docs: ship itd-3 (manual fidelity audit + planned to shipped)

.abcd/development/intents/planned/itd-3-modular-rules-loader.md → .abcd/development/intents/shipped/itd-3-modular-rules-loader.md

@@ -1,7 +1,7 @@
 ---
 id: itd-3
 slug: modular-rules-loader
-spec_id: null
+spec_id: spc-1
 kind: standalone
 suggested_kind: null
 reclassification_history: []
@@ -63,7 +63,54 @@ This is also the keystone for retiring `~/ABCDevelopment/.claude/` entirely. Wit
 
 ## Audit Notes
 
-_Empty. Populated by intent-fidelity-reviewer when intent moves to shipped/._
+> _Manual fidelity audit — the `intent-fidelity-reviewer` agent is not yet built,
+> so this audit was performed by hand. Judge: Claude Opus 4.8; date 2026-07-11;
+> reviewed against the merged loader (the rules-loader design + implementation).
+> Per-criterion verdicts per the itd-1 discipline, then a three-bucket prose
+> audit. `spec_id: spc-1` is reserved for itd-3; the future native spec store
+> adopts it rather than re-minting._
+
+### Per-criterion verdicts
+
+1. **MET.** COMMITTING's `recall`/`aliases` cover "commit", "PR", and "git add";
+   the prompt-router hook injects the domain on the turn.
+2. **MET (exceeded).** A no-match renders zero model-facing tokens — bettering the
+   `< 200` threshold. The "header only" mechanism was replaced by out-of-band
+   diagnostics (signed-off delta D3).
+3. **MET_WITH_CONCERNS.** Dedup is MET (inject on turn 1, skip 2–3). "Forced
+   refresh on prompt 5" was deliberately replaced by event-driven refresh on
+   `SessionStart(compact)` plus a fixed-N backstop (default 15, configurable) —
+   signed-off delta D1.
+4. **MET.** A `*ROADMAP` star-command injects the domain regardless of keyword
+   match (it overrides a dormant state, not the kill switch).
+5. **INCONCLUSIVE.** The marker block is delivered and sized (~40 lines) and ahoy
+   installs it, but the live "fresh clone + installed plugin + new session"
+   behaviour cannot be observed in-repo (the hook only fires in an installed
+   plugin).
+6. **NOT_MET.** The shipped defaults are a concise curated set (eight domains)
+   drawn from the working conventions plus the Pass-1 mapping in
+   `research/legacy-harvest.md`; a completeness-audited rule-by-rule port of the
+   legacy methodology file was not performed. This is the one genuine gap.
+
+**Rollup:** 3 MET · 1 MET_WITH_CONCERNS · 1 INCONCLUSIVE · 1 NOT_MET.
+
+### Three-bucket audit
+
+- **Honoured.** Just-in-time recall injection replacing the monolithic
+  force-load (the core promise); binary-bundled defaults plus a per-repo
+  override; per-domain dedup; star-command activation; the kill switch and
+  dormant state; the `abcd rules [domain]` diagnostic verb; the OPINIONS domain
+  pointing at the canonical conventions; a fail-closed, non-blocking trust
+  boundary; a marker block under 50 lines.
+- **Diverged (every item a maintainer-signed delta).** D3 — zero no-match tokens
+  plus out-of-band diagnostics instead of a `< 200`-token header. D1 —
+  event-driven refresh with a fixed-N backstop instead of forced-every-5. D2 —
+  the shipped `{schema_version, disabled, domains{}}` per-field-override shape
+  instead of the earlier `extends`/`overrides` sketch. D4 — no `.abcdignore`.
+- **Missing / outstanding.** A completeness-audited harvest of the legacy
+  methodology file into the defaults (criterion 6). Live installed-plugin session
+  verification (criterion 5). Neither blocks the shipped promise; both are
+  recorded here as follow-ups.
 
 ## References
 

.abcd/development/plans/2026-07-11-itd-3-rules-loader.md

@@ -5,7 +5,7 @@ any implementation. Reached via the [prefer-sota](../principles/prefer-sota.md)
 process: an adversary challenged generic just-in-time rule-injection SOTA (CARL,
 AGENTS.md, Cursor rules, Skills) for THIS repo's fit, then a `sota-researcher`
 verdict was taken on the fit-surviving hypothesis. This doc is the design of
-record for [itd-3](../intents/planned/itd-3-modular-rules-loader.md); it does not
+record for [itd-3](../intents/shipped/itd-3-modular-rules-loader.md); it does not
 authorise a build. Adopting a planned intent is a maintainer decision — the four
 deltas from the intent (§7) are the sign-off surface.
 
@@ -30,7 +30,7 @@ rules **point at** `.abcd/development/principles/` rather than copying them.
   `internal/core`; the capability lives in the core and renders through the
   vendor-neutral `abcd rules` verb.
 - **CARL's global `~/.carl/carl.json` source** — collides with per-repo-only
-  ([the intent's "What's Out of Scope"](../intents/planned/itd-3-modular-rules-loader.md);
+  ([the intent's "What's Out of Scope"](../intents/shipped/itd-3-modular-rules-loader.md);
   `~/.abcd/` holds only machine-wide state). No `~/.abcd/rules.json`. Ever.
 - **CARL's MCP server + `add_rule`/`toggle_domain` verbs, and the Python hook** —
   each is a new runtime/dependency (ask-first). Rule *mutation* is a file-edit of

.abcd/development/research/legacy-harvest.md

@@ -18,7 +18,7 @@ Decision interview conducted 2026-05-04 in conversation. Outcomes captured below
 
 ## Pass 1 — Methodology rules
 
-Source: `~/ABCDevelopment/.claude/CLAUDE.md`. Disposition: split into CARL-style domain rules per [`itd-3-modular-rules-loader`](../intents/planned/itd-3-modular-rules-loader.md), shipped as `scripts/abcd/defaults/rules.json` in the plugin.
+Source: `~/ABCDevelopment/.claude/CLAUDE.md`. Disposition: split into CARL-style domain rules per [`itd-3-modular-rules-loader`](../intents/shipped/itd-3-modular-rules-loader.md), shipped as `scripts/abcd/defaults/rules.json` in the plugin.
 
 | § | Section | Verdict | Domain mapping | Notes |
 |---|---|---|---|---|
@@ -208,7 +208,7 @@ abcdDev/
 }
 ```
 
-Merge semantics, conflict policy, and diagnostic surface (`abcd rules show`, `abcd rules diff`, `abcd rules lint`) defined in [`itd-3-modular-rules-loader`](../intents/planned/itd-3-modular-rules-loader.md).
+Merge semantics, conflict policy, and diagnostic surface (`abcd rules show`, `abcd rules diff`, `abcd rules lint`) defined in [`itd-3-modular-rules-loader`](../intents/shipped/itd-3-modular-rules-loader.md).
 
 ## Migration sequence
 
@@ -277,7 +277,7 @@ If the first phase starts feeling too heavy, the candidates for moving to a late
 ## References
 
 [itd-1]: ../intents/disciplines/itd-1-acceptance-gates.md "itd-1 — Acceptance gates"
-[itd-3]: ../intents/planned/itd-3-modular-rules-loader.md "itd-3 — Modular rules loader"
+[itd-3]: ../intents/shipped/itd-3-modular-rules-loader.md "itd-3 — Modular rules loader"
 [itd-7]: ../intents/drafts/itd-7-rp-workspace-portability.md "itd-7 — RP workspace portability"
 [itd-17]: ../intents/drafts/itd-17-model-effectiveness-tracking.md "itd-17 — Model effectiveness tracking (a later phase)"
 [itd-23]: ../intents/drafts/itd-23-spec-kit-interop.md "itd-23 — Spec Kit interop (a later phase)"

.abcd/development/research/related-work.md

@@ -57,7 +57,7 @@ Architecture in one paragraph:
 
 ### Where abcd overlaps
 
-abcd's 978-line `~/ABCDevelopment/.claude/CLAUDE.md` is exactly the failure mode CARL addresses. Most sessions need none of those rules; every session pays for all of them. The plugin-bundled defaults + per-repo `.abcd/rules.json` model proposed in [`itd-3-modular-rules-loader`](../intents/planned/itd-3-modular-rules-loader.md) is structurally identical to CARL.
+abcd's 978-line `~/ABCDevelopment/.claude/CLAUDE.md` is exactly the failure mode CARL addresses. Most sessions need none of those rules; every session pays for all of them. The plugin-bundled defaults + per-repo `.abcd/rules.json` model proposed in [`itd-3-modular-rules-loader`](../intents/shipped/itd-3-modular-rules-loader.md) is structurally identical to CARL.
 
 ### Why not depend on CARL
 

.abcd/work/DECISIONS.md

@@ -184,3 +184,11 @@ parallel-agent merge contention bites.
   no-match + out-of-band diagnostic log (supersedes the "<200-token header"
   acceptance criterion); **D4** `.abcdignore` rejected for v1. Build proceeds
   phased/TDD from Phase 1 (`internal/core/rules`).
+- 2026-07-11 — itd-3 **shipped manually** ahead of the intent-lifecycle pipeline.
+  Moved `planned/ → shipped/` by hand with `spec_id: spc-1` (reserved — the future
+  native spec store adopts spc-1 for itd-3, never re-mints it) and a hand-authored
+  `## Audit Notes` (the `intent-fidelity-reviewer` agent does not exist yet; judge
+  = Claude Opus 4.8). Rollup 3 MET / 1 MET_WITH_CONCERNS / 1 INCONCLUSIVE / 1
+  NOT_MET; every divergence is a signed-off D1–D4 delta, the one gap is the AC6
+  legacy-harvest completeness. Inbound links repointed planned→shipped by hand —
+  the link-drift-on-move the future reconcile pass automates.