Design science research
Document Lens is the artefact of a design science research project. The contribution is not a domain finding but a way of building analysis instruments: a configurable lens, polarity in one keyword list, and a scoring rule the researcher supplies rather than the software hard-codes.
Manuscript under review · artefact available now
Researchers asking whether an organisation talks about a topic rarely need only a mention count. They need to know whether the topic appears in the right place and whether the language delivers or merely performs. Done by hand, that reading is careful but slow and hard to reproduce; done with a bespoke script, it is fast but leaves no auditable record of how each decision was made.
Document Lens was built to hold both halves at once: deterministic, reproducible measures over the researcher's own corpus, with every finding linked back to the passage that produced it. Its first use case was sustainability reporting in Australian university annual reports; the framework — a lens — is what generalises.
The instrument is open source under the MIT licence. It runs locally: documents, extracted text, and every computed signal live in a SQLite database on the researcher's machine, with no account and no upload. The analysis engine ships with the application. A project can be exported as a portable bundle and its tables exported for independent checking.
The paper describes a pinned version of the software, because every claim about its behaviour depends on what it actually did when the numbers were computed. Cite the release below rather than a branch.
Analysis axes, polarised keywords, and a scoring rule are declared as data. Changing the lens changes the domain, without changing the software.
A set of principles, each stated with the evidence that would refute it — so a reader can judge the design record rather than take it on trust.
Research corpora are often confidential or embargoed. Keeping files and analysis on the device is a design commitment, not a deployment detail.
Three systems have done most to automate disclosure analysis, and Document Lens is usefully understood against them. We state the differences as design choices, not deficiencies: each comparator is well matched to the job it was built for.
An LLM-based tool that analyses corporate sustainability reports against the TCFD's eleven recommendations and produces a 0–100 conformity score. It grounds answers in retrieved passages with numbered source spans and refines its prompts from expert feedback. The framework is built into the software, and the analysis runs on a hosted model.
Ni et al. (2023), EMNLP System Demonstrations · doi:10.18653/v1/2023.emnlp-demo.3
A library that parses corporate sustainability reports into structural units — pages, blocks, sentences — and annotates them with climate-domain models, including environmental-claim detection. The framework is a choice among annotators the analyst loads; the output is extraction rather than a score.
Morio et al. (2024), IJCAI Demonstrations Track · doi:10.24963/ijcai.2024/1024
A hosted, searchable collection of climate laws, policies, and corporate disclosures — over 12,000 documents — with concept tags produced by classifiers and evaluated against expert-labelled passage samples that deliberately include ambiguous boundary cases. Its concept store is curated by the provider, and it serves discovery across a large public corpus.
Climate Policy Radar · climatepolicyradar.org · methodology
Document Lens differs in where the framework lives (data the researcher supplies, not a feature of the software), where the analysis runs (locally and deterministically, with optional model layers clearly flagged), and what happens to human codes (they are inputs the instrument can count and track, not only outputs to review). It scores a researcher's own corpus against a framework the researcher supplies; the comparators above either fix the framework or serve a public collection.
If you use the software, cite the pinned release. A citation for the paper will be added here once it clears peer review.
@software{borck2026documentlens,
author = {Borck, Michael},
title = {Document Lens},
version = {0.33.0},
date = {2026-10-31},
doi = {10.5281/zenodo.23864321},
url = {https://documentlens.borck.education},
license = {MIT},
note = {Local-first instrument for lens-based document analysis}
}