[adamjohnwright]

Spec 002 (analysis report), T027: the second of the two points T019 left open (research R9), decided on 14 Nov 2026: the analysis report prints a book's and a web page's references the way today's report does.

Today's form, read from its source. I read ApaStyle in reactome/analysis-report's 1.4.0-SNAPSHOT build from its compiled classes. Nothing here shows that this build is the one deployed. What it writes:

• a web page: Title. Retrieved from address, with no authors and no year;
• a book: Authors (year)., then Chapter, Book in italics, then , pages; it prints no publisher and no ISBN;
• an article: the same as spec 001's form, apart from how authors are written, which the comparison does not read.

A correction to R9: T019 noted "a book's publisher after a comma". The text after the comma is actually the book's title, which follows the chapter title. Today's report prints no publisher. Spec 001's form does print the publisher. On R-HSA-1280215 in release 98 (checked against a private copy of the graph), the report now prints (2002). Cytokines and the immune system, Cytokine Reference. and <title>. Retrieved from <address> for that pathway's two web pages.

Changes

• The report and spec 001 share reference() through sectionHtml. So the report gets its own form, reportReference (src/report/layout.ts), which it passes to sectionHtml. Spec 001's book is unchanged; referenceLead and referenceTitle are split out of reference() without changing what it writes. Follow-up, not done: whether spec 001's book should switch to this form is still open.
• The address and the publisher are read by spec 001's literature query as it stands; no field is added.
• The compare-report reader now drops pdf-export's list number (3. ) from a reference that has no year. In today's form a web page has no year, so the number used to make every web page differ.
• Spec Assumptions ("One implementation"), research R9 (its row, plus a note on today's form), and tasks.md T027 updated.

Tests (each seen failing without its change)

• test/report/body.test.ts: the fixture graph's book (which has a publisher) and its web page, read by literature, print in the report's form. Spec 001's form of the same two is asserted unchanged.
• test/compare/report.test.ts: a book and a web page printed in today's form and in the report's form agree in compare-report. The today's-form helper now writes both kinds as ApaStyle does. This test also fails without the reader's list-number fix.

All 715 tests pass locally against a freshly loaded fixture graph; types, lint and format pass as well. Locally the licence check failed because of a symlinked node_modules in my worktree; the same lock file passes it (0 offending).

This PR and #217 both add a task to tasks.md just before T021. Whichever merges second will need a rebase.

🤖 Generated with Claude Code