Checking code: invisible twice

Andrei Niasiuk · August 12, 2026 · Facts checked: August 15, 2026

Four university time-norm documents from Russia and Belarus set a formal, paid rate for grading a student’s written work: depending on the document and the kind of work, from 0.25 to 1 academic hour. Not one of the four gives programming code a line of its own: it is checked under the same rate as an essay or a typical calculation. Britain, which does give marking explicit lines of its own (thirty minutes per essay at Reading, an hour per student at Lancashire), still leaves code uncounted. Turn to the burnout studies surveyed here and grading never appears as a variable of its own: workload shows up as a single aggregate figure, and the hours a person spends reading someone else’s code disappear into it the same way they disappear into the payroll spreadsheet. The work sits in a gap that neither the accountants nor the researchers have ever measured on its own.

A companion review on this site asked what the evidence shows about whether automated grading teaches programming better; it found more than six decades of the question being asked and never quite answered. What that review left open is the cost side: what checking code takes out of the person doing it. Almost nobody has measured that.

What the norms actually say

Russia: one shared category for everything written

HSE’s time norms, in force since 2004, list checking of “essays, homework, control assignments, and reports” (“проверка: эссе, домашних заданий, контрольных работ, рефератов”) at 0.3 hour per assignment, with reports carrying their own higher rate (0.75 hour for bachelor’s and specialist programs, 1 hour for master’s and postgraduate) and a stated ratio of one graded work per 24 hours of scheduled class time (HSE, 2004). SPbSUT’s norms, updated in 2017-2018, land on the identical figure for a differently worded category: “checking, consultation, and acceptance of control, calculation, and calculation-graphic assignments,” 0.3 hour per assignment, capped at 1 hour per student per course per semester (SPbSUT). Both documents were adopted independently and arrive at the same order of magnitude. A lab report containing a program is checked, in either one, under the same rate as a written control assignment containing none.

Belarus, in two tiers

The republic-wide Ministry of Education norm, in force since 2018, sets a range rather than a fixed number: checking of “control assignments, including calculation-graphic and calculation assignments (typical calculations)… provided for by the academic program” runs “from 0.35 to 0.5 hour per assignment, no more than 1 hour per student per academic discipline (module) per semester” (Ministry of Education of the Republic of Belarus, 2018). Individual universities then fix their own rate inside that republican range. The Belarusian State Academy of Communications did so by institutional order dated 24 April 2025, built on the Ministry’s current 2023-2024 resolutions, and its version does something none of the other three documents does: it names laboratory work directly. “Checking control assignments, typical calculations, calculation-graphic assignments, essays, reports on completed laboratory research: 0.4 hour per student / 0.25 hour per student per academic discipline, module” (BSAC, order No. 126, 2025). Of the four documents examined here, this is the only one where a lab report gets its own words in the regulation, rather than hiding inside “control assignments” by implication. It still doesn’t say “code”: a lab report in electrical engineering or a lab report in a programming course draws the same 0.4 hour.

Where the norm stops and the code review starts

At the Faculty of Applied Mathematics and Informatics of Belarusian State University, an automated checking pipeline (iRunner, built in-house) has graded first- and second-year programming assignments since 2020. A 2020 account of the rollout describes what happens after the automated part finishes: “the last submitted versions from each participant undergo manual review: at this point, what cannot be formalized and automated is checked (the optimality of the idea, the quality of the source code, and so on)” (Ilyin, 2020; original: “…в этот момент проверяется то, что невозможно формализовать и автоматизировать”). The same account names a specific gap in the tooling: “the system has a significant shortcoming: the absence of a full-fledged subsystem for code review” (original: “отсутствие полноценной подсистемы для рецензирования кода (code review)”). Automating the test-running did not remove the human reading; it relocated it to a step none of the four regulations describes, because none of them was written with a step like it in mind.

Widening the lens

Russia and Belarus are not special cases. The same absence recurs in every system checked here, and it arrives by a different route each time.

The United Kingdom has no national norm; each university sets its own Workload Allocation Model, and the country runs two incompatible designs side by side. University of Reading spells checking out to the minute: “Any piece of student writing (essay, report) at the length of 2500 words: 30 minutes per [piece],” and marks exam scripts at four per hour, a rate set for one-hour exam papers (University of Reading). Lancashire UCU sets a coarser but still explicit rate: “1 hour per student per 20 credit module” (Lancashire UCU, 2023). Manchester, Aberdeen, and Bristol bundle it instead: Manchester adds “0.6 of an hour” per hour of direct teaching, for “preparation, assessment, marking and pastoral guidance” together, no line for any of them (Greater Manchester WLA Framework, 2023).

France goes past silence into a stated exclusion. The national référentiel (arrêté du 31 juillet 2009) lists exactly which activities convert into teaching hours, and correction of any kind isn’t among them. Université Toulouse III spells out why: “les corrections et évaluations… font partie intégrante de la tâche d’enseignement… et ne sont donc naturellement pas considérées dans le présent référentiel” (“checking and evaluation… are an integral part of the teaching task… and are therefore not considered in this référentiel”) (Toulouse III, 2025).

Germany regulates teaching load at state level; the ordinance examined here, North Rhine-Westphalia’s LVV, enumerates what converts into teaching hours (SWS): lectures, seminars, practicals, up to three hours for thesis supervision. “Korrektur” appears nowhere in it. A secondary source, not itself a regulation, states the logic: “Forschung, Prüfungen, Selbstverwaltung und Betreuung sind Teil des Amts, fließen aber nicht in die SWS-Zahl ein” (research, examinations, self-administration, and supervision are part of the post, but they don’t enter the SWS count) (wissenschaftsstellen.de).

Poland, seen here through Politechnika Wrocławska’s rules, comes closest to naming the work directly without pricing it. The university lists “sprawdzanie prac kontrolnych studentów” (checking students’ control assignments) as an explicit duty, but places it poza pensum, outside the counted hours, with no number attached anywhere, unlike thesis supervision in the same document, quantified at 10 to 15 hours (PWr, §1.2.2.1).

The United States barely regulates any of this. Saint Louis University’s CS department norms staffing, not hours: “at least 1 grader or TA per 30 students” (SLU CS, 2022), and the University of New Mexico’s equivalent CS workload policy doesn’t mention grading once (UNM CS, 2022). What forced the question into the open was litigation: Long Beach City College adjuncts sued, saying they “weren’t compensated for hours of work outside the classroom, including grading assignments,” and the district’s board approved an $18 million settlement in January 2026, pending final court approval (Inside Higher Ed, 2026). The plaintiffs were hourly-paid adjuncts at a community college, the corner of US higher education where unpaid hours cut most directly into take-home pay, and so the first place the missing line item turned into a legal claim rather than a footnote.

Seven systems now, and in none of them does checking a program earn a line of its own. That is the one constant. The mechanism is where they diverge. Reading spells the minutes out and still stops short of code; Manchester folds everything into one coefficient with no line for anything inside it; France writes the exclusion down; the United States had no norm at all until a lawsuit produced one. Four mechanisms in the documents examined, and nothing here says the list is complete.

What the burnout literature does and doesn’t show

The case for “workload matters”

Any workload claim about Russian faculty has to be read against the scale gap the literature itself points out: “if the ordinary teaching load of a Russian university instructor is 900 academic hours, then abroad it is approximately 200-300 hours a year” (Ershtein, 2021; original: “если обычная нагрузка российских преподавателей вузов составляет 900 академических часов, то за рубежом она примерно равна 200-300 ч в год”). Against that backdrop, “workload causes burnout” reads as close to common sense, and the international literature backs the broad shape of it. A 2025 systematic review covering sixty peer-reviewed studies and roughly 43,600 university faculty found workload among the most consistent predictors: “the strong association between burnout syndrome and excessive workload…” (Cadena-Povea et al., 2025), though the same review rates the overall certainty of that evidence as only “moderate to low.” A single-institution study at a Philippine university (N = 35) reports the correlation as a number: workload and burnout at r = 0.698, p < .001 (Beltran-Salipong, 2025).

The case against isolating it

“Workload” in every one of these studies is an aggregate: teaching, prep, administration, sometimes research, bundled into one self-reported number. Where a study did try to separate task volume from other stressors, the result cuts the other way: a 2026 study at a Philippine college found “task volume (r = 0.229) and time pressure (r = 0.309) were not significantly associated with job burnout, whereas role conflict and work schedule (r = 0.498) were” (Cañizares et al., 2026), concluding that structural factors, not sheer quantity of work, carried the explanatory weight. A Novosibirsk survey of faculty across several universities asked directly what caused burnout and ranked the answers: new requirements imposed on faculty came first (76.5%), pay came second (58.8%), and classroom workload came third (41.2%) (Bogdan & Samsonova, 2020). Workload matters in this literature. It is rarely the top-ranked cause, and nothing in it distinguishes grading from lecturing, committee work, or anything else that fills the hours.

Does grading crowd out research?

The related hypothesis is displacement: hours spent grading come directly out of hours spent on research. A 251-respondent study at Politecnico di Torino ran a correlation analysis specifically to check that and reported that the analysis “excludes… the existence of a negative link in terms of workload…” (Maisano, Mastrogiacomo & Franceschini, 2023), while flagging that the finding is limited to one Italian institution. A larger Chinese study of 917 faculty at 21 research universities found a real negative relationship between curriculum-based teaching time and research output, but only within the subgroup of assistant professors, not across the whole sample (Li & Yang, 2023/2024). And the one time-series long enough to check the trend directly, HSE’s Monitoring of Education Economics survey of Russian faculty from 2006 to 2009, found research hours declining modestly (“10.1 hours in 2006 and 9.3 hours in 2009,” a change the survey itself describes as “somewhat less”) while classroom teaching hours rose over the same period, from 16.3 to 17.8 per week (HSE Monitoring, 2006-2009). That is movement in the direction the displacement story predicts, over a window too short and a decline too shallow to confirm it. The displacement story is a real hypothesis with a real first source: practitioners of workload accounting at Ural Federal University wrote it plainly in 2015, arguing faculty have “simply no time” for research once overloaded on classroom hours (Yashin & Strukova, 2015). But that is an expert’s stated judgment, not a measurement, and where anyone did measure it, the answer split by country and by seniority rather than confirming it outright.

The gap as a measurement problem

Put the two halves together and the shape of the missing study becomes visible. No study surveyed here has logged grading, specifically the act of reading and running a student’s program, as its own time variable, separate from lecture prep or from grading a written exam. Where an institution did automate, as BSU did with iRunner in 2020, the before-and-after survives only as a memory of how the hours used to feel. Cañizares et al. showed what the alternative looks like when they ran a validated burnout instrument against task volume and role conflict separately rather than as one number; nobody has yet run one against grading.

What that would take is not exotic: submission-system timestamps already log when a program was opened, run, and scored, which is closer to an objective time record than any self-reported workload survey in this corpus. Pairing that log against a standard burnout inventory, on a real course, before and after an automation change, would answer a question the systematic reviews keep restating rather than closing. That study has not been run. Until someone runs it, the line item stays missing twice over: in the ledger and in the literature.

This piece draws on official time-norm regulations from four Russian and Belarusian universities and on burnout research collected across English- and Russian-language literature. It was compiled by Andrei Niasiuk, who is building AutoLabSuite, a platform for grading programming lab assignments in university courses.