Paper To Wiki Extraction

Paper to wiki extraction turns one paper into reusable notes without turning the site into a pile of paper summaries. The paper note keeps source-specific evidence; wiki notes keep definitions, formulas, contracts, and failure modes that future posts can reuse.

$$\text{paper}\to \text{claims}\to \text{wiki updates}\to \text{post bundle}$$

Extraction Passes

Pass	Extract	Update
Claim	one sentence claim, supported scope, unsupported scope	Claim extraction, Claim-evidence boundary
Object	modeled entity, example unit, output unit, context	Entities, Computational biology entities
Representation	token, graph, coordinate, embedding, conformer, feature cache	Representation contract
Method	architecture, learning signal, sampler, inference procedure	AI, AI paper claim patterns
Formula	objective, likelihood, estimator, metric, update rule	Formula intake, Formula pattern catalog
Data	dataset, preprocessing, split unit, leakage risk, label semantics	Benchmark intake, Computational Biology paper intake
Evidence	metric, baseline, ablation, uncertainty, strongest/weakest result	Evidence table, Benchmark claim contract
Artifact	code, data, split, weight, config, environment, reproducibility status	Reproducibility checklist
Public boundary	private information, unpublished results, collaborator details	Publishing gate
Post route	reader question, next path, minimum support notes	move to the Posts workflow only after paper and concept notes are stable

Claim Decomposition

For each important claim, write the local decomposition before editing wiki notes.

claim: to verify
paper_scope: to verify
object: to verify
representation: to verify
method: to verify
formula: to verify
data_and_split: to verify
metric: to verify
baseline: to verify
uncertainty: to verify
artifact_status: to verify
wiki_updates:
  - to verify
post_candidate: not applicable

Update Decision

If The Paper Adds	Prefer
a reusable term or boundary	concept note update
a new object or example unit	entity or data note update
a new input construction	representation contract update
a new model family or block	architecture note update
a new training signal	learning-method or objective note update
a new sampler or distribution path	generative model and Math note update
a new benchmark protocol	benchmark or evaluation note update
only a paper-specific score	paper note only
only an unverified candidate	inbox item only
several linked reusable updates	Korean synthesis post candidate

AI + Computational Biology + Math Pattern

Mixed papers usually need three synchronized updates:

1. Computational Biology: object, label, split, leakage, and deployment context.
2. AI: architecture, learning signal, sampler, inference cost, and evaluation risk.
3. Math: objective, distribution, metric, symbol table, and estimator boundary.

Do not let one axis hide the others. A protein-ligand paper can be a computational biology paper even when it uses a Transformer; a new score or velocity objective still needs Math; a benchmark score still needs an evidence boundary.

Stop Conditions

• The source is not public.
• The paper metadata is missing and cannot be marked to verify.
• The claim cannot be separated from private project context.
• The formula is copied without local symbol definitions.
• The benchmark lacks split, metric, or baseline context.
• The wiki update would only repeat a paper summary.
• The post would have no reusable support notes.

Related

• Paper review workflow
• Claim routing
• Concept update contract
• AI Computational Biology Math readiness gate
• Coverage matrix