← mariuscomper.uk
RO · EN

Romania · baccalaureate 2015–2026 · June session · 3.36 million grades

Four point nine five

For years, almost no paper in Romania's school-leaving exam stopped at 4.95, five hundredths below the pass mark. In 2015 there was one 4.95 for every 302 papers at 5.00. Since 2024, when every June paper has been marked on a digital platform, there is one for every two.

The gap below 5 filled in the same year in all 42 counties, while the control thresholds, 6.50 and 7.00, did not move. The change coincided with digitised marking; the data do not show that it caused it. Nor did the pile at 5.00 disappear: appeals now rebuild a large part of it.

2015

302 times as many papers at 5.00 as at 4.95

Outline: 2024
Papers marked in June, first marking, without recognised grades
YearAt 4.95At 5.00Papers at 5.00 per paper at 4.95
20154714,181302
20184310,044234
20206111,467188
2021668,905135
2022997,56776
2023918,39992
20249842,1022.1
20258681,8462.1
20261,0832,3912.2
Grades at first marking from 4.50 to 5.45, per 10,000 grades, June session, without grades recognised from earlier years. Solid bars show the chosen year, the outline the year it is compared with. 2016, 2017 and 2019 are missing because their files cannot be read reliably.

Correction, 6 November 2026. The first version of this page said that 5.00 was the most common grade in the 2024 exam and that it was given 7.68 times as often as 4.95. That count included recognised grades: candidates who resit the exam keep the grades of subjects they passed in earlier years, and those papers were not marked in 2024. In June 2024 alone, 3,398 grades of exactly 5.00 were of this kind, and none at 4.95. On papers marked in 2024 the ratio is 2.1, and the most common grade is 10. Part one and every June 2024 distribution are now counted only on papers marked in the year in question; for August, where the published results do not flag recognised grades, only grades carried over from June were removed. The figures on second readings are unchanged with one exception: at 5.00, 19 candidates in 100 asked for a re-mark, not 6.

I · The hole

A step almost nobody stood on

Romania's baccalaureate, the exam that ends secondary school, is graded from 1 to 10, and a candidate needs at least 5 in every written subject and an average of at least 6 to pass. Each paper's grade is the average of two teachers' marks, kept to two decimals and never rounded, so it can land on any step of five hundredths. Neighbouring steps should hold roughly the same number of papers. In June 2015, of the 454,348 grades given to papers marked that year, 14,181 were exactly 5.00. Only 47 were 4.95.

In 26 of Romania's 42 counties there was not a single 4.95. The ten steps from 4.50 to 4.95 together held under 1 per cent of all grades; the single step at 5.00 held 3.1 per cent.

The pile did not sit at 5.00 alone. Compared with a smooth curve drawn through the rest of the grades, the steps from 5.00 to 5.45 held about 19,000 papers more than they should have, and the steps below about 13,000 fewer. The distribution shows a surplus of grades above the line and a gap below it. In 2015, 5.5 per cent of all passing grades sat in that pile.

The data do not establish how the pattern formed. They show only that, close to the line, grades did not stay below it.

II · The years

From 302 to 2

The ratio of papers at 5.00 to papers at 4.95 falls slowly until 2022: 302 in 2015, 234 in 2018, 188 in 2020, 135 in 2021, 76 in 2022. In 2023 it rises to 92. In 2024 it drops to 2.1 and stays there: 2.1 in 2025, 2.2 in 2026.

The same break shows in the shares. Until 2023 the steps from 4.50 to 4.95 held between 0.85 and 1.27 per cent of grades; since 2024 they have held between 2.9 and 3.3 per cent. The step at 5.00 held between 2.2 and 3.1 per cent; since 2024 it has held 0.6 per cent.

Papers at 5.00 for every one at 4.95, year by year

The scale is logarithmic: each horizontal line multiplies. Grey bands are missing years.

302 (2015), 234 (2018), 188 (2020), 135 (2021), 76 (2022), 92 (2023), 2.1 (2024), 2.1 (2025), 2.2 (2026).

Your year

Pick the year you sat the baccalaureate and, if you like, the subject. See how many papers stood at 4.95 and how many at 5.00 that year, next to 2026.

Subject

Pick a year.

For each year, the numbers are in the table under the chart at the top.

III · What changed

Scanned in the room, sent at random

Romania's Ministry of Education tried digitised marking in the June–July sessions of 2022 and 2023, for candidates from two counties, Călărași and Dâmbovița. From the August–September 2023 resit session it used it for every paper. June 2024 was the first summer session marked this way across the country.

The ministry's release of 14 August 2023 sets out what changed. Papers are scanned in the exam room at the end of the test, with the candidate present. They are anonymised automatically on upload. The platform then sends each paper, "automatically and at random", to two markers in a county other than the candidate's. The same release says the marking schemes and the types of question do not change.

Two things were not new. Papers were anonymous before, with the corner bearing the candidate's name sealed down. And marking in another county had been ordered in 2015, when paper scripts were driven from one county to another. What was new was the paper's route: scanned in front of the candidate, no paper scripts carried between counties, assigned at random by the platform.

The 2024 break appears in every one of the 42 counties: in all of them the ratio fell from 2023. The pilot counties fit only half the story. In Călărași the ratio fell as early as 2022, to 1.8, on small numbers: 16 papers at 4.95. In Dâmbovița it fell only to 40 in 2022 and 16 in 2023, then to 1.9 in 2024. And Buzău, which was not a pilot county, fell to 4.3 in 2023, with no explanation in public documents.

A study of New York's Regents exams, once marked by teachers in their own schools, measured something similar. In New York City schools that later joined a pilot of centralised marking, it estimated that 73.2 per cent of exams scoring just below the pass mark of 65 were manipulated in June 2010; 17.2 per cent in 2011, after re-scoring near the cutoff was banned; and 0.44 per cent in 2012, under centralised marking, a figure not distinguishable from zero (Dee, Dobbie, Jacob and Rockoff, 2019). It is a parallel, not an explanation for Romania.

What can be said: the gap below 5 filled in the first June session marked on the platform across the country. What cannot be said: that the platform filled it.

Your county, year by year

Grades from 4.50 to 5.45 for candidates from one county. Under each chart: papers at 5.00 / papers at 4.95. Counties are not ranked: the differences between them did not hold from one year to the next, so they do not make a league table.

The ratio by year, for the whole country, is in the table under the chart at the top.

IV · The control thresholds

At 6.50 and 7.00 nothing happened

If the way all grades are given had changed in 2024, it should show elsewhere on the scale too. The same measurement, applied around 6.50 and 7.00, where nothing is won or lost, finds no pile in any year, before 2024 or after.

Measured against the smooth curve, the pile at 5.00 was worth 8.8 steps of papers in 2015, 7.3 in 2018 and 6.0 in 2023. In 2024, 2025 and 2026 it is worth 1.5. It did not reach zero: even on papers marked on the platform, 5.00 is still twice as full as 4.95.

V · Appeals

The pile is rebuilt after the second reading

Until 2023, appeals added about a fifth of the final pile at 5.00. In 2018, for example, the surplus of papers just above the line was 12,925 at first marking and 15,675 after appeals. Since 2024 appeals have added almost half: 2,434 at first marking, 4,468 after.

The reason shows in the number of requests. In 2023, 3,088 papers first marked between 4.50 and 4.95 were appealed. In 2024 there were 7,871, and in 2026, 9,109. Since 2024, between 42 and 45 in 100 of them have reached 5 or above. After appeals, the ratio of 5.00 to 4.95 was 253 in 2023 and 4.9 in 2024.

First marking and final grade

The outline shows grades at first marking, the solid bars the final grades, after appeals. Appealed papers below 5 climb and fill the 5.00 step again.

Year

After appeals, the ratio of 5.00 to 4.95 was 253 in 2023 and 4.9 in 2024.

VI · The average of 6

The second line was always drawn by appeals

Passing takes more than 5 in every subject: the average must be at least 6, so the three written grades must add up to at least 18.00. At first marking, sums of 18.00 and 17.95 were almost equally common in every year from 2015 to 2026: the ratio between them sits between 0.87 and 1.21. After appeals, 18.00 comes up 2.1 to 3.7 times as often as 17.95.

Candidates averaging just under 6 nearly always appeal: in 2018 and 2024, between 90 and 93 in 100 of those averaging roughly 5.75 to 5.92 asked for a re-mark. Just above 6, between one and three in 100. Unlike 5.00, the line at 6 did not change in 2024: before and after, it was drawn by appeals, not by first marking.

VII · The cost

How many ended up on the other side

Until 2023, between 1,093 and 1,634 candidates a year failed the baccalaureate because of a single subject, with a final grade between 4.50 and 4.95 on a paper marked that year. Since 2024 there have been between 3,164 and 4,026.

Of these, about 1,140 to 1,300 a year would also have reached an average of 6 had that subject been 5.00; before, it was 375 to 587. This is an estimate under the exam's rules, not an observation: nobody knows what grade those papers would otherwise have received.

The pass rate did not fall. In the June session, 75.0 per cent of candidates who sat the exam passed in 2023, and 78.3 per cent in 2024.

For a candidate with a grade between 4.50 and 4.95, the practical advice has not changed: appeal. Since 2024, about four in ten such appealed papers have reached 5.

Part two · 2024, grade by grade

The page’s first version, redone

What follows is the page published in August 2026 about the 2024 exam, with its figures redone on papers marked that year. The chapters on second readings use both 2024 sessions, June and August.

At Romania's 2024 school-leaving exam, 5.00, the lowest mark that counts as a pass, was given twice as often as 4.95, the step immediately below it. Apart from 10, no other whole grade breaks away from its neighbour like that.

On the written papers marked in the June 2024 session, the grade 5.00 appears 2,102 times and 4.95 appears 984 times. The steps before the line fall towards it instead of rising, and right after the line the count doubles. The place where the scale breaks is exactly the grade the diploma hangs on.

4.95 → 5.00 Grades given at first marking in June 2024, between 4.00 and 6.00, without grades recognised from earlier years. Each bar is one step of five hundredths of a point. The red line is the border between passing and failing.

Grades at first marking on written papers, June 2024
GradePapers
4.801,295
4.851,211
4.901,136
4.95984
5.002,102
5.051,672
5.101,801
The column at 5.00 is 2.1 times taller than the one at 4.95, though five hundredths of a point separate the two grades.

2024 · The line

An exam that ends in a single number

Every written paper at the Romanian baccalaureate ends in a number with two decimals. The paper is scanned in the exam room at the end of the written test, with the candidate present, and travels without a name to two teachers in a county other than the candidate's own, assigned automatically. The two mark it separately, and if their marks are no more than a point apart, the paper's grade is their arithmetic mean, written to two decimals and never rounded. That is where the exam's five-hundredth steps come from.

In the June 2024 session, teachers marked 378,635 such numbers, leaving out grades recognised from earlier years. Laid out side by side they draw a predictable landscape: a broad peak among the high marks, a thin tail down towards 2 and 3, a pile-up at 10 where the scale runs out. Everywhere, neighbouring steps resemble each other. Apart from the top grade of 10, there is one place where they do not, and it sits exactly where the rules draw the border between passing and failing.

The grade 5.00 was given 2,102 times. The step below it, 4.95, was given 984 times. The most common grade in the exam is 10, with 8,553 papers.

Five hundredths of a point separate the two, and a ratio of two to one.

2024 · The slope that falls

The closer a paper sits to passing, the fewer companions it has

In an ordinary distribution the number of papers rises and falls smoothly, without jumps between adjacent steps. On the last four steps before the line, the movement runs against the shape of the hill: 1,295 papers at 4.80, then 1,211 at 4.85, then 1,136 at 4.90, then 984 at 4.95. Getting closer to a pass empties the steps instead of filling them.

The ten steps between 4.50 and 4.95 hold 11,104 papers between them. The single step at 5.00 holds 2,102, almost a fifth of everything awarded across the whole band below it.

2024 · No other whole number

Round numbers draw a few extra papers everywhere, and 5 draws differently

Whole grades have a small pull of their own, visible at every value: 4.00 appears nine per cent more often than 3.95, and 9.00 ten per cent more often than 8.95. Apart from 5 and 10, the ratios in the table sit between 1.01 and 1.11, a modest gathering. At 5 the ratio is 2.14.

Each whole grade against the step below it · written papers, first marking, June 2024
Grade0.05 lowerExactlyTimes more often
4.007097721.09
5.009842,1022.14
6.002,0162,2331.11
7.002,6272,6581.01
8.003,4803,8391.10
9.004,6275,1011.10
10.002,6078,5533.28

The only other place where papers pile up is the top mark, and there the pile explains itself: above 10 there is no step for anyone to stand on. Above 5 there are a hundred.

The ruler

Every grade in the exam, step by step. Change the subject, the session and the moment, then move in close to the line to see what happens around 5.00.

Subject
Session
Moment

Grades recognised from earlier years are left out of June. The published August results do not flag them, so only grades carried over from June were removed there; the August and Both views still contain some recognised grades.

The tables above show the same scale.

2024 · The second reading

The same paper, two other teachers, a different number

A candidate unhappy with a grade can ask for the paper to be read again. The scanned paper then goes to two different teachers, again in a county other than the candidate's own, and whatever comes out becomes the final grade, higher or lower. The student signs a form acknowledging this beforehand. In 2024, 64,383 papers went through a second reading.

Of those, 61,177, or 95 in every 100, came out with a different number than they went in with. Half moved by at least three tenths of a point, and 4,522 moved by a whole point or more. Grades rise more often than they fall, 59 per cent against 36, and the rest stay put.

These papers are not an ordinary sample: they reach a second reading because someone was unhappy with the first. What they show is how far a paper's grade can travel when it is read again from scratch, by people who did not have it in their hands the first time.

Grades after the second reading also jump at 5.00: 1,390 papers got 5.00 and 441 got 4.95, so the ratio between the passing step and the one below it stays at three to one.

What happened to papers marked like this

Pick a grade. Below you see how many papers received it at the first marking in 2024, how many were read a second time and where they ended up. For grades under 5, the last figure counts the ones that crossed the line.

Grade 4.95
Subject
—papers got this grade
—were read a second time
—of those went up
—of those went down
—came out at 5.00 or above

—

Where they landed after the second reading

The main figures on second readings are in the text above.

2024 · Who crosses the line

Three thousand two hundred and twenty-three diplomas

At the second reading, 5,467 papers went in with a grade below 5 and came out at 5 or above. Another 1,505 travelled the other way and fell under the line.

Counted in people rather than papers, the figures read like this: 3,223 candidates passed the 2024 baccalaureate because their papers were read a second time, and would have failed on the first marking. Another 24 had passed on the first marking and lost it on the second.

Of the 110,797 candidates who passed, 1,281 have at least one subject marked in 2024 that ended, after appeals, at exactly 5.00. For them, five hundredths of a point less would have meant a failed subject.

2024 · Who asks

Below the line people ask, above it far fewer do

The appetite for a second reading grows as a grade climbs towards the line without reaching it. In the June session, at 4.50, 65 candidates in 100 asked for one; at 4.80, 73 did; at 4.90 and at 4.95, 78.

At 5.00, 19 did.

Five hundredths of a point separate four requests in five from one in five.

The line, by county

The share of grades of exactly 5.00 among the grades given at first marking to each county's candidates in the June 2024 session, without recognised grades. Counties are listed in alphabetical order of their code.

The share of grades of exactly 5.00 runs from 0.3 to 1.0 per cent, depending on the county.

The share runs from 0.3 to 1.0 per cent. It depends mostly on how many of a county's papers come to rest near the line, and per county the numbers are small, in the tens of papers, so the table is not a ranking.

2024 · What is visible

A measurement written to two decimals that moves by tenths

A baccalaureate grade carries every sign of a measurement: two decimals, two independent readings, an arithmetic mean without rounding. The 2024 figures show that when the same paper reaches two other teachers a second time, the number changes in 95 cases out of 100, and half of the changes are of at least three tenths. The decimals describe more precision than the procedure behind them can deliver.

Apart from the top grade, around one single value the mechanism behaves unlike anywhere else on the scale. At the grade that separates a diploma from its absence, the step above is twice as full as the step below, and the steps beneath it thin out as they draw closer to it. The same landscape shows up at the first marking and at the second, and in 41 of the 42 counties 5.00 comes up more often than 4.95.

For a student looking at a grade between 4.50 and 4.95, the practical conclusion is simple: that grade moves, and in 2024 it moved up more often than down. For one looking at a 5.00, all that can be said is that they landed on a step twice as full as the one below it.

How to read the figures

  • The figures in part two come from the individual results of Romania's 2024 national baccalaureate, the June–July session and the August one, as published by the Ministry of Education: 144,950 distinct candidates, 23,130 of whom sat both sessions, and 495,635 grades on written papers. Of these, 23,732 in June and at least 36,610 in August are grades recognised from earlier sessions, which were not marked in 2024. For June they are flagged in the individual results file published on data.gov.ro and were removed from every distribution; for August only the grades carried over from June were removed. The sections "2024 · The second reading" and "2024 · Who crosses the line" use both sessions.
  • “First marking” is the grade before any appeal. “Final grade” is the one that enters the average. “Second reading” exists only for papers that were appealed.
  • A candidate passes with at least 5 on every written paper and an average of at least 6. The average is computed to two decimals by cutting off the rest, not by rounding.
  • The 3,223 candidates saved by a second reading are those who appear as passing in the official results but would not have met the pass condition on their first-marking grades.
  • Papers read a second time get there because the candidate asked, so they are not a random draw from all the exam's papers.
  • The 12-year story (part one) uses the individual-results files for the June session published on data.gov.ro for 2015–2026: a grade means the grade on one written paper, without grades recognised from earlier years. "First marking" is the grade before any appeal; "final grade" is the one after.
  • 2016, 2017 and 2019 are missing: their files write decimals with unquoted commas, and a read-back test recovered only 85 per cent of the 5.00s. They point the same way (ratios of 251, 247 and 168, understated) but are not printed.
  • The test was fixed before counting. The pile is measured as the surplus of papers from 5.00 to 5.45 over a smooth curve (a seventh-degree polynomial with a correction for round grades) drawn through the other grades, divided by the height of one step. The control thresholds 6.50 and 7.00 stay below 0.55 in every year. The control thresholds 4.00 and 5.50, also chosen in advance, lie in the zone the pile emptied and failed the "three times smaller" rule in 2018 and 2021–2024; the result is reported as it came out.
  • The same test, applied to the average of 6, failed: the 5.50 control came out as large as 6.00. Chapter VI rests on the ratio of mark sums of 18.00 and 17.95, with control sums of 21.00 and 19.50, which stay close to 1.
  • A county ranking failed the stability test fixed in advance (correlation between odd and even years 0.50, between early and late years 0.31), so the page does not rank counties. The candidate's county comes from the school code; the papers were marked in another county.
  • The cost estimate (chapter VII) counts candidates who sat the exam and failed with a single subject below 5, that one between 4.50 and 4.95 on a paper marked that year, and checks whether a 5.00 there would have given them an average of 6.

Sources

  • Ministry of Education, individual results of the 2024 national baccalaureate, published at bacalaureat.edu.ro. The public archive of both sessions was collected and made available by tacheometry.
  • Ministry of Education, individual results of the June 2024 session, published at data.gov.ro, with recognised grades flagged.
  • Ministerial order 6,156/2023 on the organisation of the 2024 national baccalaureate, for the rules on marking, on computing the grade and on resolving appeals.
  • Ministry of Education briefings on digitised marking, for how papers are scanned in the exam room and assigned automatically to teachers in another county.
  • Ministry of Education, individual results of the national baccalaureate, June session, 2015–2026, published at data.gov.ro.
  • Ministry of Education, press release no. 71 of 14 August 2023 on digitised marking: the 2022 and 2023 pilots in Călărași and Dâmbovița, nationwide from the August–September 2023 session.
  • Order 6,156/2023, articles 10, 12 and 13: the digitised marking platform, how the grade is computed, appeals.
  • Order 4,265/2015, on marking in another county, and the methodology approved by Order 4,799/2010, annex 2, article 56, on sealing the corner of the paper.
  • Thomas S. Dee, Will Dobbie, Brian A. Jacob and Jonah Rockoff, "The Causes and Consequences of Test Score Manipulation: Evidence from the New York Regents Examinations", American Economic Journal: Applied Economics, 2019, pp. 407–408.
Back to top ↑