White Paper · Lacefield Pedagogical Framework · v5.0
Every other component of the framework depends on knowing where a student actually is — not where a placement test says they are, not where they think they are, but where their knowledge genuinely begins and where its structure is sound versus corrupted. This paper documents the intake diagnostic that produces that picture.
Standard placement assessment tells you what level a student tests at. It does not tell you where their knowledge structure is sound, where it is partial, where it is wrong, or what foundational operations they can perform automatically versus effortfully. These distinctions are not refinements of placement data — they are categorically different information, and the absence of them means that instruction begins from a guess rather than a diagnosis. This paper documents a three-component intake diagnostic — foundational fluency assessment, schema adequacy mapping, and subject-specific reading comprehension baseline — that produces the inputs the Lacefield Pedagogical Framework requires to initialize correctly. The output is a Dynamic Learning Profile: a structured representation of the student's current state that directly sets the parameters for session calibration, assignment generation, and difficulty adjustment. The diagnostic protocol described here is largely novel; where established research applies it is cited, and where the territory is genuinely new that is stated explicitly. The protocol is a design proposal intended to be refined through implementation and empirical observation.
A placement test tells you approximately which level of content a student can engage with. That is useful information but it is not sufficient to initialize adaptive instruction. Three things a placement test does not reveal matter more for instruction than the level score itself.
It does not distinguish gaps from wrong schema. A student who scores at a fifth-grade mathematics level may have fifth-grade gaps — things they simply have not learned yet — or may have learned earlier material incorrectly, producing a schema that actively interferes with correct understanding at higher levels. These require completely different instructional responses. Gaps are filled by building forward. Wrong schema requires the more demanding process of conceptual change: building adjacent correct structure, demonstrating the contradiction with the existing incorrect structure, and then replacing it — a process documented by Strike and Posner (1992) as requiring the student to first become dissatisfied with what they currently believe. A placement score cannot distinguish between these two conditions. Treating a wrong-schema student as a gap student — simply teaching the material they appear to be missing — will produce the synthetic model problem documented by Vosniadou (1994): additional instruction that gets incorporated into the incorrect framework rather than replacing it.
It does not measure automaticity. A student who can eventually produce the correct answer to a basic arithmetic problem is not the same as a student who retrieves it automatically. The difference is working memory cost. The companion paper on foundational fluency (Lacefield, 2026c) documents the mechanism: non-automated operations consume working memory that would otherwise be available for the higher-level reasoning the current task requires. A placement test that gives full credit for a slow correct answer misses the constraint entirely. A student who scores at the seventh-grade level but computes basic arithmetic effortfully will perform below their placement level on any task that requires both arithmetic and higher-level reasoning simultaneously — which is most tasks.
It does not measure reading comprehension relative to the subject's language. Applied mathematics, science, economics, and most other academic subjects present their problems in language. The reading comprehension required to extract a mathematical structure from a word problem is not the same as general reading comprehension, but it is closely related, and it sets a ceiling on applied performance that is independent of computational or reasoning ability. The companion paper on reading as the hidden core of mathematics (Lacefield, 2026d) documents this at N = 111,346. A placement test that measures mathematical computation without controlling for reading comprehension conflates two distinct constraints and cannot tell you which one is the binding bottleneck for this student.
These three gaps — between placement and calibration, between accuracy and automaticity, between content coverage and schema health — collectively explain why most educational software produces what practitioners recognize as the death spiral: a student placed slightly too high, given tasks that are ten percent too difficult for three weeks, who quietly disengages before any measurement system registers the problem. The dominant approach to student modeling in educational software is Knowledge Tracing — originally Bayesian Knowledge Tracing (Corbett & Anderson, 1994), extended through Deep Knowledge Tracing and its variants. These systems model the probability that a student has acquired a skill, updated after each problem attempt. They are well-validated for predicting whether performance on a specific skill will be correct on the next attempt. What they do not model is the structure of the knowledge — whether the student's understanding is built on correct foundations or on a corrupted schema that produces correct-looking answers in familiar contexts and collapses on novel ones. A student who has memorized a procedure that produces correct answers on standard problem formats will show high knowledge tracing scores while having no genuine schema for the underlying concept. The diagnostic described in this paper is doing something structurally different: not tracking performance probability, but mapping the health and architecture of the mental model itself. Schema tracing. The distinction matters most in the cases that matter most — the student who appears to be performing adequately but whose understanding will not transfer, and the student who appears to be struggling but whose schema is sound and whose errors are execution rather than conception.
"The diagnostic is not a better placement test. It is a different kind of instrument entirely — one that maps the structure of what a student knows rather than the level at which they can perform."
Before the diagnostic can run, the subject must be mapped. This mapping is not a curriculum outline or a list of topics. It is a dependency structure — a representation of which concepts are genuinely prerequisite to which others, organized into tiers by their logical sequence in the subject's development. Only load-bearing concepts belong on the map. A concept is load-bearing if its absence or corruption produces compounding errors in everything built above it. Peripheral concepts — those that can be missing without systematically affecting the rest of the structure — are not included.
The distinction between load-bearing and peripheral is critical and non-obvious. Most curriculum documents list all concepts as if they were equally foundational. The subject tier map deliberately excludes concepts that are merely useful, including only those whose absence or corruption is structurally consequential. Building this map correctly is itself a significant piece of subject-matter expertise — specifically what Ball, Thames, and Phelps (2008) term horizon content knowledge: the awareness of how mathematical topics relate across the span of the curriculum, and which relationships are structurally necessary versus incidental. A teacher or administrator who understands the subject only at the level they are currently teaching cannot build a valid tier map. They can identify what topics appear in the curriculum — any curriculum document does that. What they cannot reliably identify is which of those topics are load-bearing and which are peripheral, because that distinction requires understanding what happens to the structure above when a concept is missing or wrong, which requires knowing the structure above. The depth requirement for building a valid tier map is substantially greater than the depth required to teach the content at that level. This is the same depth requirement documented in the companion paper on incorrect correction (Lacefield, 2026h) — and it applies to the diagnostic instrument itself before it applies to any student.
Chi, Feltovich, and Glaser (1981), in their foundational study of expert-novice differences in physics problem-solving, established that experts organize knowledge around deep structural principles — the load-bearing relationships — while novices organize around surface features of problems. The subject tier map is a formalization of the expert's organizational structure, made explicit so that the diagnostic can assess which level of organization a student has internalized.
Foundational fluency, as defined in this framework, is not the ability to perform foundational operations correctly. It is the ability to perform them automatically — accurately, quickly, and without conscious deliberation — so that they impose minimal working memory cost during higher-level tasks. The foundational fluency assessment measures this specifically, not correctness alone.
Every foundational operation exists at two levels simultaneously, and a student can have one without the other. This distinction determines both what the assessment must measure and what the instructional response to each failure pattern looks like.
Implicit fluency is the ability to use an operation correctly in context without being able to explicitly name, define, or articulate it. Most language users have implicit fluency in grammatical agreement — they know when a sentence sounds wrong without being able to state the rule. Most arithmetic-competent students have implicit fluency in order of operations — they apply it correctly in familiar contexts without necessarily being able to state why. Implicit fluency is sufficient for many purposes and is often present when explicit fluency is absent.
Explicit fluency is the ability to deliberately retrieve, name, and apply the operation — to do it on demand, in unfamiliar contexts, and under conditions that require conscious selection of the right operation among alternatives. Explicit fluency is what allows a student to recognize that a problem requires a specific operation rather than pattern-matching to a familiar problem type. Without explicit fluency, performance collapses on novel problems even when implicit fluency is intact.
The assessment must test both layers for the same foundational operations, because the instructional response differs. A student with strong implicit but weak explicit fluency needs explicit instruction on naming and articulating what they can already do — a relatively tractable intervention. A student with weak implicit fluency needs the operation built from the ground up, which takes substantially longer and requires different instructional sequencing.
Accuracy alone is insufficient to assess foundational fluency. A student who produces a correct answer after four seconds of visible deliberation has not demonstrated automaticity — they have demonstrated that the operation is within their capability but not within their automatic retrieval range. The working memory cost of the four-second deliberation is real and consequential regardless of the correct outcome.
Response speed therefore must be tracked alongside accuracy throughout the fluency assessment. In a human-administered diagnostic, visible deliberation is directly observable — the pause, the scratch work, the uncertain expression before committing to an answer. In an automated implementation, response latency serves as the primary proxy: the time between a problem being presented and the student's first committed response. This is not a perfect substitute for direct observation, but it captures the relevant signal adequately for calibration purposes. Approximately two seconds is the working threshold — responses faster than this are treated as automatic; responses slower, regardless of accuracy, are treated as effortful. This threshold is a starting heuristic subject to individual calibration, not a universal standard.
It is worth being direct about the comparison this implies. Most tutors do not track visible deliberation at all. The best human tutors do it intuitively as part of reading the student in real time. The app does it systematically through latency logging on every item. The relevant comparison is not between an automated system and a perfect human tutor — it is between an automated system and the actual distribution of instruction students receive, which almost never includes any deliberation monitoring whatsoever. Systematic latency tracking is not an approximation of something better. For most students in most instructional contexts, it is substantially more than they have ever had.
The limitation that a live tutor can read a student's face for frustration, confusion, or incomprehension — and that a current app cannot — is real and should be stated honestly. Real-time affective state detection from facial expression exists as a research capability: systems built on Ekman's action unit coding and commercial implementations such as Affectiva already detect confusion, frustration, and engagement states from facial video in controlled conditions. The obstacles to deployment are not primarily technical — they are regulatory (COPPA makes facial data collection from minors extremely sensitive), consent-related, and ethical. A reasonable estimate is that within three to five years this will be deployable in adult-facing applications with appropriate consent frameworks. The system architecture should therefore be designed so that richer affective signals can be integrated as they become available and ethically deployable, without requiring restructuring of the core diagnostic and calibration logic. In the current implementation, latency and distractor patterns are the available proxies. They are meaningful proxies. They are not the ceiling.
For the schema adequacy diagnostic to function as an algorithm rather than a human conversation, it requires a mechanism for distinguishing wrong-schema errors from careless errors and from random errors. The mechanism is distractor analysis.
In the multiple-choice components of the diagnostic, wrong answer options are not arbitrary distractors. They are the specific conclusions a student would reach by applying common misconceptions to the problem. If a student incorrectly adds the numerators and denominators of two fractions separately — a predictable and well-documented misconception — they will produce 2/6 from 1/2 + 1/4 rather than the correct 3/4. This answer is not random. It is the specific output of a specific wrong schema. When a student selects it, the diagnostic has information: not just that the answer is wrong, but which misconception produced it.
This means the subject tier map must specify not only the correct understanding at each tier but also the predictable misconceptions — the specific wrong schemas that students commonly hold at each level. These misconceptions become the distractor set. A wrong answer that matches a predictable misconception is logged as a wrong-schema signal. A wrong answer that matches no predictable misconception is logged as a possible careless error or random error, and triggers a repeat item before a conclusion is drawn.
For open-response items — where the student produces a free answer rather than selecting from options — the same logic applies in reverse. The system checks whether the wrong answer produced corresponds to any known misconception for that item. If it does, it is treated as a wrong-schema signal. If it does not match any known misconception pattern, it is treated as a non-schema error and handled accordingly.
Redundancy is built into the diagnostic design to prevent noise from being mistaken for signal. No concept is assessed only once. Each concept appears in multiple items — at minimum one item testing the concept in isolation and one testing its relationship to adjacent concepts — and the wrong-schema flag is not set on a single wrong answer. The flag requires a consistent pattern across multiple items involving the same concept. A single wrong answer might be a careless error, an outlier, or an interface problem. A consistent pattern of the same misconception-driven wrong answer across multiple items is a schema signal.
The foundational fluency assessment uses two formats in sequence for each foundational operation, corresponding to the implicit/explicit fluency distinction.
Recognition format — implicit fluency. The student encounters the operation embedded in a context and must respond correctly without being asked to identify or name it. In a multiple-choice implementation, wrong answer options are misconception-driven distractors constructed from the specific errors that incorrect implicit fluency produces. A student who selects a misconception-driven distractor on a recognition item is showing implicit fluency failure with wrong-schema characteristics — they are not just guessing, they are applying an incorrect operation consistently. A student who selects a random wrong answer is showing a different failure pattern. The distinction is logged.
Recall format — explicit fluency. The student is asked to produce, explain, or apply the operation explicitly, out of context. This tests whether the student can deliberately retrieve and apply the operation, not just recognize its correct use. Open-response items in this format check produced answers against known misconception patterns as described in section 3.3.
Both formats track response latency alongside accuracy. Each foundational operation is assessed in at least two items per format — once in isolation and once in relationship to adjacent operations — so that no conclusion rests on a single data point. The assessment stops at the point where a consistent pattern of slow or inaccurate responses on both layers identifies a foundational bottleneck, which is then logged as a primary instructional target.
The schema adequacy diagnostic is the most novel component of the intake protocol and the one least supported by existing assessment literature. It is documented here as a design proposal derived from the framework's theoretical commitments, intended to be refined through implementation.
The schema adequacy diagnostic as described here does not have a direct precedent in published assessment literature. The theoretical basis — Chi et al. (1981) on expert/novice knowledge organization, Vosniadou (1994) on synthetic models, Strike and Posner (1992) on conceptual change — supports the diagnostic's design rationale. The specific protocol is original to this framework. Its validity and reliability as an assessment instrument have not been empirically established and constitute a primary target for future research.
The schema adequacy diagnostic is a structured conversation, not a test. It moves upward through the subject tier map, tier by tier, assessing the student's ability to describe elements at each tier and to describe the relationships between tiers. The framing throughout is explicitly low-stakes: the student is told at the outset that this is not a test of what they know, but a map of where to start — that most of what follows is things they are going to learn, not things they are expected to already understand. This framing is not courtesy. It is a diagnostic necessity: a student who is trying to perform rather than accurately report will produce distorted signal, and the framing is designed to remove the incentive to perform.
At each tier, two questions are asked. First: can the student describe the elements at this tier — name them, gesture at them, characterize them in any way? Precision is not required. Approximate, directional, or partially correct descriptions carry diagnostic information. Second: can the student describe the relationships — why these concepts matter, what they prepare for, how they connect to each other or to the tier below? Again, precision is not the standard. The diagnostic is looking for whether any structural understanding is present, not whether it is complete.
The conversation at each tier is calibrated to the student's apparent level. If a student demonstrates solid understanding at Tier 1, the diagnostic moves quickly to Tier 2. If a student demonstrates partial understanding at Tier 2 — some elements present, some absent, relationships partially but not fully understood — the diagnostic continues to Tier 3 to check for the wrong-schema condition before stopping.
The primary stopping rule is: when a tier is found to have partial, incorrect, or absent schema, that tier is the diagnostic floor. The student's effective starting point is at or below that tier, regardless of what they claim to know above it.
But the diagnostic does not stop immediately at the floor. It continues one tier higher specifically to check for the wrong-schema condition — the case where a student has a label or apparent familiarity with a higher-tier concept whose foundation at the current tier is missing or incorrect. A student in this condition appears to have higher-level knowledge that they do not actually have in the structurally sound sense. Their Tier 3 language is built on a corrupted Tier 2 foundation, which means the Tier 3 understanding is also corrupted even if it sounds plausible.
This is the most dangerous diagnostic finding and requires explicit flagging in the Dynamic Learning Profile. The student who has a wrong schema at Tier 2 and apparent knowledge at Tier 3 is not a student who needs Tier 2 instruction followed by Tier 3 instruction. They are a student who needs Tier 2 conceptual change work — the more demanding process of replacing incorrect structure rather than filling a gap — before any higher-tier instruction can be reliably built. Treating them as a gap student will produce the synthetic model problem: new instruction that gets incorporated into the wrong Tier 2 structure and produces confident incorrect Tier 3 understanding.
For subjects with distinct major branches, the schema diagnostic maps each branch separately. A student may have solid Tier 2 schema in algebra and no Tier 2 schema in geometry. A student may understand macroeconomic concepts and be unfamiliar with microeconomic ones. These are not equivalent to a single partial Tier 2 finding — they represent different instructional starting points for different parts of the subject, and the Dynamic Learning Profile must represent them separately rather than averaging them into a single level estimate.
The reading comprehension baseline is the most thoroughly research-supported component of the diagnostic and is documented in detail in the companion paper on reading as the hidden core of mathematics (Lacefield, 2026d). Its inclusion in the intake diagnostic reflects the finding that language comprehension is a unique predictor of applied performance in any subject where problems are presented in language — which includes mathematics, science, economics, history, and most other academic domains.
The baseline is subject-specific rather than generic. General reading comprehension measures — passage comprehension tests, reading fluency assessments — do not capture the precision with which a student can extract mathematical, scientific, or domain-specific logical structure from a problem statement. At foundational and secondary levels, the primary reading bottleneck is general comprehension: can the student extract what is being asked from a prose description? At advanced levels, the bottleneck shifts to lexical ambiguity: the interference caused by words that carry both everyday meanings and precise domain-specific meanings. In mathematics, "volume," "product," "mean," "prime," "rational," and "domain" all have everyday meanings that are more practiced and more automatically activated than their mathematical counterparts. A student who allows the everyday meaning to shape problem interpretation before the mathematical definition can override it produces comprehension errors that are definitional rather than contextual — and these require different remediation than general comprehension errors. The reading baseline therefore includes two distinct components: subject-specific prose comprehension items for foundational and secondary levels, and lexical ambiguity probes for advanced levels — items that test whether the student applies the domain-specific or everyday meaning of high-ambiguity terms when both meanings are contextually plausible.
The output of the reading baseline is a language ceiling estimate: the point at which imprecise reading is likely to become the binding constraint on applied performance, independent of the student's reasoning or computational ability. This ceiling is entered directly into the Dynamic Learning Profile and informs both the assignment design and the session framing — a student at or near their language ceiling on a given problem type needs reading-precision intervention before mathematical instruction on that problem type will be reliably effective.
The three diagnostic components combine into a single structured output: the Dynamic Learning Profile. The profile is not a score. It is a map of the student's current state, expressed in the terms the framework's calibration system directly operates on. Every field in the profile is an input to a specific instructional decision.
The profile's calibration start field directly sets the parameters for the first session's gradient assignment, as documented in the companion paper on the gradient lesson system (Lacefield, 2026g). The session update log ensures that the profile becomes more precise over time — each session is both instruction and diagnostic, and the information produced by observing a student work is fed back into the profile before the next session is designed. The longer a student is in the system, the more accurately the profile represents their actual state.
The schema adequacy diagnostic has not been empirically validated as an assessment instrument. Its design is theoretically grounded in Chi et al. (1981), Vosniadou (1994), and Strike and Posner (1992), and it is consistent with seven years of practitioner observation in which diagnostic conversations of this type informed effective instructional sequencing. But its reliability — the degree to which two administrators would produce the same profile from the same student — and its validity — the degree to which the profile accurately predicts instructional starting point and rate of progress — have not been formally tested. These are primary targets for empirical research as the framework moves into systematic implementation.
The fluency latency thresholds are working heuristics, not empirically derived cutoffs. The approximately two-second threshold distinguishing automatic from effortful retrieval is a reasonable starting estimate based on the processing speed and working memory literature, but the precise threshold may vary by subject domain, operation type, student age, and individual processing speed baseline. The threshold should be treated as a starting calibration point subject to individual adjustment, not as a universal standard.
The subject tier map must be built correctly for the diagnostic to be valid — and building it correctly requires substantial subject-matter depth. The quality of the diagnostic output is bounded by the quality of the tier map it is administered against. A tier map that includes non-load-bearing concepts as if they were load-bearing will produce a diagnostic that identifies false bottlenecks. A tier map that omits genuinely load-bearing concepts will miss real ones. This is not merely a technical requirement — it is a subject-matter knowledge requirement. The specific knowledge needed is Ball et al.'s (2008) horizon content knowledge: the awareness of how topics in the subject relate structurally across the full span of the curriculum. A teacher or administrator who knows the subject only at the level they are teaching cannot reliably identify which concepts at that level are structurally consequential and which are peripheral, because identifying structural consequence requires knowing what is built on top of each concept. The tier map requires looking down at what is currently being taught from the vantage point of what comes after it — which requires knowing what comes after it at sufficient depth. In automated implementations, this requirement is addressed by having subject-matter experts build and validate the tier maps before they are deployed — the map construction is a one-time expert task, not a real-time teacher judgment.
The diagnostic framing is a necessary but not sufficient condition for accurate signal. The low-stakes framing is designed to remove the incentive for students to perform rather than accurately report. It will not succeed with every student. Students with acute test anxiety, strong social performance motivations, or significant prior experience of academic shame may produce distorted signal despite the framing. In these cases the diagnostic should be understood as an approximate starting estimate, with the session update log doing more of the calibration work in the first few sessions than the intake diagnostic alone.
The intake diagnostic is the input layer of the framework. Its output — the Dynamic Learning Profile — directly initializes every other component: the gradient assignment's difficulty calibration, the fluency protocol's target operations, the session structure's reading-comprehension framing, and the cultural approach that will be most effective for this student's specific prior history with academic environments. Without the diagnostic, the framework begins from a guess. With it, the framework begins from a map.
The diagnostic is the component of the framework most in need of empirical development. The other papers in this series document principles with substantial existing research support. This paper documents a protocol that is theoretically grounded but largely novel in its specific design. The research agenda it implies — reliability and validity studies for the schema adequacy diagnostic, empirical calibration of the fluency latency thresholds, validation of the tier-map methodology across subject domains — is substantial and constitutes the primary original research program the framework generates.
That is not a weakness of the diagnostic or of the framework. It is the honest description of where the frontier is. The diagnostic is where the most important and least explored work remains to be done.