Record-line codes carry their titles, and lists get lines

[dmarx]

SOTA-136, staged and then rendered through real Quartz:

Status	Proposed
Filed	2026-09-05
Consensus	contested
Source	LIT-140 — mHC: Manifold-Constrained Hyper-Connections
Extends	SOTA-137 — Widen the residual stream into several streams with freely learned mixing (hyper-connections)
Compared against	SOTA-133 — Replace fixed residual accumulation with learned attention over preceding layers
Contested by	LIT-151 — oHC: Orthogonal Hyper-Connections on SO(4) via Quaternions LIT-181 — Beyond the Birkhoff Polytope: Spectral-Sphere-Constrained Hyper-Connections
Supersedes	SOTA-137 — Widen the residual stream into several streams with freely learned mixing (hyper-connections)

A code alone asks the reader to already know the record — LIT-141 says nothing about what it is, and being followed by someone who doesn't yet know where it goes is the whole point of a backlink. That makes the items long enough that the center dot stops working, so several values become a real bulleted list, one per line. A single value stays plain: a one-item bullet is a bullet about nothing.

Bullets in a table cell need HTML, so I checked rather than guessed

A markdown table cell can't hold a block-level list, so this is <ul><li> — which only works if the renderer passes raw HTML through, and guessing would ship a page full of visible angle brackets.

Cloned Quartz v4.5.2 (what actions/site pins) and built a probe. Raw HTML passes through, and the part that actually mattered is that markdown inside the <li> is still parsed:

<td><ul><li><a href="../literature.d/LIT-151" class="internal alias"
    data-slug="record/literature.d/LIT-151">LIT-151</a> — oHC…</li></ul></td>

Then the whole anthology through the same build — 410 files, 1217 pages emitted, bullets are a real <ul>, links inside them come out class="internal alias", and zero pages contain an unexpanded [[. That last check is what a fixture can't give you.

The title that reads as syntax

Splicing titles in broke luria site on this repository:

record/decisions.d/ADR-005.md → unresolved in frontmatter: [[CODE]]
record/decisions.d/ADR-024.md → unresolved in frontmatter: [[CODE]]

ADR-025 is titled Wikilinks: `[[CODE]]` is a typed reference, and typing it changes the rules. Both those decisions carry influenced_by: ADR-025, so both record lines now contained a literal [[CODE]], and the expander went looking for a document called CODE.

A title is data being spliced into two surfaces that read syntax — a table cell where | ends the cell, and a string about to be wikilink-expanded where [[ opens a reference. Both are escaped now; markdown renders \[ as [, so it costs the reader nothing.

Worth noting: the backticks around it in the title are not protection — the expander doesn't mask code spans, consistent with ref_status.scan being deliberately unmasked.

What caught it was test_links_out_of_the_site_go_to_the_repository, which asserts report.unplaced == [] over the whole real record. A fixture with tidy titles passes either way. There's now also a test naming ADR-025 specifically, so the escaping can't be quietly removed by someone who doesn't know why it's there.

Shape

Rows changed from (label, str) to (label, list[str]) — that's what lets one renderer choose between a plain cell and a list without parsing its own output back. Titles are read once per stage() through ref_status.load_docs(), the reader that already answers "what documents are there and what are they called", rather than a second one that could disagree.

Checks

python -m pytest tests -q → 926 passed. luria link --fix, luria index, luria lint → clean. Plus the real Quartz build above.

🤖 Generated with Claude Code

https://claude.ai/code/session_01YP4P3m8rzVFb8idnTE4FfT

---

Generated by Claude Code