register-predicates

name	register-predicates
description	Author the DeLP/Datalog predicate declaration artifacts a paper's rules will use. Required before author-rules if the paper's rules reference predicates not yet declared.
argument-hint	<papers/Author_Year_Title>
disable-model-invocation	false
compatibility	Claude Code, Codex CLI, and Gemini CLI.

Register Predicates: $ARGUMENTS

Declare the DeLP predicates a paper's rules will use. Predicates carry a name, arity, per-argument sort, optional description, and optional derived_from grounding DSL.

Use pks predicate add (propstore >= 0.2.0). The CLI handles YAML authoring, schema validation, and git commit uniformly. Do NOT hand-author YAML or run git commit yourself.

Theoretical Background

DeLP (Garcia & Simari 2004) and the Diller/Borg/Bex 2025 grounding work require every rule atom to reference a declared predicate. The declaration fixes arity and per-position sort, which the grounder uses to enumerate well-typed Herbrand substitutions. Arity 0 is admitted as propositional facts; arity 1 (bird/1) is the canonical defeasible-reasoning example.

Step 0: Validate

paper_dir="$ARGUMENTS"
ls "$paper_dir"/notes.md 2>/dev/null || echo "MISSING: notes.md"
ls knowledge/.git 2>/dev/null || echo "MISSING: knowledge/.git"

Step 1: Derive Predicate Set From Notes

Read notes.md. Identify the paper's causal, mechanistic, or clinical predicates — the atomic predicates the paper's stated reasoning steps turn on. Examples for RCT papers:

• Study-scope predicates: <paper>_like_case/1 or <paper>_like_setting/1 when the paper really needs a source-specific applicability predicate.
• Effect predicates: intervention_reduces_outcome/1, intervention_increases_outcome/1, method_improves_metric/1, method_increases_cost/1.
• Recommendation predicates: intervention_indicated_for_use_case/1, method_preferred_for_task/1, or model_suitable_for_setting/1.

Arity selection: most clinical-paper predicates are unary (parameterized by cohort). Binary may appear for relations (supersedes/2). Arity 0 is for global propositional facts.

Prefer descriptive predicate names over opaque abstraction. intervention_reduces_target_outcome/1 beats reduces/3 when the rule language is source-local. Do not include a paper, author, venue, or collection-specific prefix unless the predicate is genuinely about that source's applicability boundary.

Step 2: Choose arg_types

arg_types is a tuple of sort names, one per position. For cohort-style unary predicates, use entity (the propstore seed concept for generic entities). For domain-specific sorts, use a concept name that exists in the registry.

Step 3: Register Predicates Via CLI

Use the authoring group stem <author>_<year> (e.g., author_2024) as optional source metadata with --file. Each pks predicate add call writes one canonical predicate artifact named for the predicate id under the knowledge/predicates/ directory. Duplicate predicate ids are rejected across canonical predicate artifacts.

cd knowledge  # or pass -C to each pks call

pks predicate add \
  --file author_2024 \
  --id intervention_reduces_target_outcome \
  --arity 1 \
  --arg-type entity \
  --description "The studied intervention reduces the rate or level of the target outcome in this setting."

pks predicate add \
  --file author_2024 \
  --id study_like_setting \
  --arity 1 \
  --arg-type entity \
  --description "Entity belongs to a setting comparable to the paper's study setting."

Optional flags:

• --derived-from <dsl-string> — how propstore should materialize ground atoms from repo data. Recognised forms live in propstore.grounding.predicates. Omit for abstract (non-grounded) predicates whose truth will be asserted by rules, not data.
• --description — short human-readable explanation. Always supply for RCT predicates.

Step 4: Verify

pks build

Expect: Build rebuilt: or Build unchanged: with zero warnings. If build fails, a predicate declaration has a schema error — diagnose and re-run pks predicate add with corrections.

Output

Predicates registered as one artifact per predicate id
  Predicates: N total (one emit_success per add)

When To Rerun

Rerun this skill if author-rules needs predicates you haven't declared. Use additional pks predicate add calls with the same --file authoring group metadata for readability. Canonical storage is one predicate artifact per predicate id, not a per-paper predicates bucket.