api.anthropic.com) does the reading. It extracts the comparison fields, assessment drafts and lineage mapping, triages reference lists and drafts chapter text from the bank only. Attaching a PDF lets Claude cite real figure numbers and captions. An abstract alone cannot support “smoking-gun” evidence. A typical extraction costs a few cents with a Sonnet-class model. PDF extractions cost more.Operating wavelength (log scale) against year, for papers with a defined band. Dot size scales with citations (OpenAlex). Colour shows the tunability level, following your editorial rule 3. The shaded band is your HORIBA iHR320 measurement window: Si 0.4–0.9 µm, InGaAs 0.95–1.7 µm.
Each edge is a verified citation: the later paper's OpenAlex reference list contains the earlier one. Lanes group papers by role. Hover a node to light up its links, and click to open it. Papers that share no edges are the ones to expand with the horizontal reference builder.
Seven chapters. Papers sit under the sections they serve, and one paper can serve several. Click a name to open the paper.
The field is shown as chains of questions rather than a list of papers. Each node records what was known, what the linking papers established, and the question left over. Statuses are draft judgements to revise. The chains are illustrative, as your brief notes, and do not claim the questions arose in exactly this order. For every arrow, ask: did they resolve it, or only propose it? Which limitation drove the next step? Was it reproduced or challenged?
One row per paper. Measured and simulated values are kept in separate columns (editorial rule 5). Click a header to sort. The six assessment scores (0–2) stay separate and are never summed. Values in italics with “verify” come from the curator, not the abstract. Check them against the PDF.
Select a paper from the map, chapters, lineage or matrix.
Claude drafts only from the records in this bank. It cites them as [Author Year] and flags anything the bank cannot support. Use it for essay outlines, evidence-led paragraphs, gap analyses, or a question you want answered across papers. To work in your own Claude chat or project instead, export the LLM context pack: a compact, citable digest of the bank.