Industries / Industrial equipment and services / Industrial services / Calibration services
Scheduling and records
Scheduling and records is the office side of calibration: keeping a register of every instrument, deciding how often each must be checked, recalling it when it falls due, and keeping each result as evidence.
5
Milestones achieved, from 1980 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1980
Recall made mandatory
A US military standard, MIL-STD-45662, required defense contractors to keep a written calibration system, to set each instrument's interval by its stability, purpose and use, to shorten the interval when earlier results showed drift, and to run "a recall system for the mandatory recall" of instruments when due. It replaced MIL-C-45662A of 9 February 1962. Its 1988 revision defined reliability as the probability that an instrument stays in tolerance throughout its interval, and in February 1995 the Defense Department cancelled it and pointed buyers to ISO 10012-1 and ANSI/NCSL Z540-1. A NASA paper of 1984 shows how uneven practice still was: NASA's own field centers ranged from "no instrument recall program" to systems tracking several thousand instruments.
Credit
US Department of Defense, with the US Army Missile Command as the standard's keeper; Earl German and Frederick Kern of NASA Langley Research Center for the 1984 survey
1987
Intervals from incomplete records
A calibration record says only whether an instrument was in or out of tolerance on the day it was checked, not when it drifted out. Dennis Jackson of the Naval Weapons Station at Seal Beach and Howard Castrup of Science Applications International Corporation gave a maximum-likelihood method, which picks the drift rate that makes the observed record most probable, for exactly this kind of data, so that a lab could set a group's interval to meet a reliability target such as 95% still in tolerance at recall. Statistical methods of this kind now sit in RP-1, the recommended practice on intervals from NCSL International, the calibration labs' association; Sandia National Laboratories' free, open-source Suncal program implements its test-interval method (A3) and binomial method (S2).
Credit
Dennis H. Jackson and Howard T. Castrup; Collin Delker of Sandia's Primary Standards Lab for Suncal
2007
Four ways to set intervals
ILAC, the international body of laboratory accreditors, and OIML, the intergovernmental organization for legal metrology, issued a joint guide, ILAC-G24/OIML D 10, replacing OIML's edition of 1984. It describes four methods: lengthen or shorten each instrument's interval after every calibration (the "staircase"), chart its results over time, count hours of use instead of calendar time, or check a few key readings with a simple test box between full calibrations. It says there is no "universally applicable single best practice" and that intervals fixed by "engineering intuition" and never reviewed are not reliable enough. The December 2022 edition kept the four and added a group of other statistical methods.
Credit
ILAC's Accreditation Committee and OIML Technical Committee 4
2018
A skipped step voids tests
New Jersey's Alcotest breath-testing machines, made by Draeger, must be recalibrated every six months. Marc Dennis, a New Jersey State Police coordinator who did this in five counties for over seven years, was indicted in 2016 for skipping one step and certifying that he had done it: measuring the test solution's temperature with a thermometer traceable to NIST standards, the only device in the procedure not made and calibrated by Draeger. On 13 November 2018 the New Jersey Supreme Court held that results from machines calibrated without that check are inadmissible, which put in question the breath samples of 20,667 people.
LessonA calibration record is worth only the steps actually done, and the one check that is independent of the instrument's maker is the one that cannot be skipped.
Credit
New Jersey Supreme Court in State v. Cassidy, adopting the report of Special Master Joseph F. Lisa
2021
NIST tests the method
NIST applied the RP-1 interval method to its own calibration records for two models of radiation survey meter and an electronic dosimeter, aiming at 95% in tolerance. For one survey meter, 107 of 128 readings fell within the 20% limit set by Nuclear Regulatory Commission rules and the method kept the one-year interval; with a 30% limit only one reading, which NIST thought was a data-entry mistake, fell outside, and the method would have recommended two years. NIST also said it keeps paper logbooks for all its instruments "dating from decades ago" and left them out because converting them was too labor intensive.
LessonAn interval method is only as good as the records fed to it, and records kept on paper rarely get used.
Credit
L. Pibida, B. Estes, M. Mejias and G. Klemic, NIST
Watch for
Intervals by serial number
The US Air Force Metrology and Calibration program (AFMETCAL) looks after about 700,000 test instruments with 500,000 calibrations a year in 65 labs. Its staff wrote that its intervals are set for broad groups of models from pass-or-fail results, and that it plans to set them instead from each serial number's test-point readings, using the RP-1 exponential method plus machine-learning tools that flag unusual instruments (isolation forest) and judge temporary extensions (gradient-boosted trees).
Achieved whenthe Air Force or an outside reviewer publishes in-tolerance rates for instruments on the new intervals next to those on the old ones.
Claimed by
Jeremy Latsko and three colleagues at AFMETCAL, at the IMEKO TC6 conference, September 2025
Open
01
Test intervals on air monitors
A public test bed for interval-setting methods, built on US air-quality monitors. At least every two weeks, each regulatory gas analyzer must be challenged with a known concentration and the result reported to the EPA, so each analyzer has a long public record of as-found checks. Thin each record to longer intervals and count how many out-of-limit checks each method's recommended interval would have missed. This is analysis of public data only and proposes no change to any monitoring rule.
The briefCredit · unclaimed02
Logbooks into interval data
A pipeline that turns scanned pages of handwritten calibration logbooks into a table of instrument, date, reading, tolerance and in-or-out result, flags every value it is unsure of for a person to check, and passes the result to an interval method.
The briefCredit · unclaimed
Open 01 · the brief
Test intervals on air monitors
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test bed for interval-setting methods, built on US air-quality monitors. At least every two weeks, each regulatory gas analyzer must be challenged with a known concentration and the result reported to the EPA, so each analyzer has a long public record of as-found checks. Thin each record to longer intervals and count how many out-of-limit checks each method's recommended interval would have missed. This is analysis of public data only and proposes no change to any monitoring rule.
Why it is still open
Interval methods have been tested on small or invented data. NIST's 2021 test used three instrument types; the Air Force says its own analysis lacks serial-number test-point data; KRISS's open interval code and a synthetic set published in September 2026 (961,908 check readings) use made-up records. A search found no study that uses the EPA's one-point QC records to compare interval methods.
Built with what exists
The EPA's AQS Data Mart API (free key) for "QA One Point QC Raw Data" and monitor records that name each analyzer's method; the EPA QA Handbook's ozone limit of 1.5 ppb or 7%, whichever is greater; Sandia's open-source Suncal for the RP-1 A3 and S2 methods; KRISS's open interval code (github.com/hyungseokshim/cal_intv_mcmc); Python with pandas and scikit-learn.
First step
Register for an AQS API key, download one state's ozone one-point QC checks for 2015 to 2025, and plot each analyzer's percent difference against days since its previous check.
Done looks like
For every ozone analyzer with at least five years of checks, a published table giving its observed out-of-limit rate; each method's recommended interval per analyzer and per analyzer model; and the share of out-of-limit checks that each recommended interval would have missed when the record is thinned to it, with code and data public.
Open 02 · the brief
Logbooks into interval data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A pipeline that turns scanned pages of handwritten calibration logbooks into a table of instrument, date, reading, tolerance and in-or-out result, flags every value it is unsure of for a person to check, and passes the result to an interval method.
Why it is still open
NIST said in 2021 that decades of its paper calibration logbooks went unused because converting them was too labor intensive. Public work on handwritten logbooks is general: an open pipeline for industrial process logbooks (2026) and a hobby comparison of eight vision models on three handwritten flight-logbook photos with 37 flights (October 2026). Neither reads calibration readings or reports how often a misread turns a pass into a fail.
Built with what exists
A vision-language model that reads page images, such as Claude (see the Note "Seven more flagged") or Gemini; Transkribus or Microsoft's open TrOCR model for handwriting; a person re-keying a random sample as a check; Sandia's Suncal for the interval step.
First step
Ask a university radiation-safety office or instrument workshop for permission to scan one old calibration logbook, and transcribe 20 pages both by hand and by model to measure a first error rate.
Done looks like
At least 2,000 logbook entries from one or more labs, transcribed with the owner's permission; a random 10% re-keyed independently by two people; the field error rate and the rate of wrong pass-or-fail results published, with and without the person-check step; and, for each instrument model with enough history, the interval the A3 method recommends next to the interval in use.