Changelog

2.1.0 — 2026-08-01

Added explicit visual and text-only execution routes for models without image understanding.
Added inventory --text-only, which emits page text, per-visual text cards, body-reference contexts, and a text evidence ledger without rendering PNG assets.
Added A/B/C/D text evidence grades and strict rules against presenting caption/OCR inference as direct visual observation.
Added validate_report.py --text-only checks for disclosure, text-review completion, source recording, unverified-crop handling, and source-map execution metadata.
Upgraded the source-map template to schema v3 with visual capability and verification fields.
Documented structured-source, PDF-text, OCR, user-description, and human/vision-model handoff paths.

Broadened the default audience from computer-vision Ph.D. readers to research-trained readers across disciplines.
Added structured routing for domain, audience, goal, depth, and language.
Added project-level .paper-reader.yaml preferences and an example configuration.
Separated method, theory, empirical/observational, dataset, system, and review paper types from domain-specific evidence norms.
Added domain lenses for AI/CS, biomedicine, physics/mathematics, chemistry/materials, engineering, social science, earth/environment, and humanities/qualitative research.
Kept computer vision as a deeply supported optional lens.
Generalized report, reading, quality, and source-map protocols from model/experiment language to methods, theory, observation, qualitative evidence, and other research designs.
Extended PDF caption detection to Scheme, Plate, Box, Chart, supplementary visuals, and Extended Data visuals.
Added multilingual elevator-pitch validation and reader-profile validation for source-map schema v2.

Initial open-source release with source-grounded deep reading, visual extraction, report validation, and computer-vision-focused review guidance.