Coverage Matrix
Coverage matrix is a routing sheet for new papers, posts, and concept notes. It helps decide which reusable notes should exist before a topic becomes a public post or curated paper note.
Intake Routes
| Need | Use | Start |
|---|---|---|
| Reusable wiki note | definition, boundary, contract, checks, links, public boundary | Wiki note quality gate |
| AI paper claim | architecture, objective, evidence, and system boundary | AI paper intake |
| AI claim pattern | architecture, learning, generation, evaluation, scaling, systems, or agent workflow | AI paper claim patterns |
| Computational biology paper claim | object, representation, label, split, leakage | Computational Biology paper intake |
| Computational biology claim pattern | property, activity, docking, generation, protein design, genome sequence modeling | Computational Biology paper claim patterns |
| Formula or objective | symbols, distributions, derivatives, metrics, constraints | Formula intake |
| Formula pattern | common objective or metric pattern before detailed derivation | Formula pattern catalog |
| AI + computational biology + Math claim | object, representation, model, objective, evidence, public boundary | AI Computational Biology Math contract |
| Benchmark score | data, task, split, metric, allowed information, reporting | Benchmark intake |
| Korean synthesis post | reader question, main axis, minimum formulas, wiki links | AI Computational Biology Math post intake, Wiki bundle checklist |
| Topic map | reader question, primary axis, object boundary, methods, evidence, next path | Topic map contract |
| Paper review workflow | metadata, claims, evidence, artifacts, reproduction | Paper review workflow |
| Claim route | primary axis, secondary axes, bucket, concept updates | Claim routing |
| Paper to wiki extraction | claim, object, representation, method, formula, data, evidence, artifact updates | Paper to wiki extraction |
| Concept update | reusable definitions, formulas, contracts, evidence boundaries | Concept update contract |
| Promotion readiness | route, representation, objective, evidence, and public boundary | AI Computational Biology Math readiness gate |
| Fillable paper note | reusable skeleton for AI/computational biology/Math papers | AI Computational Biology Math paper template |
| Paper bucket | decide whether the paper belongs in SBDD, protein modeling, architecture, generation, learning methods, or systems | Paper triage |
Axis Coverage
| Axis | Question | Canonical Area |
|---|---|---|
| Object | What entity is being modeled? | Entities, Computational biology entities |
| Modality | What form is the input or output? | Modalities |
| Representation | How does the raw object become tokens, graph, coordinates, embedding, or features? | Representation contract, Tensor shape notation |
| Graph construction | What nodes, edges, pair features, and inference-available context define the graph? | Graph construction, Graph neural networks, Graph Transformer |
| Task | What output space and validity rule define success? | Tasks |
| Coordinate contract | What frame, symmetry, mapping, loss, and metric define coordinate claims? | Coordinate modeling contract |
| Data | What example, label, split, and preprocessing contract define the dataset? | Data |
| Architecture | What inductive bias and complexity does the model use? | Architectures, AI architectures |
| Learning method | What supervision, pretraining, transfer, or preference signal is used? | Learning methods, AI learning methods |
| Objective-metric link | Does the optimized loss, reward, score, energy, or constraint support the reported metric and utility claim? | Objective-metric alignment |
| Model selection | Which checkpoint, hyperparameter, threshold, seed, prompt, filter, or decoding setting was selected? | Model selection |
| Generative model | What distribution, score, energy, velocity, or sampling path is modeled? | Generative models, AI generative models |
| Math | Which formula, object type, distribution, constraint, or estimator is needed? | Math, Math foundations, Constrained optimization |
| Evaluation | Which metric, baseline, split, uncertainty, and failure mode support the claim? | Evaluation, Evaluation concepts |
| Uncertainty and calibration | Is the result larger than uncertainty, and are probabilities meaningful? | Confidence interval, Calibration, Cross-validation |
| Claim boundary | What exactly does the evidence prove, and where does it stop? | Claim-evidence boundary |
| Systems | Does the claim depend on training, serving, tools, reproducibility, or compute? | AI systems, Infra |
| Scaling | Does quality depend on data size, model size, compute, memory, latency, or inference budget? | Scaling claim contract |
| Agent workflow | Does an LLM use tools, memory, planning, or verification? | Agents |
Computational Biology Coverage
| Topic | Must Cover | Route |
|---|---|---|
| Molecule or ligand model | standardization, identity, chemical state, representation, scaffold split, property or interaction label | Molecules, Chemical state contract |
| Protein model | sequence, structure, PLM/MSA/template policy, domains, residue indexing, family split, representation | Proteins, Protein language model |
| Protein-ligand complex | protein context, ligand context, pocket, pose, interaction, split on both axes | Structure-based modeling |
| Docking or virtual screening | receptor/ligand preparation, pose generation, scoring, filtering, enrichment metric | Docking |
| Geometry-heavy model | coordinate frame, invariant/equivariant target, rotation/translation behavior, leakage risk | Geometry for Structure Modeling |
| Genome sequence model | region, k-mer, variant, sequence context, split and annotation boundary | Genome Sequence Modeling |
| Computational biology benchmark | label semantics, assay/source, split unit, leakage, baseline, metric | Data and evaluation |
| Benchmark trap | negative construction, activity cliffs, applicability domain, assay harmonization, measurement ceiling | Negative set, Activity cliff, Applicability domain, Assay harmonization, Boltzmann ceiling analysis |
Minimum Note Bundle
Before promoting a paper cluster into a Korean post, check that the bundle has:
- One route page: AI paper intake, Computational Biology paper intake, or Formula intake.
- One claim pattern when the paper is in computational biology: Computational Biology paper claim patterns.
- One wiki-note quality pass before a concept becomes a reusable anchor: Wiki note quality gate.
- One data or benchmark page when a score is central: Benchmark intake.
- One topic-map contract when a cluster spans several axes: Topic map contract.
- One object/modality/task page when the input or output is nontrivial.
- One representation contract when preprocessing or featurization changes the object seen by the model.
- One tensor-shape pass when a formula depends on batch, token, graph, coordinate, head, or candidate axes.
- One formula-pattern pass when the paper introduces or relies on an objective or metric: Formula pattern catalog.
- One constraint pass when feasibility, validity, projection, repair, or filtering is central: Constrained optimization.
- One energy/score pass when a method uses energy, force, score matching, or Langevin-style sampling: Energy-based model.
- One coordinate modeling contract when a paper predicts poses, conformers, structures, vectors, forces, or coordinate updates.
- One architecture or learning-method page when the method is central.
- One objective-metric alignment check when the training loss and reported metric differ.
- One evaluation page explaining metric, split, baseline, uncertainty, or failure mode.
- One scaling claim pass when model size, data size, compute, memory, latency, or throughput is part of the contribution.
- One claim boundary when the result could be overread: Claim-evidence boundary.
- One paper note or paper-analysis note when a specific paper claim is being discussed.
- One claim-routing pass when a paper or post candidate spans multiple axes: Claim routing.
- One paper-to-wiki extraction pass before updating several support notes: Paper to wiki extraction.
- One AI claim-pattern pass when the paper’s method contribution is central: AI paper claim patterns.
- One cross-axis contract when AI, computational biology, and Math all matter: AI Computational Biology Math contract.
- One concept-update pass when a paper changes reusable definitions, formulas, contracts, or evidence boundaries: Concept update contract.
- One readiness gate pass before promoting a multi-axis candidate: AI Computational Biology Math readiness gate.
- One fillable template pass for mixed AI/computational biology/Math papers: AI Computational Biology Math paper template.
- One Korean post intake pass when the result is reader-facing: AI Computational Biology Math post intake.
- One post bundle pass before writing a synthesis post: Wiki bundle checklist.
Checks
- Is the central axis clear: object, method, formula, benchmark, paper cluster, or project?
- Are AI, computational biology, and Math links separated instead of collapsed into one vague topic?
- Does every performance claim have a route to benchmark intake or evaluation protocol?
- Does every computational biology claim name object, label, split, and leakage risk?
- Does every formula-heavy claim define symbols and the sampled distribution?
- Does every constrained or filtered-generation claim count invalid outputs?
- Does every probability claim distinguish accuracy, ranking, probability metric, and calibration?
- Is the public boundary clear before adding a post or paper note?