feat: in-text citations — visual styling, per-unit splitting, ADS search fallback

fix: preserve non-referee cache key format — only add ctx_hash element for referee mode

feat: referee rework — paper context injection, grounded critique, concrete checks, caveat ban

fix: close pymupdf doc in extract_equations; guard short anchor snippets; use grid layout for equation tag

feat: equation rendering — heuristic detector, PNG clip, anchor to paragraph, Reader display

fix: timer cleanup on unmount in MarginNote/NotebookDrawer; wire live userNotes/paperNote into onSaveSession

feat: reader notes — margin note cards, notebook drawer, paper-level note, export

fix: document _CITE_UNIT.lastIndex reset intent — module-level /g regex footgun

feat: in-text citations — splitCitations, cite spans, slim ref popover, muted styling

feat: references parser + session v2 — corpus matching, citeIndex, new schema models

fix: heading leakage in flush — clear pending_headings unconditionally; strip trailing stars from list-item headings

fix: parser fidelity — pending buffer, refs capture, heading chain, hygiene, caption gate, length basis

docs: Beacon v1.1 implementation plan — 6 tasks, parser fidelity through referee rework

docs: beacon v1.1 spec — parser fidelity fixes, referee rework, equations, citations, notes

fix: regenerate package-lock with complete optional-dep entries for linux builds

feat: Hugging Face Spaces packaging — corpus bootstrap, static serving, Dockerfile

Merge build-v1: Beacon — annotated astronomy paper reader

fix: final-review notes — true stage labels, sanitized network errors, dead line

docs: beacon addendum — variable retrieval, text input, rename

feat: retrieved-works preference, text input, beacon rename

feat: top-10 retrieval pool; plaintext/markdown manuscript input

fix: scrollable popovers; default annotation mode is historical context

fix: filter picture text, strip markdown debris, merge open-sentence continuations

fix: bypass pymupdf4llm layout analysis on vector-heavy pages

feat: figure mention popovers with caption annotator

test: make figures traversal test exercise actual traversal payloads

feat: figure extraction, caption retrieval, figures endpoint

docs: figures & captions spec addendum and plan (tasks 19-21)

fix: drop author/affiliation front matter; show first three authors et al.

fix: parse CRLF-framed SSE events from sse-starlette

chore: gitignore local corpus data

docs(spec): record compact-corpus amendment from Task 11

feat: compact local corpus — fp16 faiss index + text-only parquet

feat: sessions, autosave, display preferences

fix: move generate calls out of setState updaters

feat: streaming annotation generation and annotator panel

feat: reader screen ported with real data flow

fix: restore screen flex centering; skip button only when result ready

feat: upload and pipeline screens wired to real backend

feat: pure note-alignment layout math

feat: frontend session, SSE api client, modes

fix: raise ProviderError on OpenAI mid-stream error objects

feat: FastAPI endpoints — parse job SSE, annotate proxy with cache

feat: streaming multi-provider LLM router

feat: prompt builder and sqlite completion cache

fix: epsilon guard for zero-distance similarity conversion

fix: use arxiv_id column (not ads_id) for arXiv link in LitPaper

feat: corpus wrapper and per-paragraph retrieval

feat: keyword extraction and pathfinder keyword weighting

feat: paragraph segmentation and parse_pdf