Art 0.1.0

Built from art-dev 78fb10b.

prompts/README.md

@@ -0,0 +1,15 @@
+# Prompts
+
+One file per prompt, named `<task>_v<N>.txt`. Placeholders are `{name}`;
+anything else in the file is sent as written. `art/llm.py` records the
+prompt name and a version string (`v<N>+<content hash>`) with every call,
+so an edit is tracked even if the file is not renamed. Make a new version
+file for a deliberate change you want to compare against the old one.
+
+| File | Task | Used by |
+|---|---|---|
+| summarize_v1.txt | 2-4 sentence summary from the abstract | digest |
+| label_v1.txt | 3-6 word topic label, for the field-pulse section | digest |
+| explain_v1.txt | one-sentence match reason plus a 0-3 relevance rating (JSON) | digest |
+| llm_score_v1.txt | LLM-as-classifier baseline: 0-3 rating against the whole profile | evaluation only |
+| profile_draft_v1.txt | draft a profile name and description from seed papers | bingham init |

prompts/explain_v1.txt

@@ -0,0 +1,12 @@
+A researcher follows this topic:
+
+{profile_name}: {profile_description}
+
+Here is a paper:
+
+Title: {title}
+Abstract: {abstract}
+
+In one sentence, state what in this paper connects it to the topic, naming the specific system, method, or finding. Then rate how relevant the paper is to the topic: 0 = not relevant, 1 = tangentially relevant, 2 = relevant, 3 = highly relevant.
+
+Respond in JSON with two keys: "reason" (a string) and "relevance" (an integer from 0 to 3).

prompts/label_v1.txt

@@ -0,0 +1,12 @@
+Assign a short topic label (3-6 words) to this paper.
+Examples: colloidal gel rheology, polymer viscoelasticity, emulsion stability,
+yield stress fluids, microrheology techniques, discontinuous shear thickening,
+suspension microstructure, interfacial rheology, active matter rheology,
+AI tutoring in physics, retrieval-augmented generation, LLM evaluation benchmarks,
+scientific information extraction.
+Return only the label, with no punctuation and no explanation.
+
+Title: {title}
+Abstract: {abstract}
+
+Topic:

prompts/llm_score_v1.txt

@@ -0,0 +1,15 @@
+You are evaluating whether a paper is relevant to a researcher's interests.
+
+Researcher interests:
+{profile_text}
+
+Paper title: {title}
+Paper abstract: {abstract}
+
+Rate relevance on this scale:
+0 = not relevant
+1 = tangentially relevant
+2 = relevant
+3 = highly relevant
+
+Reply with a single digit (0, 1, 2, or 3). Nothing else.

prompts/profile_draft_v1.txt

@@ -0,0 +1,10 @@
+Below are papers that a researcher chose as typical of one of their research interests.
+
+{papers}
+
+Write a profile of this research interest for a literature screener that will look for new papers on the same topic.
+
+- "name": a short topic name, at most five words.
+- "description": two or three sentences in the field's own technical vocabulary, describing the systems, methods, and questions these papers have in common. Begin with the subject itself (for example "Rheology of dense suspensions ..." or "Language models as ..."), not with "This research" or "This topic". Describe the topic, not the individual papers. Do not mention authors, titles, or years.
+
+Respond in JSON with two keys: "name" (a string) and "description" (a string).

prompts/summarize_v1.txt

@@ -0,0 +1,6 @@
+Summarize the following paper abstract in 2-4 sentences. Be specific about the methods, materials or systems, and key findings. Use precise scientific terminology. Do not add context not present in the abstract.
+
+Title: {title}
+Abstract: {abstract}
+
+Summary: