An automatic level showing more than about 0.05 ft per 100 ft (6 mm per 30 m) of collimation error on a two-peg test has crossed the common field recalibration threshold for general construction work, and that single number drives most site decisions about when to pull an instrument off the job [S5].
The drift chain is mechanical and well understood: shipping impact, site vibration, temperature swings, and dropped tools shift the line of sight away from true horizontal, and the compensator that is supposed to absorb that disturbance has a finite correction range. Once the error exceeds the project tolerance, every subsequent reading is suspect until the instrument is verified, adjusted, and re-verified [S1][S3].
What "Out of Spec" Actually Means on an Automatic Level
An out-of-tolerance finding on a calibrated level is defined as the as-found reading exceeding the stated accuracy specification on at least one test point, before any adjustment is performed, and that as-found value is the legally and technically important number for assessing prior work [S6]. The lab under ISO/IEC 17025 documents the observed error, the measurement uncertainty, the environmental conditions, and the standards used, so the certificate can be compared across facilities and accepted by auditors worldwide [S2]. The as-found versus as-left distinction matters because the as-found result tells you whether the measurements taken between the last calibration and this one were reliable, while the as-left result only documents the instrument after the technician has touched it [S3]. For a builder's automatic level rated to ±1.5 mm per km, even a 2 mm drift per double-run is enough to invalidate a final elevation submittal on a precision slab pour.
The Two-Peg Test as the Working Tolerance Gate
The two-peg test is the standard field check for collimation error, and it is the gate that decides whether a recalibration ticket gets cut. Set two pegs 100 to 200 ft apart on level ground, take rod readings at each peg from each of two instrument setups, and the four readings reveal any systematic angular error between the line of sight and true horizontal [S5]. The pass or fail then rides on a tolerance chosen for the work: 0.05 ft/100 ft (6 mm/30 m) for general construction, 0.02 ft/100 ft (2 mm/30 m) for road and site grading, and 0.005 ft/100 ft (0.5 mm/30 m) for precise leveling [S5]. Topcon recommends monthly field checks for instruments in daily service, while Leica specifies a minimum of quarterly verification, so most quality programs layer a monthly two-peg test on top of an annual or semiannual accredited lab calibration [S3][S5]. Reading the spec without grounding it in a peg test is the classic mistake: a clear 32x magnified image does not fix a misaligned compensator, and the resolution versus accuracy gap is where most field disputes start [S1].
Compensator Quality Sets the Real-World Drift Rate

The compensator, not the telescope magnification, is the part that determines how fast an automatic level drifts out of spec under site vibration. Magnetic-damped and air-damped compensators differ in how well they absorb vibration, and that difference is the dominant variable in long-term field accuracy between two instruments that share the same ±1.5 mm/km factory rating [S1]. For a deeper look at how magnetic and air-damped compensators compare in arc seconds, the spec-decision map at automatic level compensator accuracy in arc seconds breaks the rating down into usable field numbers. Drift compounds quietly: a small error that stays under the project tolerance for a 30 m sight becomes a much larger cumulative error on a 1 km level run, which is why a 180-degree reversal test on a spirit level, or its kinematic equivalent on a digital level, has to be repeated on a fixed schedule rather than performed once at receipt [S2][S3].
What Triggers a Recalibration Outside the Calendar
Scheduled intervals are a backstop, not the primary trigger. A level should be pulled and recalibrated after any drop or impact, after a vial, sensor, or frame repair, after shipping or relocation, after a field accuracy check fails, and any time an audit or customer contract requires a current certificate [S3]. For an automatic level, a two-peg test that fails the project tolerance is the same event as a "field accuracy check fails," and the instrument should go straight to the lab rather than be field-adjusted and put back in service, unless the technician can demonstrate the field adjustment by re-running the peg test inside tolerance [S3][S5]. Out-of-tolerance events are not rare, and the cost of each one runs into thousands of dollars when you include the investigation, the affected product quarantine, and the retrospective review of measurements taken since the last good calibration [S4].
Comparison: Common Field Tolerances vs. Required Work

The tolerance chosen for the two-peg test has to match the work, and the four practical bands used across construction and precision leveling are tight enough to pick the right one without a calculator. General construction and utility grading run at 0.05 ft/100 ft (about 6 mm/30 m, equivalent to roughly 2 mm per standard 20 m sight), road and site grading tighten to 0.02 ft/100 ft (2 mm/30 m), and precise leveling, where first-order benchmarks and machine alignment live, sits at 0.005 ft/100 ft (0.5 mm/30 m) [S5]. A factory-fresh automatic level typically ships tighter than any of those, but the moment it leaves the case the spec is a baseline, not a guarantee, because transit shock and field vibration move the line of sight long before the technician sees a bubble shift [S1]. Choosing the looser tolerance on a slab pour is one of the more expensive ways to fail a pre-pour check, because the rework cost dwarfs the cost of running a tighter peg test the morning of the pour.
How Labs Verify, and What the Certificate Must Show
An accredited calibration under ISO/IEC 17025 verifies four distinct behaviors on a digital or automatic level, and the certificate has to record each one for the result to be defensible. The lab checks zero and level-point accuracy against a traceable horizontal reference, then runs angle, slope, and range checks across the stated measuring range because a level that reads correctly at zero can still accumulate measurable error at 5 or 10 degrees [S2]. Repeatability and display stability are verified by taking multiple readings at the same test point without moving the instrument, which catches jitter that a single-shot reading would miss entirely, and a base-condition check confirms that the reference surface the instrument sits on is not biasing the result [S2]. The final certificate must list observed error at each test point, measurement uncertainty, environmental conditions, the reference standards used, and a clear pass or fail against the specified tolerance, otherwise it cannot be compared against another lab's result or accepted by an external auditor [S2][S6].
OOT Response: The Six-Step Path Back to a Valid Certificate

Once an instrument is found out of tolerance, the response is a defined procedure, not a field decision. The six-step plan documented for out-of-tolerance findings starts with quarantining the instrument and any work performed since the last in-tolerance calibration, then notifying the customer or quality owner, then performing a failure investigation on the suspect test points, then assessing the impact on previously reported results, then adjusting and re-verifying the instrument, and finally updating the calibration program with whatever the failure mode revealed [S6]. The retrospective assessment is the step most often skipped, and it is the step that determines whether the company is paying for a recalibration or paying for a product recall, because the as-found data is what tells you how long the instrument was out of spec before the discovery [S3][S6]. To keep OOT events down in the first place, the most cost-effective moves are usually the unglamorous ones: proper handling, correct storage, scheduled preventive maintenance tied to the calibration interval, and a measurement program that Pareto-analyzes failure modes by instrument family rather than by vendor [S4].
Track these signals over the next quarter: any tightening of factory collimation tolerance below 1.5 mm/km on new automatic-level SKUs, any update to ISO/IEC 17025 lab requirements that changes the uncertainty reporting format on calibration certificates, and any movement in the monthly-versus-quarterly field-check guidance from the major automatic-level OEMs, since those three together set the practical recalibration cadence on most sites. For a broader look at how an automatic level's self-compensating telescope establishes the horizontal reference in the first place, the working-principle walkthrough at automatic level: how a self-compensating telescope sets the horizontal reference covers the optical chain that the tolerance numbers are trying to protect.
Detailed specification references: automatic level, automatic molding line, and infrared level.