AI Computational Biology Math Readiness Gate
Use this gate before promoting a paper candidate, topic map, or Korean synthesis post that mixes AI, computational biology, and Math. The goal is to keep a public note honest: what is verified, what is only a paper claim, and what still needs support.
Gate Order
| Gate | Pass When | Start |
|---|---|---|
| Source and metadata | Public source, title, authors, year, venue, and links are checked or marked to verify | Paper review workflow |
| Claim route | Primary and secondary axes are chosen before writing | Claim routing |
| Paper to wiki extraction | Paper claims are decomposed into reusable object, method, formula, evidence, and artifact updates | Paper to wiki extraction |
| Cross-axis contract | Object, representation, model, objective, evidence, and public boundary are all accounted for | AI Computational Biology Math contract |
| Representation contract | Raw object, representation, unit, preprocessing, split, and leakage risk are explicit | Representation contract |
| Coordinate contract | Frame, symmetry, atom or residue mapping, coordinate loss, and metric are explicit when 3D claims matter | Coordinate modeling contract |
| Objective-metric alignment | Training loss, selection metric, reported metric, and utility claim are connected | Objective-metric alignment |
| Objective family | Likelihood, contrastive, score, energy, velocity, reward, constraint, or metric objective is named | Formula intake, Energy-based model, Constrained optimization |
| Benchmark claim | Dataset, split, task, metric, baseline, uncertainty, and allowed information support the stated claim | Benchmark claim contract |
| Claim-evidence boundary | The supported scope is narrower than the headline claim when evidence is limited | Claim-evidence boundary |
| Artifact readiness | Code, data, splits, configs, weights, logs, predictions, and environment are checked for the scoped claim | Artifact availability, Implementation readiness |
| Computational biology intake | Molecule, ligand, protein, pocket, conformer, complex, genome region, assay, split, and leakage fields are handled when relevant | Computational Biology paper intake |
| Formula intake | Symbols, distributions, objectives, estimators, and metrics are rewritten in local notation | Formula intake |
| Coverage matrix | Missing object, method, math, data, evaluation, system, or agent notes are identified | Coverage matrix |
| Post route | Korean reader-facing synthesis has one question, one main axis, minimum formulas, and next links | Use the post workflow from the Posts section when a public essay is needed |
Paper Note Minimum
| Part | Minimum Content |
|---|---|
| Metadata | Public source, title, authors, year, venue, DOI/arXiv/project link if available |
| Route | Primary axis, secondary axes, paper bucket, concept updates |
| Claim list | Claims written as paper claims, not as established facts |
| Object and representation | Raw unit, model input, preprocessing, split unit, leakage risks |
| Objective | Key loss or estimator with symbol definitions |
| Evidence | Benchmark, split, metric, baseline, uncertainty, strongest evidence, weakest evidence |
| Artifact status | Code, data, splits, weights, configs, logs, predictions, environment |
| Implementation readiness | Scoped claim, minimum viable reproduction, compute feasibility, failure taxonomy |
| Public status | inbox, reading, verified, archived, or to verify |
Claim-Type Gates
Different mixed AI, computational biology, and Math papers fail in different ways. Use the matching row before promotion.
| Claim Type | Must Name | Minimum Evidence |
|---|---|---|
| Architecture improvement | input representation, block changed, complexity, baseline architecture | fair baseline, ablation, compute or parameter boundary |
| SSL or pretraining | pretraining unit, corruption/positive pair, transfer protocol | downstream split, linear probe or fine-tune protocol, leakage check |
| Generative model | sampling distribution, objective, sampler budget, validity definition | validity, diversity, novelty, utility, invalid-sample denominator |
| Energy or score model | energy, score, force, sampler, and negative-sample or noise process | partition/sampler handling, score-energy interpretation, validity after filtering |
| Constrained generation | hard constraint, penalty, projection, repair, or filter | invalid-output denominator, deployment-available constraints, constraint ablation |
| Molecular property/activity | molecule state, target, assay, endpoint, unit, threshold | split unit, assay/source handling, negative construction, calibration if probabilistic |
| Protein modeling | sequence/structure source, PLM/MSA/template policy, residue mapping | family split, template leakage check, likelihood-vs-function boundary |
| Docking or pose | receptor state, ligand state, pocket definition, pose metric | pose quality, failed docking denominator, leakage/template policy |
| Benchmark paper | task definition, allowed information, metric, selection rule | baseline strength, uncertainty, saturation, failure modes |
| Formula or estimator | random variables, distributions, objective, estimator | operational form, assumptions, metric relation, numerical budget |
| Agent or tool workflow | task state, tool boundary, verifier, memory, stopping rule | success/failure definition, evaluator leakage, human review boundary |
If the paper spans multiple claim types, all relevant rows apply.
Evidence Ladder
Use the weakest true level when writing status:
| Level | Meaning |
|---|---|
mentioned | paper text states it, but the note has not checked support |
specified | dataset, split, metric, and formula are identified |
supported | figure/table/ablation or experiment supports the claim |
bounded | limitations, failure modes, and applicability domain are stated |
reproducible | public artifacts are sufficient to rerun or audit the result |
Do not write a claim as established fact unless it reaches at least supported, and do not imply deployment confidence unless it reaches bounded.
Missing-Information Defaults
| Missing Item | Default Status |
|---|---|
| unknown metadata | to verify |
| unavailable code or weights | not released or to verify |
| unclear split | claim not ready |
| unclear label semantics | claim not ready for computational biology tasks |
| formula without symbol definitions | formula intake needed |
| metric without baseline or uncertainty | weak evidence |
| probability metric without calibration check | weak evidence |
| constraint without invalid-output accounting | claim not ready |
| private or unpublished detail needed | do not publish |
Synthesis Post Minimum
- One reader question.
- One primary axis: AI method, computational biology object, Math objective, benchmark, system, or agent workflow.
- Minimum formulas with symbol definitions when equations explain the claim better than prose.
- Links to wiki notes instead of repeating long definitions.
- Evidence boundary for benchmark, split, metric, leakage, baseline, and uncertainty claims.
- Next reading path across AI, Computational Biology, Math, Papers, Projects, Infra, or Agents.
Stop Conditions
Do not promote the candidate yet when:
- The source is not public.
- Core metadata is unknown and cannot be marked clearly.
- The note depends on private infrastructure, accounts, internal task names, collaborator details, or unpublished results.
- The main claim cannot be separated from unsupported interpretation.
- The object, representation, split, or metric is unclear for a computational biology claim.
- The formula is copied without local symbol definitions.
- The post would only summarize a news item without reusable wiki notes.
Readiness Record
source: to verify
status: inbox
primary_axis: to verify
secondary_axes: to verify
paper_to_wiki_extraction: to verify
representation_contract: to verify
coordinate_contract: not applicable
objective_metric_alignment: to verify
objective_family: to verify
benchmark_claim_contract: to verify
claim_evidence_boundary: to verify
computational_biology_intake: not applicable
formula_intake: to verify
coverage_matrix: to verify
artifact_availability: to verify
implementation_readiness: to verify
post_route: not applicable
public_boundary: to verify
claim_type_gate: to verify
evidence_level: mentioned
missing_information_defaults_applied: to verify