Exam Readiness Index (ERI): A Theoretical Framework for a Composite, Explainable Index

Ananda Prakash Verma · 2025

Read Exam Readiness Index (ERI): A Theoretical Framework for a Composite, Explainable Index by Ananda Prakash Verma (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-00718.txt.

Introduction

Ananda Prakash Verma, writing alone, builds theory for an Exam Readiness Index. The index is a single score from zero to one hundred. It sums up how ready a learner is for a high-stakes exam. The score stays interpretable and actionable by design. It is also blueprint-aware, so it respects the exam's topic plan.

Verma builds the index from six signals. The signals are mastery, coverage, retention, pace, volatility, and endurance. They come from practice tests and mock exams. He states axioms for each part and for the whole. He proves the composite is monotone and stable. Drift stays bounded when blueprint weights change. A unique best linear composite exists under convex design limits. Empirical study is left to future work.

Catalog record

CREATORAnanda Prakash Verma
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE14 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-00718

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Exam Readiness Index (ERI): A Theoretical Framework for a Composite, Explainable Index Ananda Prakash Verma Abstract We present a theoretical framework for an Exam Readiness Index (ERI): a composite, blueprint-aware score R ∈ [ 0,100 ] R\in[0,100] that summarizes a learner’s readiness for a high-stakes exam while remaining interpretable and actionable. The ERI aggregates six signals—Mastery ( M ) (M) , Coverage ( C ) (C) , Retention ( ℛ ) (\mathcal{R}) , Pace ( P ) (P) , Volatility ( V ) (V) , and Endurance ( E ) (E) —each derived from a stream of practice and mock-test interactions. We formalize axioms for component maps and the composite, prove monotonicity, Lipschitz stability, and bounded drift under blueprint re-weighting, and show existence and uniqueness of the optimal linear composite under convex design constraints. We further characterize confidence bands via blueprint-weighted concentration and prove compatibility with prerequisite-admissible curricula (Knowledge Space / learning-space constraints). The paper focuses on theory; empirical study is left to future work. 1 Introduction High-stakes exam preparation hinges on answering four questions: Am I ready now? Ready for what? What’s limiting me? What should I do next? We propose a principled, explainable readiness index that is blueprint-aware : it respects an official syllabus with section/topic weights and pacing rules, and it decomposes into human-interpretable components that align with cognitive and operational constructs (ability, coverage, forgetting, timing, stability, endurance). This paper makes four contributions: 1. Axiomatic components. We specify component maps from interaction streams to [0,1] with regularity properties (boundedness, Lipschitz continuity, monotonicity). 2. Blueprint-aware composite. We define a linear ERI R=\sum_{i}\alpha_{i}X_{i} with design constraints and show stability and interpretability guarantees. 3. Confidence and drift. We derive blueprint-weighted concentration bounds for component estimates and a bound on ERI drift under blueprint version change. 4. Prerequisite compatibility. We provide a sufficient condition ensuring ERI-driven recommendations respect admissibility in a learning space (outer-fringe). Scope. Our goal is a deployable theoretical scaffold. We do not perform parameter estimation or empirical validation; those are orthogonal and deferred to future work. 2 Related Work Knowledge Space Theory (KST). KST models feasible knowledge states as union-closed families of item sets with natural learning properties (accessibility, well-gradedness) [ 1 , 2 ] . ERI is complementary: it provides a continuous readiness summary while remaining compatible with KST’s admissible-next (outer fringe) principle (Section 6 ). Item Response Theory (IRT). IRT [ 3 , 4 ] models a latent ability parameter and item characteristics to predict responses. ERI does not require a single latent trait; instead, it aggregates multiple blueprint-weighted signals. When available, IRT-calibrated difficulties can instantiate the Mastery component M . Spaced Repetition and Retention. A large body of work—tracing to Ebbinghaus’ forgetting curve—shows that spaced retrieval improves long-term retention [ 5 , 6 , 7 ] . ERI’s Retention component \mathcal{R} formalizes a blueprint-weighted exponential decay that is compatible with trainable schedulers (e.g., Leitner/FSRS-style models). 3 Blueprints, Signals, and Notation Let \mathcal{S} denote sections and \mathcal{T} topics. A blueprint B provides nonnegative weights \{w_{t}\}_{t\in\mathcal{T}} with \sum_{t}w_{t}=1 , a duration T_{\mathrm{exam}} , marking rules, and section-level pace targets \{\tau_{s}\}_{s\in\mathcal{S}} (seconds per item). Interaction logs induce per-topic sequences \mathcal{D}_{t} (attempt outcomes, timestamps, times), and per-section sequences \mathcal{D}_{s} . We define six normalized component maps (all into [0,1] ): \displaystyle M \displaystyle:=\sum_{t\in\mathcal{T}}w_{t}\,m_{t}(\mathcal{D}_{t}),\quad C:=\sum_{t\in\mathcal{T}}w_{t}\,c_{t}(\mathcal{D}_{t}),\quad\mathcal{R}:=\sum_{t\in\mathcal{T}}w_{t}\,r_{t}(\mathcal{D}_{t}), \displaystyle P \displaystyle:=\frac{1}{|\mathcal{S}|}\sum_{s\in\mathcal{S}}p_{s}(\mathcal{D}_{s}),\quad V:=v(\mathcal{D}),\quad E:=e(\mathcal{D}). Assumption 1 (Component axioms) . For each topic t and section s , the maps satisfy: (A1) Normalization: m_{t},c_{t},r_{t},p_{s},v,e\in[0,1] for any admissible data stream. (A2) Monotonicity: m_{t} is nondecreasing in difficulty-adjusted success; c_{t} is nondecreasing in (evidence-weighted) coverage; r_{t} is nonincreasing in recency gaps; p_{s} is nonincreasing in pace deviation; v is nonincreasing in score variance; e is nonincreasing in late-session degradation. (A3) Lipschitz regularity: For appropriate metrics d_{t},d_{s} on streams, there exist L_{m},L_{c},L_{r},L_{p},L_{v},L_{e} with |m_{t}(\mathcal{D}_{t})-m_{t}(\mathcal{D}^{\prime}_{t})|\leq L_{m}\,d_{t}(\mathcal{D}_{t},\mathcal{D}^{\prime}_{t}),\quad\text{etc.} (A4) Blueprint separability: Topic-level components aggregate linearly with weights \{w_{t}\} . Instantiation examples (non-essential). One may take r_{t}(\mathcal{D}_{t})=\exp(-\lambda_{t}\Delta_{t}) with \Delta_{t} the time since last successful retrieval and \lambda_{t}>0 ; map section pace to p_{s}=\phi(z_{s}) for a clipped linear \phi of a pace z -score; and define coverage gates by evidence indicators. Our results require only Assumption 1 . 4 The ERI Composite and Design Constraints We define the readiness index as a convex combination of components: R(\mathcal{D};B,\alpha)\;=\;\alpha_{M}M+\alpha_{C}C+\alpha_{R}\mathcal{R}+\alpha_{P}P+\alpha_{V}V+\alpha_{E}E, (1) with weights \alpha\in\mathbb{R}^{6}_{\geq 0} and \sum_{i}\alpha_{i}=1 . 4.1 Axioms for the composite Definition 1 (ERI axioms) . A composite R is valid if it satisfies: (i) Boundedness R\in[0,1] ; (ii) Monotonicity in each argument; (iii) Blueprint coherence \partial R/\partial w_{t}\geq 0 whenever m_{t},c_{t},r_{t} weakly increase; (iv) Scale-invariance to monotone reparameterizations of individual components (achieved by pre-normalization). Proposition 1 (Validity of the linear ERI) . Under Assumption 1 , the linear R in ( 1 ) is valid per Definition 1 . Proof. Boundedness and monotonicity follow from nonnegativity and normalization. Blueprint coherence follows by linearity and separability. Scale-invariance holds as each component is normalized to [0,1] . ∎ 4.2 Weight selection as convex design Let \mathcal{C} denote design constraints (e.g., minimum emphasis on mastery and coverage, fairness caps per section), and J(\alpha) a strictly convex penalty (e.g., quadratic deviation from a prior \alpha^{0} and entropy regularization for spread). Consider \min_{\alpha\in\Delta_{5}}\;J(\alpha)\quad\text{s.t. }\alpha\in\mathcal{C}, (2) where \Delta_{5} is the 5-simplex. Theorem 1 (Existence and uniqueness) . If J is strictly convex and \mathcal{C} is convex and nonempty, the solution \alpha^{\star} to ( 2 ) exists and is unique. Proof. Standard convex-optimization result. ∎ Lemma 1 (Lagrangian KKT form) . Let J(\alpha)=\tfrac{1}{2}\|\alpha-\alpha^{0}\|_{2}^{2}-\eta\sum_{i}\alpha_{i}\log\alpha_{i} . Then the KKT conditions yield \alpha_{i}^{\star}\propto\exp\left(\alpha_{i}^{0}/\eta-\lambda_{i}\right), with multipliers \lambda_{i} enforcing the linear constraints in \mathcal{C} and the simplex. 5 Stability and Confidence 5.1 Lipschitz stability Theorem 2 (ERI Lipschitz constant) . Let d(\mathcal{D},\mathcal{D}^{\prime}) be a metric that dominates per-topic/section metrics. Under Assumption 1 , |R(\mathcal{D})-R(\mathcal{D}^{\prime})|\leq\Big(\alpha_{M}L_{M}+\alpha_{C}L_{C}+\alpha_{R}L_{R}+\alpha_{P}L_{P}+\alpha_{V}L_{V}+\alpha_{E}L_{E}\Big)\,d(\mathcal{D},\mathcal{D}^{\prime}). Proof. Triangle inequality and linearity. ∎ 5.2 Blueprint drift Suppose a blueprint update changes weights from w to w^{\prime} with total-variation \delta=\tfrac{1}{2}\sum_{t}|w_{t}-w^{\prime}_{t}| . Proposition 2 (Bounded ERI drift under re-weighting) . With components fixed, |R(\mathcal{D};w)-R(\mathcal{D};w^{\prime})|\;\leq\;\delta\cdot\max\{\|m\|_{\infty}+\|c\|_{\infty}+\|r\|_{\infty},\,1\}. If all topic components lie in [0,1] , the drift is at most 3\delta for the topic-aggregated terms. Proof. By linearity, the difference is a weighted sum of bounded topic terms, bounded by the \ell_{1} -distance of weights. ∎ 5.3 Confidence band via blueprint-weighted concentration Let \widehat{m}_{t} estimate m_{t} from n_{t} independent item outcomes in topic t with bounded range. By Hoeffding’s inequality, \Pr\Big(|\widehat{m}_{t}-m_{t}|\geq\epsilon_{t}\Big)\leq 2\exp(-2n_{t}\epsilon_{t}^{2}). Aggregating with blueprint weights gives, for any \epsilon>0 , \Pr\Big(|\widehat{M}-M|\geq\epsilon\Big)\leq 2\exp\!\left(-\frac{2\epsilon^{2}}{\sum_{t}w_{t}^{2}/n_{t}}\right). (3) Analogous bounds hold for \widehat{C},\widehat{\mathcal{R}} when estimated from bounded per-topic statistics. A union bound yields an overall ERI confidence band whose half-width shrinks with blueprint-weighted effective sample size \left(\sum_{t}w_{t}^{2}/n_{t}\right)^{-1} . 6 Prerequisite-Admissible Recommendations Let (Q,\mathcal{K}) be a learning space (union-closed, accessible) over items/skills with outer fringe F^{+}(X)=\{q\notin X:X\cup\{q\}\in\mathcal{K}\} . A recommendation policy is admissible if it selects only items whose concepts lie in F^{+}(X) or their immediate prerequisites. Assumption 2 (ERI-compatible gating) . Action maps use only concepts with nonzero coverage evidence and admissible with respect to the current state estimate X . Proposition 3 (ERI actions respect admissibility) . Under Assumption 2 , any action chosen by minimizing a separable surrogate loss of the form \sum_{t}\psi_{t}(1-M_{t},C_{t},\mathcal{R}_{t}) over admissible concepts respects the outer fringe and cannot introduce prerequisite violations. Proof. By construction, the feasible set is F^{+}(X) ; separability

Read the complete text