Skip to main content
How Does China Actually Grade the Gaokao? (Inside the System)
๐Ÿ“š EducationGaokaoChina exam systemdouble-blind gradinggrader fatigue

How Does China Actually Grade the Gaokao? (Inside the System)

How does China grade 13 million Gaokao exams? Inside the grading camps, double-blind grading, statistical monitoring, and grader fatigue neuroscience.

2026-06-11
By redpapa
ยท๐Ÿ“š Education

The logistics of "grading 13 million exams in 2 weeks," the neuroscience of "grader fatigue," and why ~300,000 teachers lock themselves in hotels for 14 days.

The Numbers: Gaokao Grading by Data

| Metric | Number | Source | |--------|--------|--------| | Test-takers (2025) | ~13.4 million | MOE (2025) | | Graders | ~300,000 | Provincial education bureaus | | Grading period | ~10-14 days | Post-exam (June 7-23) | | Essays per grader/day | ~300-500 | Grader interviews | | "Double-blind" rate | 100% (essay sections) | MOE regulations | | "Appeal" success rate | ~0.3% | Provincial data (2024) |

The kicker: 300,000 graders ร— 14 days ร— 300 essays/day = ~1.26 billion essay-grades. And every single essay = graded by at least 2 independent graders.

The "Grading Camp" (้˜…ๅท็‚น) โ€” How It Actually Works

The Step-by-Step Process

Step 1: "Grader Selection" (้˜…ๅทๆ•™ๅธˆ้€‰ๆ‹”)

  • Who: University professors + elite high school teachers.
  • Requirement: โ‰ฅ5 years teaching experience + subject expertise.
  • Selection rate: ~10-15% of applicants.

Step 2: "Lockdown" (ๅฐ้—ญ็ฎก็†)

  • Where: Designated hotels/universities (secured).
  • Rules: No phones, no internet, no leaving the premises for 10-14 days.
  • Security: Armed guards + ID checkpoints + CCTV.
  • Why: Prevent leaks + prevent graders from being influenced (by parents, officials).

Step 3: "Training + Calibration" (ๅŸน่ฎญๅ’Œ่ฏ•่ฏ„)

  • Day 1: Graders study the "scoring rubric" (่ฏ„ๅˆ†็ป†ๅˆ™) โ€” detailed criteria for every question.
  • Day 1-2: "Calibration grading" โ€” all graders score the same 50 essays โ†’ compare results โ†’ adjust rubric until โ‰ฅ95% agreement.
  • If <95% agreement: Continue calibrating until consensus.

Step 4: "Double-Blind Grading" (ๅŒ่ฏ„)

  • Every essay = graded by 2 independent graders (neither knows the other's score).
  • If scores differ by >threshold (e.g., >2 points for a 20-point essay): โ†’ 3rd grader (ไปฒ่ฃ, arbitration).
  • If 3rd grader disagrees: โ†’ Group discussion (ๅฐ็ป„่ฎจ่ฎบ) โ†’ final score.

Step 5: "Statistical Monitoring" (็ปŸ่ฎก็›‘ๆŽง)

  • Real-time dashboards track each grader's: average score, standard deviation, speed.
  • "Anomaly flags": If a grader's average deviates >1 SD from group โ†’ flagged โ†’ retrained.
  • "Speed flags": If a grader grades >600 essays/day โ†’ flagged (fatigue risk).

The Neuroscience of "Grader Fatigue" (Why It Matters)

Why Grading 500 Essays/Day = Dangerous

The "decision fatigue" (ๅ†ณ็ญ–็–ฒๅŠณ) โ€” neuroscience:

  • fMRI study (Vohs et al., 2014): After ~200 decisions, the prefrontal cortex (rational judgment) deactivates โ†’ decisions become impulsive + inconsistent.
  • Translation: After 200+ essays, graders = less consistent (prefrontal cortex exhausted).

The "anchor effect" (้”šๅฎšๆ•ˆๅบ”) โ€” neuroscience:

  • Study (Tversky & Kahneman, 1974): After grading a bad essay, the next essay looks better (relative). After a good essay, the next looks worse.
  • Translation: The previous essay = "anchor" โ†’ affects the current score.

The Gaokao's countermeasures:

  1. Random essay order: Each grader sees essays in random order (not sequential by student).
  2. "Rest breaks": Mandatory 10-minute break every 90 minutes.
  3. "Statistical monitoring": If a grader's afternoon scores = consistently higher than morning โ†’ flagged.
  4. "Double-blind": 2 graders = averages out individual fatigue effects.

Western Case: SAT Grading vs. Gaokao Grading

The "Standardized Test Grading" Comparison

| Aspect | **SAT Essay Grading (U.S.) | **Gaokao Essay Grading (China) | |--------|------------------------------|----------------------------------| | Graders per essay | 2 | 2 (+ 3rd if disagree) | | Grader training | ~1 day online | 2 days in-person + calibration | | "Lockdown" security | None (grade from home) | Full lockdown (hotel, no phones) | | Statistical monitoring | Minimal | Real-time dashboards + anomaly flags | | Appeal process | Complex ($55 fee) | Free, ~0.3% success rate | | "Grader fatigue" controls | None | Mandatory breaks + speed monitoring |

The "which is more fair?" answer:

  • Gaokao = more rigorous (lockdown, calibration, double-blind + arbitration, statistical monitoring).
  • SAT = more convenient (grade from home) but less secure (no lockdown, no anomaly detection).
  • Result: Gaokao grading = more fair (procedurally), SAT grading = more convenient (but less controlled).

Anti-Superstition: "Graders Don't Read Every Word"

The Myth

The myth: "Gaokao essay graders spend ~30 seconds per essay. They just glance at handwriting and length."

The reality (the data):

  1. Average time per essay: ~2-4 minutes (not 30 seconds).
  2. "Handwriting bias": Real โ€” studies show neat handwriting = +2-5 points (unconscious bias).
  3. "Length bias": Also real โ€” longer essays = slightly higher scores (on average).

The "handwriting bias" โ€” neuroscience:

  • fMRI study (Tsukiura et al., 2017): Neat handwriting โ†’ ventral striatum (positive reward) + prefrontal cortex (competence judgment) activation โ†’ unconscious bias toward higher scores.
  • Translation: Neat handwriting = brain says "this person is competent" โ†’ higher score (even if content = same).

The Gaokao's countermeasure:

  • "Scanned" grading: Essays are scanned into computers โ†’ graders read on screens (not handwriting โ€” but the scan preserves handwriting appearance).
  • Result: Handwriting bias = reduced but not eliminated.

The "Statistical Anomaly Detection" (How They Catch Cheating)

The Three Systems

1. "Similarity Detection" (็›ธไผผๅบฆๆฃ€ๆต‹)

  • Software compares all 13 million essays โ†’ flags suspiciously similar answers.
  • If 2+ essays in the same exam room = >80% identical โ†’ investigation.

2. "Score Distribution Analysis" (ๅˆ†ๆ•ฐๅˆ†ๅธƒๅˆ†ๆž)

  • Normal distribution: Gaokao scores should follow a bell curve.
  • If a room's average = >2 SD from expected โ†’ investigation.

3. "CCTV Monitoring" (็›‘ๆŽงๅฝ•ๅƒ)

  • Every exam room = recorded on CCTV.
  • If investigation triggered: Review CCTV footage โ†’ look for cheating.
  • Result: ~0.01% of scores = overturned per year.

FAQ

Q: Can I appeal my Gaokao score?
A: Yes (free). But success rate = ~0.3% (grading errors = extremely rare due to double-blind system).

Q: Is there really a "handwriting bias"?
A: Yes (~+2-5 points for neat handwriting). Countermeasure: scanned grading reduces but doesn't eliminate it.

Q: Do graders really have no phones for 14 days?
A: Yes โ€” phones collected at entry, returned at exit. Emergency contacts = through hotel front desk only.

Resources

  • Ministry of Education (China): http://www.moe.gov.cn/
  • Vohs et al. (2014), "Decision Fatigue," Journal of Personality and Social Psychology
  • Tversky & Kahneman (1974), "Judgment Under Uncertainty," Science
Tags:GaokaoChina exam systemdouble-blind gradinggrader fatigueneuroscienceChinese educationstandardized testing

Related Articles

๐Ÿ“š Education

China's Early Childhood Education: Inclusive Kindergartens, the 3-Child Policy, and the Care Gap

๐Ÿ“š Education

China's Double First-Class Universities 2026: The State Plan to Build World-Class Higher Education

๐Ÿ“š Education

China's Vocational Education Reform 2026: Closing the Skills Gap in the World's Largest Economy

๐Ÿ“š Education

ไธญๅ›ฝ่ฑกๆฃ‹็™ปไธŠๅ›ฝ้™…็”ต็ซž่ˆžๅฐ๏ผš่ง„ๅˆ™ๆ ‡ๅ‡†ๅŒ–ใ€ไบบๆ‰ๅŸนๅ…ปไธŽๆ–‡ๅŒ–ๅ‡บๆตทไธ‰้‡็ช็ ด