Parts

feature · paper proposed

An earlier declaration was accepted by Zach Mainen; it has moved since — this is 80f95d915c47, unaccepted. · no paper read under it

Which claims are components of other claims, and which whole does each belong to?

A capability that revises existing artifacts. It runs once, and then it is part of the schema.

Part of Tree — What does the paper assert, and how do its claims depend on each other?

How it works

Reads the tree, not the paper — each claim's slug, role, panel, sentence and existing edges — and returns the part-of edges it holds: every claim that states one comparison, condition, measure or study of a proposition another claim states whole. The graph half-says composition already, since a part usually carries a supports or tests edge into its whole, but support is evidence and composition is not, so the definition decides, not the edge. It then writes part-of: into each part's claim file, editing the frontmatter in place.

--dump-prompt writes the exact request and exits, and --answer feeds a reply back through the same validation — an unknown slug, a self-edge, a second whole for one part, or an edge that would close a cycle is dropped rather than written — so a model with no backend here can still answer it.

Rests on

Feeds

Nothing — this is a leaf.

How it is defined

A model answers this layer, so the prompt is the layer. It is reproduced below from the committed file, and it is a declared input — editing it makes every run that used it stale.

extract/prompts/parts.mdthe prompt it runs undernot in the repository

The declaration names this path and the repository does not have it. An input that does not exist hashes to nothing, so it cannot make a run stale — the layer is declared to depend on something it is not in fact tracking.

extract/prompts/contract/vocabulary.mdthe prompt it runs undernot in the repository

The declaration names this path and the repository does not have it. An input that does not exist hashes to nothing, so it cannot make a run stale — the layer is declared to depend on something it is not in fact tracking.

What it produces10 parts

One paper, as the worked example — Gadeke, v3. Read from runs/gadeke-2026-guilt-insula/parts.output.json · 3 KB. model supplied:runs/gadeke-2026-guilt-insula/parts.answer.v3.json

# partwholewhy
1 during-outcome-phase-responses-low insula-rois-responded-more-low The raw low-outcome insula responses (Social > Partner) are one measure of the insula ROI guilt result, which subtracts the high-outcome response from this.
2 difference-response-between-low-high insula-rois-responded-more-low The low-minus-high difference being larger in Social is the 'even after subtracting high outcomes' clause of the insula ROI guilt result stated on its own.
3 responsibility-model-yielded-higher-values likelihood-ratio-test-showed-responsibility The R-squared comparison across models is one leg of the model-selection verdict that the Responsibility model fitted the happiness data best.
4 participant-momentary-happiness-varied-rewards responsibility-redux-model-incorporating-expected That happiness varied with the participant's own rewards is one component of the reward dependencies the Responsibility Redux model was fitted to capture.
5 participant-momentary-happiness-varied-rewards-2 responsibility-redux-model-incorporating-expected That happiness varied with the partner's rewards is the other reward-source component the Responsibility Redux model captures.
6 participants-chose-risky-option-lottery participants-showed-very-similar-risk The frequency of risky choice across Solo and Social is one of the three measures the 'similar risk preferences' conclusion integrates.
7 participants-slightly-more-risk-averse participants-showed-very-similar-risk The risk-aversion parameter rho comparison is the second risk measure the 'similar risk preferences' conclusion integrates.
8 risk-premiums-not-differ-between participants-showed-very-similar-risk The risk-premium comparison is the third measure feeding the 'similar risk preferences' conclusion, reconciled with the small Study 1 rho tendency.
9 guilt-effect-occurred-whether-participant both-studies-participants-felt-worse That the guilt effect held whether the participant's own outcome was high or low is one robustness condition of the behavioural guilt effect.
10 responsibility-choices-not-influence-happiness both-studies-participants-felt-worse The null for high partner outcomes states the boundary condition that the guilt effect is specific to low outcomes.

Across the corpus

9 not run · 1 stale·a paper links to its own cell, where this layer's output for it is rendered

Inputs and outputs

Reads, besides its dependencies
Produces
  • runs/{paper}/parts.output.json

One per paper — the table above links each one that exists.

Views
  • table — rendered above, over the 10 parts in the artifact
  • graph — on the paper page, as the claim graph

Running it

The command comes from the declaration, so this text and what actually runs cannot diverge. pipeline.py run also runs the unmet dependencies first.

python3 scripts/pipeline.py run <paper> parts

Underneath, that runs cd extract && python3 -m claim_graphs.cli parts --paper {paper}.