that deflection is not a finishing defect the FL number was built to judge.

FF behaves differently. Floor flatness, the short-wave bumpiness, stays meaningful on a suspended slab even after shoring comes out, because deflection over a 12 inch interval is small. So on an elevated structural deck you can still hold an FF, but FL has to be handled another way, by a survey to a tolerance the structural spec sets, not by an E1155 levelness number.

E1155 also covers only randomly trafficked floors. For floors where the wheels run a defined path, a different method applies. So before you write FF and FL into a data center spec, settle two questions: is the slab on grade or suspended, and is the traffic random or defined. Get those wrong and the numbers you specified cannot legally be measured the way you wrote them.

When do you test floor flatness?

Test fast. ASTM E1155 measurement is meant to be taken within 72 hours of the slab being finished, and on a suspended slab it has to be done before any shoring is removed. The window is short on purpose. F-numbers describe the floor the finisher produced, and the longer you wait, the more curling, drying shrinkage, and traffic damage contaminate the reading and muddy who owns a failure.

The 72 hour rule is also a contractual fairness point. Measured early, the number reflects the placement and the finishing crew owns the result. Measured weeks later, after the trade stacked material on it and forklifts ran across green concrete, the floor can read worse for reasons that have nothing to do with the pour, and the argument over cause is unwinnable. The crew that wants to be judged fairly wants the test inside the window.

Build it into the pour schedule, not the punch list. The F-number test gets scheduled with the placement so the technician is on site the next morning, before shores come out and before the floor takes traffic. Confirm the exact timing language in the adopted edition of E1155 and in the project spec, because the spec can tighten it.

Field example: a data hall slab tested to FF35/FL25

Take a slab-on-grade data hall poured in a single placement, specified FF35/FL25 as the SOV with an MLV of FF24/FL17. The technician runs the dipstick the morning after finishing, inside the 72 hour window, on a layout of sample lines covering each test section. The composite comes back FF38/FL27. The overall passes.

Then read the sections. One bay near the construction joint reads FF21/FL16, both under the MLV. The placement made its average because the other bays were strong, but that one bay is below the local floor, and it sits right where two rows of racks will land. That section fails, average or not.

The decision is not reject the slab. It is fix the section. Map the low and high spots in that bay from the profile data, grind the highs that are dragging the flatness, and check whether the levelness miss is a tilt that grinding can chase or a dish that needs fill. Then retest the repaired section to the MLV. The record shows FF38/FL27 overall, the one section out, the remediation, and the passing retest. That is a defensible acceptance, and it is the section data, not the headline number, that drove it.

What FF/FL does a data center need?

There is no single data center number, because two different floors live in a data center. The structural slab under an access floor and a hard floor where racks roll and sit directly carry different tolerances, and the spec should call them separately. The values below are common ranges to set expectations, not acceptance criteria. The project specification governs.

A subfloor under a raised access floor is usually the looser of the two. The adjustable pedestals take up slab variation, so a conventional to moderately flat slab, often in the FF25/FL20 neighborhood, is common, with the real constraint being uniform plenum height and pedestals that seat without excessive shimming. A hard-floor data hall is tighter, because nothing between the rack and the concrete absorbs the error. Flat to very flat ranges, FF35/FL25 and up, are typical, and high-density halls with rolling deployment can push higher.

Two cautions on the numbers. First, FL only applies to slabs on grade, so a suspended structural deck needs a survey tolerance instead of an FL. Second, a number copied from a warehouse spec is usually wrong for a hall, in both directions. Set the FF and FL to what the racks, carts, and access floor actually need, and write the SOV and MLV that match.

The subfloor under a raised access floor

Under an access floor, the slab is a base for pedestals, not a finished surface, and that changes what you are protecting. The pedestals adjust in height, so they swallow a fair amount of slab variation, which is why the slab tolerance can be looser here than on a hard floor. What you actually care about is that every pedestal seats clean on the concrete and that the plenum height stays uniform across the room.

Tilt is the enemy more than bumps. A slab that falls across the hall forces the pedestals on the low side to run tall and the ones on the high side to run short, and past the pedestal's adjustment range the installer starts shimming or cutting, which is slow and weakens the grid. A broad levelness miss can also pinch the plenum where airflow and cable depend on it. So on a subfloor, FL and the overall fall across the room often matter more than chasing a high FF.

Local high spots still bite. A bump under a pedestal base means the base does not sit flat, the locking nut does not bear evenly, and the stringer grid carries a built-in rock. Those are the spots to grind. The access-floor installer will hand you a punch list of pedestals that would not seat, and those map straight back to slab defects the F-number average never flagged.

The slab-on-grade hard-floor data hall

High-density halls increasingly run hard floors, slab on grade with the racks sitting and rolling directly on the concrete, and the tolerance gets tighter because nothing buffers the error. There is no pedestal to take up a bump and no access floor to span a dip. The caster wheel and the leveling foot meet the slab as it was finished.

Two things drive the spec. Racks roll in on casters during deployment, and a wavy floor catches the wheels, makes a heavy rack hard to push straight, and can rack the frame on a hump. Once parked, the rack sits on leveling feet, and a local dip or tilt means a foot hangs or the cabinet leans, which throws off door alignment, aisle containment seals, and the load path on the feet. Flatness handles the rolling, levelness handles the sitting, and the high-density hall wants both held tighter than a subfloor.

Hard floors also push the conversation past FF and FL alone. The mid-wavelength roughness that affects a rolling rack, the waves in the 2 to 10 ft range, is not fully captured by FF, which is why a hall built for rolling traffic may add a waviness check on top of the F-numbers. Specify the rolling concern, not just the static one.

Access-floor levelness is a separate check from slab F-numbers

The finished access floor has its own levelness tolerance, and it is a different measurement from the slab F-numbers, taken at a different time, against a different number. Mixing them up is a common handoff failure. The slab gets FF and FL to E1155 before the access floor goes in. The finished access floor gets a planar levelness check after it is installed and the pedestals are locked.

Access-floor levelness is commonly specified as a planar tolerance, often on the order of plus or minus 0.060 in in any 10 ft, about 0.10 in over the entire floor area, and roughly 0.030 in across panel joints. Those figures trace to CISCA recommended practice and the access-floor manufacturers, and the project spec and the manufacturer's instructions control the exact values. Notice these are straightedge-style tolerances in inches per distance, not F-numbers. You do not run an E1155 dipstick to accept the finished access floor.

Keep the two records separate and sequenced. The slab F-number report accepts the concrete. The access-floor levelness report accepts the installed floor, and the across-joint number catches a panel that rocks even when the room reads level overall. The raised-floor acceptance packet ties these together; treat the slab FF/FL and the finished-floor levelness as two distinct line items, because they are.

Measurement methods: profiler, straightedge, and 3D scan

Three methods show up on a data center floor, and they answer different questions. The F-number profiler is the acceptance tool for the slab. A walking dipstick or equivalent F-meter reads slope at a fixed spacing along the sample lines, and the software computes FF and FL to E1155. This is the method a spec means when it calls F-numbers, and it is the one that produces a defensible acceptance record.

The straightedge and feeler-gauge method is older and still useful for spot checks and for tolerances written that way. You lay a straightedge, commonly a 10 ft or a 2 ft, on the floor and measure the gap underneath with a feeler gauge or wedge. ACI 117 carries straightedge-based flatness classes, and many access-floor and resilient-flooring tolerances are written as a gap under a straightedge rather than as an F-number. It is quick, it is local, and it does not give you a statistical floor value the way E1155 does. The waviness method, ASTM E1486, is a related straightedge-family approach aimed at the mid-wavelength ride that FF misses.

Laser scanning and 3D survey produce an as-built surface map, a dense elevation model of the whole floor. They are strong for showing where the highs and lows are, for planning grinding, and for verifying levelness over the room. Used well, the scan complements the F-number test; it shows the shape so you know where to work. It does not replace E1155 for acceptance unless the spec says it can, so confirm what the contract accepts before substituting a scan for the dipstick.

Random traffic vs defined traffic, and superflat floors

E1155 and the FF/FL system cover randomly trafficked floors, where carts and people move in any direction. When the traffic runs a fixed path, the rules change. A defined-traffic floor is one where wheels track the same lines every pass, like a narrow-aisle warehouse or, in a data center, a deployment aisle where racks roll the same route in and out. Defined traffic is judged on a different basis, often the F-min system and the waviness and wheel-path criteria in ASTM E1486, not on FF and FL.

The defined-traffic numbers run higher and tighter because the wheel feels every wave on the path it actually uses. A superflat floor specified as F-min 100 is roughly comparable to FF140/FL100, a far flatter surface than any random-traffic hall needs. You would only reach for that on a hall where very tall, heavily loaded racks roll on a fixed path and ride quality on that path is the controlling concern.

The practical call for a data center is usually this. A general hard floor is random traffic, so FF and FL to E1155 are right. A specific high-density deployment lane where racks track a defined path may warrant a defined-traffic or waviness criterion on that lane. Do not specify superflat for the whole room out of caution; it is expensive and most of the floor does not need it.

Waviness and the rolling-traffic problem

Flatness as FF does not catch everything a rolling rack feels. FF is built on the 12 inch slope change, so it is sensitive to short bumps. The waves that actually rock a tall rack on casters are longer, roughly the 2 to 10 ft wavelengths, and a floor can hold a respectable FF while still having gentle long waves that make a heavy rack pitch and wander as it rolls.

That is the gap ASTM E1486 and the waviness index were built to close. The waviness index looks at deviation from the midpoints of 2, 4, 6, 8, and 10 ft chords, so it sees the mid-wave ride that FF and FL miss. On a high-density hall where racks are heavy and rolled into place, that mid-wave ride is what bends a frame or makes a cabinet hard to position, so a waviness or ride criterion can matter more than the headline FF.

If the hall moves big racks on wheels, say so in the spec and pick a criterion that measures the ride, not just the bumps. A floor accepted on FF alone can still ride badly, and you find that out the day the racks roll in, which is the worst possible time to learn the floor was measured for the wrong thing.

The survey grid and the as-built surface map

An F-number test tells you the floor passed or failed. An as-built surface map tells you where. Both belong in the turnover, and on a hard-floor hall the map is what the rack-layout team actually uses. Tie the elevation survey to the building grid so a reading has an address, not just a value.

Run the survey on the column grid or a defined coordinate net, record elevation at each node relative to a fixed benchmark, and flag the highs and lows against the target plane. The result is a table and a contour map a crew can stand on. When a rack location reads low or a pedestal run will not seat, you point to the coordinate and the elevation, not to a vague low spot near the middle of the room.

The map also closes the loop on remediation. Grind a high, fill a dip, then resurvey those coordinates to show the fix took. The before-and-after at the same grid points is the evidence the floor was brought into tolerance where it mattered.

Fixing a floor that misses: grind, fill, and grout

A floor that misses is fixed by taking the highs down and bringing the lows up, and the two cost very differently. Grinding the high spots is the workhorse repair. A planetary grinder knocks down the humps that drag the flatness, and on a hard floor it also restores ride for rolling racks. Grinding is fast, it is local, and you grind to the map, hitting the flagged coordinates rather than working the whole floor.

Lows are the expensive direction. You cannot grind a dip up, so you fill it, with a cementitious self-leveling underlayment for a broad low area or a skim and feather for a shallow one. Self-leveling fixes levelness and flatness together when it is placed right, but it adds a bonded topping that has its own thickness, cure, and bond requirements, and a feathered edge that is too thin will debond and you will be back. Under an access floor, a localized low under a pedestal is often handled by grouting or shimming under that pedestal rather than fixing the slab, which is cheaper when only a few pedestals are affected.

The blunt part: fixing a floor after the fact costs multiples of getting the pour right, and it eats schedule at the worst time, after the slab is supposed to be done and other trades are waiting on the room. Grinding a few highs is cheap. Re-topping a whole bay that was finished out of tolerance is a real number and a real delay. The cheapest flatness is the kind you specify, finish, and test inside 72 hours, so the finishing crew owns and corrects it while they are still on site.

What to document

An F-number result with no test record behind it cannot be defended when a rack location reads low months later. Capture the conditions, the layout, the values against both the SOV and the MLV, and the as-built map, so a reviewer can reproduce the call and a deployment team can use it.

Record the floor type and whether it is on grade or suspended, the specified SOV and MLV for FF and FL, the test date and the elapsed time from finishing, the E1155 edition and instrument, the sample-line layout, the composite result, the per-section results with any section below the MLV called out, the remediation and the retest, and the as-built survey tied to the grid. For the finished access floor, keep the separate levelness report with the in-10-ft, overall, and across-joint numbers.

Common mistakes

• Testing too late, past the 72 hour window, so curling and traffic damage muddy who owns the result.
• Testing a suspended slab after the shoring is removed and reporting an FL that the method does not support.
• Confusing FF and FL, or specifying one and assuming it covers the other.
• Applying FL to a suspended structural deck instead of a survey tolerance set by the structural spec.
• Accepting on the SOV alone and signing for a section that fails the MLV where the racks will sit.
• Treating the finished access-floor levelness as the same check as the slab F-numbers.
• Specifying superflat or defined-traffic numbers for a whole random-traffic hall that does not need them.
• Accepting a hard floor on FF alone when rolling racks need a waviness or ride criterion too.

Field checklist

Run the floor through this before the slab is accepted and before the access floor goes in.

Standards and references

ASTM E1155 is the test method for the F-number system, the procedure that produces FF floor flatness and FL floor levelness from a profiler run along sample lines. It is the method a data center spec means when it calls F-numbers, and it carries the limitations that matter: the measurement is taken within 72 hours of finishing and before shoring removal, FL is valid only on slabs on grade and on shored suspended slabs, and the system covers randomly trafficked floors. Confirm the adopted edition, because the layout rules and timing language have been revised across versions.

ACI 117, the tolerances specification, frames the F-number tolerance levels and the specified overall value and minimum local value relationship, while ACI 302, the guide for concrete floor and slab construction, covers how the floor is built and finished to reach them. Both are referenced by topic here; verify the section and edition the project adopted before citing a specific value. For defined-traffic and mid-wavelength ride, ASTM E1486 covers waviness, wheel-path, and levelness criteria that FF and FL do not sense.

The finished access floor is a different document set. Access-floor levelness traces to CISCA recommended practice and the access-floor manufacturer's installation requirements, written as a planar tolerance in inches per distance rather than as an F-number. Across all of it, the project specification and the manufacturer's instructions control where they are stricter than the reference standards, and the structural engineer sets the survey tolerance for any suspended deck where FL does not apply.

Units, terms, and conversions

Floor tolerance numbers come in two unrelated systems, and the same floor reads differently depending on which the spec wrote. F-numbers (FF and FL) are unitless statistical indices from ASTM E1155, higher being better and linear in scale. Straightedge tolerances are a physical gap, written as inches in a given length, like 0.060 in in 10 ft, and these are what access-floor and many flooring specs use.

Keep the two straight, because a floor specified in F-numbers cannot be accepted with a straightedge gap and the reverse, and a target written in one system and judged in the other is a dispute waiting to happen. Metric work uses millimeters for the gap and 3 m for the 10 ft reference length; E1155 also exists in a metric companion.

FF (floor flatness)

ASTM E1155 index of short-distance bumpiness from slope change over 12 in intervals; higher is flatter

FL (floor levelness)

ASTM E1155 index of tilt and dishing from elevation difference over 10 ft; valid only on grade or shored slabs

F-number

Unitless, linear statistical index for flatness or levelness; an FF50 floor is twice as flat as FF25

SOV (specified overall value)

The F-number required for the whole placement, the composite across all test sections

MLV (minimum local value)

The lowest F-number allowed in any single test section, so an average cannot hide a local failure

Waviness index

ASTM E1486 measure of mid-wavelength deviation, 2 to 10 ft, that affects rolling ride and FF does not sense

Random vs defined traffic

Random traffic moves any direction and uses FF/FL; defined traffic follows a fixed wheel path and uses F-min or waviness

in per 10 ft

Straightedge-style levelness tolerance, the allowed gap in a length, used for finished access floors, not an F-number

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What load rating does a data center raised floor need?
• What is a concentrated load rating on an access floor?
• Cast-in vs post-installed anchors: which should I use?
• What is the difference between cold and hot commissioning of a BESS?
• What is a CDU in a data center?