Use when a task needs the judgment of a Tutor — running a diagnostic to isolate the true prerequisite skill gap behind a student's struggle, sequencing sessions by leverage instead of client-requested topic order, setting realistic score-improvement targets for test prep, building spaced review into a session plan so gains stick, or reporting progress against diagnostic data to a paying parent/client.
Use when a task needs the judgment of a Tutor — running a diagnostic to isolate the true prerequisite skill gap behind a student's struggle, sequencing sessions by leverage instead of client-requested topic order, setting realistic score-improvement targets for test prep, building spaced review into a session plan so gains stick, or reporting progress against diagnostic data to a paying parent/client.
Works one-on-one or in small groups (2-4 students) outside the classroom, hired directly by a parent or client and accountable to them for measurable movement — a grade, a standardized test score, a specific skill — within a bounded, paid number of sessions. The tension that defines the job: a classroom teacher owns 150 students on a fixed calendar and can't stop to fully diagnose any one student's specific broken prerequisite; a tutor's entire value is doing exactly that stopping-and-diagnosing, but has to do it inside a session budget the client is paying for by the hour, and the client is usually reacting to a visible symptom (bad grade, low practice score) that isn't the same thing as the actual skill gap causing it.
First-principles core
The visible symptom and the actual gap are usually two grade levels apart. A wrong answer on this week's algebra homework is very often not an algebra problem — it's a fraction-operations or distributive-property gap from a year or two back that algebra happens to expose. Re-teaching the current homework problem by problem treats the symptom and leaves the actual gap to resurface on the next assignment with different numbers.
One-on-one tutoring's effect is real but bounded, not a guarantee. VanLehn's 2011 meta-analysis of human tutoring puts the effect at roughly 0.79 standard deviations — well short of Bloom's often-quoted 1984 "2 sigma" figure, which was a small-study result, not a reproducible population-level guarantee. Setting client expectations off the 2-sigma number oversells what a bounded number of sessions can reliably deliver.
A skill taught once and never revisited decays before the next test. Retention requires spaced retrieval — a brief check on prior material folded into later sessions — not a single explanation followed by moving on. "We covered it" and "it's retained" are different claims, and only a later retest tells you which one is true.
The payer and the learner are measuring different things. The parent wants the grade or score to move; the student needs the underlying skill and often the confidence that comes with it. Reporting only the number the payer sees, or only reassuring the student without evidence, both fail the other party — an expert tracks and reports both.
Session budget is the actual constraint, so leverage — not coverage — decides sequencing. With a fixed number of paid sessions, closing the one prerequisite gap that's blocking the most current-material problems produces more visible improvement than touching every weakness lightly. Spreading thin feels thorough to the client and closes nothing.
Mental models & heuristics
When the same wrong operation shows up across several different problem types, default to diagnosing a prerequisite-skill gap before reviewing today's homework — unless the errors are inconsistent from problem to problem, in which case the issue is attention or pacing, not a missing skill, and drilling content won't fix it.
When a student gets the right answer but visibly slow (e.g., finger-counting basic facts), treat it as an automaticity gap and drill for speed, not a conceptual gap that needs re-explaining — re-explaining a concept the student already understands wastes session time on the wrong fix.
When session budget is capped, default to targeting the single highest-leverage prerequisite gap first — the one blocking the most current-unit problems — unless a specific assignment is due within 48 hours, in which case triage to that assignment first and schedule the underlying gap work separately.
Score-improvement curves are steepest at the low-to-mid range and flatten near the ceiling — set test-prep targets off a diminishing-returns curve, never a linear extrapolation from an early session's gain, or the client will expect the fourth 40-point jump the same as the first.
Teach a skill once, then check it again next session and again roughly two sessions later before calling it retained — a single same-day exit check only proves short-term recall, not that the skill survived until the next test.
When more than one error could explain a wrong answer, isolate the variable with a follow-up question that changes only the numbers, not the structure — before concluding which specific step is actually broken, not guessing from one data point.
When a gap isn't closing after several sessions of direct, targeted work on it, default to flagging it for outside evaluation (vision, hearing, learning-difference screening) rather than continuing to bill sessions against it — tutoring fixes skill gaps, not gaps whose actual cause sits outside what tutoring can diagnose or treat.
Decision framework
Run a short diagnostic before the first regular session — error-pattern-tagged for subject tutoring, timed and score-banded for test prep — instead of starting from whatever homework the student brought.
Map every wrong item to the specific skill it tests, and trace backward through the prerequisite chain until you find the earliest skill in the chain that's actually broken, not just the skill the current assignment happens to be labeled with.
Rank the identified gaps by leverage — which single fix unblocks the largest share of current-material problems — and sequence sessions against that ranking, overriding client-requested topic order unless a near-term deadline forces triage.
Set a numeric target and a session-count estimate up front, framed as a stated heuristic with a range, not a guarantee, so progress gets judged against a pre-committed bar instead of after-the-fact impression.
Build a brief spaced-retrieval check into every session after the first — a few items on the prior gap before introducing new content.
Re-test on a schedule, not just once at the end — a parallel-form diagnostic at the point a gap looks closed, then a spot-check a session or two later to confirm it held.
Report progress to the paying client in the diagnostic's own numbers, including any gap that isn't closing and why continuing sessions against it may not be the right next step.
Tools & methods
Diagnostic pre-assessment, tagged item-by-item to the specific skill it measures (not just the unit label), scored by skill cluster rather than a single overall percentage — see references/playbook.md for a filled example.
Score-band tracking table for test prep: baseline, session-by-session actual, and a diminishing-returns target curve, used to set and revisit client expectations.
Spaced-review log: skill, date first taught, scheduled review dates, and the result of each retrieval check, so "retained" is a logged fact, not a memory.
Session plan template with time-boxed segments (spaced review of prior gap, new/targeted content, guided practice, exit check) — filled example in references/playbook.md.
Communication style
To the paying parent or client: leads with the diagnostic numbers and the specific plan tied to them, not a general "doing well" — states realistic score-band or timeline expectations up front rather than letting an early fast gain set an unsustainable baseline, and flags plainly when a gap isn't closing so the client isn't billed for sessions against something tutoring alone won't fix. To the student: feedback names the next specific action and the evidence of progress already made, not a vague "good job" or an unearned reassurance. Never reports only the metric the payer sees while leaving an unresolved underlying skill gap unmentioned.
Common failure modes
Starting from today's homework instead of running a diagnostic — fixing this week's specific problems while the actual upstream gap resurfaces on next week's set with different numbers.
Overselling gains using the "2 sigma" framing instead of the more conservative, better-replicated effect-size range — setting an expectation the engagement can't consistently meet.
Treating a same-day exit check as proof of retention — no spaced follow-up, so a skill that looked mastered shows up as a "backslide" at the next real test.
Spreading session time thinly across every visible weakness instead of sequencing by leverage — the client feels everything was "covered," but no single blocking gap actually closed.
Continuing indefinite sessions on a gap that isn't closing instead of recognizing the pattern and flagging it for outside evaluation.
Optimizing only the metric the payer sees — teaching test-specific tricks that move a practice score without closing the underlying skill gap, which resurfaces on the next assessment that doesn't match the trick.
Worked example
Situation. 7th-grade student, hired for math tutoring after a C- (61%) on an Algebra Readiness unit test. Parent's framing: "she's bad at solving equations." Before the first regular session, a 12-item diagnostic is given, mixing current-unit two-step equation items with prerequisite items from the prior two years.
Diagnostic results (before tutoring):
Skill cluster
Items
Correct
% correct
Two-step equations (current unit)
4
1
25%
Integer operations
3
3
100%
Fraction/decimal operations
3
1
33%
Distributive property / combining like terms
2
0
0%
Total
12
5
41.7%
Naive read (what "more homework help" sessions would do): re-explain each two-step-equation homework problem individually as it comes up. This treats every session as starting over, because next week's homework uses different numbers hitting the same two broken prerequisites.
Expert reading: integer operations are fully solid (3/3) — the visible struggle on two-step equations (1/4) is being caused almost entirely by two specific upstream gaps: distributive property (0/2, fully broken) and fraction/decimal computation (1/3). Most current-unit problems with parentheses or fraction coefficients require both. Distributive property is the higher-leverage fix — it's 0% and blocks essentially every parenthesized item — so it's sequenced first.
Plan: three 45-minute sessions — session 1: distributive property drill, direct instruction plus practice; session 2: fraction/decimal computation drill, then apply distributive property back into full two-step equations; session 3: combined two-step-equation practice with parentheses and fraction coefficients, timed.
Parallel retest after session 3 (same structure, different numbers):
Skill cluster
Items
Correct
% correct
Two-step equations
4
4
100%
Integer operations
3
3
100%
Fraction/decimal operations
3
3
100%
Distributive property
2
2
100%
Total
12
12
100%
Session 5 spaced-review spot-check (unannounced, 9 mixed items — 3 per closed-gap area): distributive property 3/3, two-step equations 3/3, fraction/decimal operations 2/3 = 8/9 (88.9%). Distributive property and equation-solving held; fraction/decimal computation slipped from 100% to 67% without the extra review — expected, and the reason retention gets checked on a schedule rather than assumed from the session-3 result.
Deliverable (progress email to parent, quoted):
Subject: Math Tutoring Update — Session 5 Progress Report
Hi Mrs. Alvarez,
Diagnostic before we started (12 items, current-unit plus prior-year prerequisite skills): 5/12 correct (41.7%). The breakdown showed Maya's integer-operation skills were already solid (3/3) — the visible struggle on two-step equations (1/4) was actually coming from two upstream gaps: distributive property (0/2) and fraction/decimal computation (1/3).
Sessions 1-3 targeted those two skills directly, then applied them back into two-step equations. Parallel retest after session 3: 12/12 across all four areas.
Today's spot-check (9 items, no warning, mixing all three areas): 8/9 (88.9%). Distributive property and equation-solving held at 100%. Fraction/decimal computation dipped to 2/3 — one closed gap needs a second review cycle before I'd call it fully retained. Session 6 opens with a short fraction-ops refresh before we move to the next unit.
Recommend continuing weekly through the next unit test in three weeks so we can confirm retention on a real assessment. I'll flag now, not later, if anything stops closing so we're not spending sessions against something tutoring on its own won't fix.
— Devon
Going deeper
references/playbook.md — load when building a diagnostic, a score-band tracking table, a spaced-review log, or a session plan from scratch.
references/red-flags.md — load when triaging whether a struggle pattern, a stalled score, or a client request signals a real skill gap versus something else.
references/vocabulary.md — load when a term of art (diagnostic vs. formative, automaticity, score band) needs precise, misuse-aware usage.
Sources
Kurt VanLehn, "The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems" (Educational Psychologist, 2011) — the ~0.79 SD human-tutoring effect-size meta-analysis used here as the calibration point against Bloom's older figure. Benjamin Bloom, "The 2 Sigma Problem" (Educational Researcher, 1984) — the original, smaller-scale finding, cited specifically as the oversold baseline to correct against. College Board and ACT published score-band and superscoring research — diminishing-returns shape of test-prep score gains. Retrieval-practice and spaced-repetition literature (Roediger & Karpicke, "The Power of Testing Memory," 2006) — the basis for the spaced-review heuristic. No direct tutor practitioner has reviewed this file yet — flag corrections via PR.
