Move ADRs from README into Atlas/References/ADR/, adopt MADR format

Context and Problem Statement

By April 2026 the vault README.md had grown to ~4400 words, of which ~3000 were the ADR log. The “single growing list” approach from 0003-adrs-as-single-growing-entry-list reached its threshold: the file conflated current spec (the durable rules) with historical decisions (chronological log), making it hard to scan and bloating context for both human and agent consumers.

Considered Options

• Move ADRs to per-file MADR records under Atlas/References/ADR/NNNN-title-in-dashes.md. README keeps Zen + Spec + a brief Decisions section linking the ADR folder.
• Keep ADRs in README.md (status quo).
• Move ADRs to a separate decisions.md at vault root (revert 0015-merge-decisions-into-readme).
• Custom ADR format.

Decision Outcome

Chosen option: “Per-file MADR records under Atlas/References/ADR/”, because (a) separates current spec from decision history (different lifecycles, different audiences), (b) MADR is an established standard with clear section structure, (c) per-file granularity enables wikilink references between ADRs and from elsewhere in the vault, (d) Atlas/References/ is the established home for own-authored reference material.

Filename convention: NNNN-title-in-dashes.md (zero-padded 4 digits, e.g. 0017-atomic-note-convention-overhaul.md). Frontmatter carries date, title, status, and tags. Body follows MADR section structure (Context and Problem Statement / Considered Options / Decision Outcome / optional Consequences and Pros and Cons).

Template lives at x/templates/adr-template.md, deployed via Nix and wired into Templater for Atlas/References/ADR/ folder creation.

Consequences

• Good, because README shrinks from ~4400 to ~1400 words (estimated). Spec-vs-history separation is clean.
• Good, because MADR standardizes section structure across decisions.
• Good, because per-file ADRs enable wikilinks between decisions (0008-kill-archive-folder referenced from 0013-archive-triaged-and-removed, etc.).
• Bad, because vault file count grows by ~17 ADR files plus the template.
• Bad, because external references to [[README#adr-nnn|ADR-NNN]] anchors break — must be updated to [[NNNN-title-in-dashes]].

Supersedes the location aspect of 0002-combine-zen-spec-adrs-in-readme, 0003-adrs-as-single-growing-entry-list, and 0015-merge-decisions-into-readme. The original spirit (decisions are tracked, single source of truth) is preserved — only the location of the ADR collection moves.

Pros and Cons of the Options

Per-file MADR under Atlas/References/ADR/

• Good, because separates current spec from decision history.
• Good, because uses an established standard (MADR).
• Good, because per-file granularity supports wikilink referencing.
• Bad, because vault file count grows.

Keep ADRs in README

• Bad, because file bloated to ~4400 words.
• Bad, because conflates spec with history (different lifecycles).

Separate decisions.md at vault root

• Bad, because 0015-merge-decisions-into-readme showed two root-level files have discoverability cost.
• Bad, because doesn’t address the per-ADR-granularity concern.

Custom ADR format

• Bad, because reinventing the wheel; MADR exists and works.