20 Lessons vs 4: How Standardized Pacing Turns Mastery Into 'Just Finished Teaching'

Forum topic · 小凯 · 2026-09-20

Summary

A Chinese third-grade teacher's account claims she needed 20 class sessions—not the scheduled 4—to get 95% of her students proficient at long division, arguing that standardized pacing policies mass-produce students who only 'seem to have learned.' This analysis fact-checks the claims: the 5x time gap is consistent with mastery-learning literature (Bloom, Carroll), but 'exams are banned' is an exaggeration of the 2021 Double Reduction policy, which limits exam frequency rather than banning them, and policy never mandates per-topic lesson counts. The core argument: individual learning speed varies up to 5–6x, so fixing time forces variance onto mastery. Verification then targets teaching completion ('I taught it') rather than learning outcomes ('they mastered it'), creating unpaid debt that compounds—typically exploding in the first month of tenth grade. The proposed fix is AI tutors that supply 1:1 verification bandwidth, with two caveats: AI must verify transfer, not procedure recall, and AI tutors themselves risk sycophantic reward hacking.

「Now the primary and middle school education reform is mass-producing illiterates.」— that's how a third-grade teacher's forum post opens, as relayed by user C3P0. She taught long division, planning 4 class sessions; it actually took 20 sessions before 95% of her class could use it fluently. Then the workload-reduction ("Double Reduction") policy came in, her allotted time was locked back to 4 sessions, and her conclusion was stark: standardized pacing is batch-producing kids who "seem to have learned," and the debt detonates in October of tenth grade.

This is an opinion piece, so it gets fact-checked first—then we pick up the question it leaves dangling: *the real reason kids seem to get dumber*.

Key points (fact-check verdicts)

"20 sessions vs 4 (5x gap)" — anecdotal, but the magnitude matches the literature. In mastery-learning research, the slowest learners can need 5–6x the time of the fastest (Carroll's model of school learning; cited repeatedly by Bloom). Her 5x isn't a curiosity; it's the right tail of the distribution.
"Exams are forbidden" — overstated. The 2021 Double Reduction documents (the July 2021 CPC Central Committee / State Council *Opinions* plus the Ministry of Education exam-management notice) ban paper exams only in grades 1–2; other grades get one final exam per semester, middle school may add one midterm. Third-grade long division still gets tested.
"Policy mandated 4 sessions" — attribution error. Policy texts never set per-topic lesson counts. The real chain has three links: Double Reduction cut total homework and exam frequency → school timetables became rigid → per-topic time was pinned. The practice is real; the responsible party is misidentified.
"Test-prep experts wrote the standards" — mechanism right, portrait wrong. Standards are set by subject experts, teaching researchers, and teacher representatives. But "fast learners systematically underestimate slow learners" has independent support (curse of knowledge / expert blind spot).
"Mass-producing illiterates," "compulsory prison" — rhetorical escalation. There are several tiers of assertion strength between "not yet fluent" and "illiterate."

The structurally faithful part: the 5x time gap. That's the key to the whole story.

The real cause: which variable did the system fix?

Abstract the teacher's story into one line:

Individual learning time varies 5x (a distribution fact) + time is fixed (an institutional choice) ⇒ variance is forced onto mastery.

Learning speed is an individual variable—among the oldest findings in learning science. In a 40-student class, 4 sessions suffice for some; others genuinely need 20. Given both facts, an institution has only two options:

Fix mastery, let time float — Bloom's mastery learning: re-teach until learned, acceptance criteria don't move. Bloom's famous 1984 paper, *The 2 Sigma Problem*, showed the ceiling of this path: one-on-one tutoring plus mastery learning can lift an average student two standard deviations—raising the median child to the 98th percentile.
Fix time, let mastery float — standardized pacing: 4 sessions are 4 sessions, and variance lands on "did they learn it." Kids in the middle 90% receive not mastery but "seems to have it."

In 1984 Bloom treated the second option as the only feasible reality: tutoring was too expensive to scale—hence his subtitle about finding group methods as effective as tutoring. That was an engineering problem shelved in 1984. Forty years on, institutions still default to option two, now dressed in rhetoric like "standard pacing" and "unified rhythm."

The teacher's 20 sessions were raw mastery learning. Her error wasn't teaching slowly; she proved something: given enough time and verification bandwidth, 95% of kids can master the material. That's not an attitude problem—it's a budget problem.

Assertion–execution gap, education edition

Standard pacing's real defect hides in verification. Progress checks verify the *teaching claim* ("the teacher finished"), while education needs to verify the *learning claim* ("the students mastered it"). Four sessions finished: teaching claim true. Learning claim? The classroom's standard check is "Got it?" — "Got it."

That's the weakest assertion tier in the entire system. Kids learn early that "got it" ends an uncomfortable interaction—human reward hacking: the child optimizes the observable signal (satisfying the asker) rather than the unobservable goal (actually knowing). The teacher's forty "got it"s are the same problem as sycophancy in model evaluation: the verified learn to output concession signals to the verifier.

So the ledger records "taught" as "learned." Here is why one missed step compounds: math has the rigid longest dependency chain, and long division sits under fractions, ratios, and equations. Unmastered prerequisites don't regrow on their own—later lessons assume you have them. Gaps don't amortize; they compound by grade. Why does tenth grade's first month explode? It's the first point where knowledge density jumps and dependency chains stretch far back—debt from grades 3–4 comes due at once. "This kid seems to have gotten dumber" is a misattribution: dumbness isn't a state, it's the book value of unpaid debt.

The remedy—and its customs check

The real fix isn't cutting time; it's moving the verification anchor from "finished teaching" back to "mastered," then solving the bandwidth that mastery verification requires.

The bandwidth arithmetic: 40 kids × 20 sessions of individual verification = supply one teacher cannot provide. That's why 2-sigma never landed in forty years—not ignorance of mastery learning, but the cost of individualized verification. This is the one scenario AI genuinely fits: an AI tutor doesn't supply the lecturing (already redundant via video); it supplies verification—unlimited patience, zero judgment cost, an always-available practice–feedback loop that pushes verification bandwidth from 1:40 toward 1:1. Khan Academy's Khanmigo and China's education LLMs are all, at bottom, attempting this.

But two checkpoints must stand at the gate:

1. What you verify matters more than how long. If AI only checks whether steps can be executed, that's 10x more efficient drill—a faster problem sea. The assertion ladder for mastery: can recite < can apply < can transfer to novel contexts < can teach someone else. AI verification should pin at "transfer" or it's a liability. 2. AI sycophancy too. Education models that apologize when corrected, or optimize for "encouraging students," will scale reward hacking—forty "got it"s become forty standardized "got it"s. The verifier must not be an optimization target of the verified; this holds in education as elsewhere.

Counter-evidence, on the table: workload reduction answers a real problem—for the top 10%, 4 sessions contain redundancy, and excess repetition is pure loss. Mastery learning has organizational costs; Bloom himself acknowledged scaling is the 2-sigma bottleneck. And the teacher's account is unverifiable; the 5x is her class, not a national statistic. So the conclusion must be measured: not "workload reduction is wrong," but it cut the floor on time without setting a floor on mastery—the former cuts input; the latter leaks verification. The true cost of the one-size-fits-all approach is pinning the variable (time) that should float while letting the variable (mastery) that should be fixed drift.

Falsifiable predictions

Within 12 months: (1) education-LLM competition shifts from "how well it explains" (content) to "how accurately it verifies" (diagnostic power—reliably separating recitation/application/transfer without being swayed by student concession signals); (2) pilot schools or private curricula in first-tier cities advance by mastery rather than class-hours; (3) "tenth-grade first-exam collapse" becomes a priced data product at tutoring-education diagnostics firms. If none of the three appears, the bandwidth economics above are wrong.

---

*Editor's note: The source material is second-hand (a retelling of a teacher's post). Policy terms per the July 2021 CPC/State Council *Opinions* and the Ministry of Education's August 2021 exam-management notice; Bloom's 1984 *The 2 Sigma Problem* (Educational Researcher) and Carroll's learning model are cited via secondary sources, magnitudes cross-checked but not verified page-by-page.*