Markdown for LLMs
docassemble
The source Markdown for this article. Copy it into your assistant or download it as a text file.
# docassemble **In short:** docassemble asks the questions and builds the document; Arxo decides the case with grounds. On one charities interview they meet in observable traces: how many questions to an answer, what the file contains, and what happens at an evaluative term. This page is for access- to-justice builders who interview users into legal outcomes. ## What docassemble is for docassemble is a guided-interview and document-assembly platform: dependency-driven questioning that orders itself, legal logic as plain author Python, and three assembly paths from templated documents to filled forms. Its strengths are zero distance from question to rule, mobile-ready interviews out of the box with versioned answer snapshots and review screens, legally significant output files, mature deployment, and an interview style lawyers write directly without a separate cards layer — backed by a community with external test tooling. ## Where it meets Arxo Fact-gathering through interviews plus rule-driven assembly on one fixed task: question count to answer, redundant questions, byte-reproducibility of the assembled file, and handling of the evaluative feature — compute, request judgment, or stop. The Arxo side runs its charities package with question cards; the interview side is an independently written charity-test flow; the pinned charities act arbitrates. The engine and its rules are, in the docs' own words, effectively one thing in one place — so the comparison reads observable traces, never internals head-on. ## Key differences - **Author patterns are not platform properties.** Plain-Python legal logic means every interview's behavior is the author's as much as the platform's; the page attributes each claim to its side. Silence as a boundary on one side versus a coded answer on the other is a purpose difference, not a win. - **Evaluative terms go to a human by branching.** Arxo returns requires-judgment with a named request, stays undecided, and names missing facts; the interview asks a person and branches, with no first-class judgment or refusal outcomes and no derivation trace — the nearest equivalent is author-written screen text. - **Time, editions, and numbers are author code.** Python dates and author-computed amounts carry no calendrical or editorial norm semantics; the platform has no edition awareness — change means manual re-read, interview edit, and version bump. Day counts and risk levels from the studied bank are non-comparable by contract. - **Compare fields, not bytes across formats.** Reproducibility means exact hashes across two runs of the same side; question metrics are exact integers with no tolerance. String heuristics for legal connection (a place name in an address) are not legal connection under the studied articles — such cases stay non-comparable with grounds. ## A concrete scenario Twelve interview cases — a full education purpose, empty purposes, an analogous purpose without the deciding view, a benefit statement, control structures with exemption orders, the jurisdiction link, governor convictions including spent-but-reportable ones, a skipped answer, plus day-count and risk-level refusal controls — with an edit axis across two text revisions: redefined officers, a new corporate form, re-routed oversight, and a stability control. Interview fixtures and container digests are created at run time, not before. ## Choosing and combining Choose docassemble when the front door is an interview: users answer questions and receive a document, on mobile, tomorrow. Look to Arxo when the decision behind the document must cite its edition, keep conflicts visible, and refuse cleanly on thin facts. Combined, the interview gathers and assembles while Arxo decides the hard cases — with the question log and the assembled fields as the shared surface. ## Evidence and open questions - Sources checked: September 2026 (docs index, changelog, interview-logic and assembly guides, the full license text, the external tester; one official source set was partially blocked, and HTML plus flat extractions were used instead — marked as such). - Studied profile: docassemble at its September release with a fixed stand runtime. - Basis: confirmed by documentation plus a prepared protocol with a frozen neutral case bank (no interview-side answers pre-set); comparative run not performed. - Open: transitional provisions outside the studied slice, interview fixtures at run time, and the run itself. ## Sources and reproducible materials - Shared scenario: [One charities text, five systems](/comparisons/jersey-charities/). - Documentation: [docassemble.org/docs.html](https://docassemble.org/docs.html) - Interview logic: [docassemble documentation](https://docassemble.org/) - External tester: [github.com/SuffolkLITLab/alkiln](https://github.com/suffolklitlab/alkiln)