SoundMindsAI · SoundMindsAI · May 24, 2026 · May 24, 2026
@@ -6,7 +6,7 @@ _Top-level index across MVP1 → GA v1+ as of **2026-05-24**. Click a release na
 
 | Release | Theme | Progress | Status |
 |---|---|---|---|
-| [MVP1 / v0.1](MVP1_DASHBOARD.md) | The Loop | 74 / 75 scoped done · 11 remaining | **In progress** |
+| [MVP1 / v0.1](MVP1_DASHBOARD.md) | The Loop | 75 / 75 scoped done · 12 remaining | **In progress** |
 | [MVP1.5 / v0.1.5](MVP1_5_DASHBOARD.md) | Real Signals | 1 item(s) queued | **Held / queued** |
 | [MVP2 / v0.2](MVP2_DASHBOARD.md) | Observable | 1 / 1 scoped done · 1 remaining | **In progress** |
 | MVP3 / v0.3 | Production Stacks | — | **Not yet scoped** |

@@ -384,7 +384,7 @@ <h2>Releases</h2>
 <div class="roadmap-row">
   <div class="release-name"><a href="mvp1_dashboard.html">MVP1 / v0.1</a></div>
   <div class="theme">The Loop</div>
-  <div class="progress">74 / 75 scoped done · 11 remaining</div>
+  <div class="progress">75 / 75 scoped done · 12 remaining</div>
   <span class="state-pill in_progress">In progress</span>
 </div>
 

@@ -8,15 +8,15 @@
 - [`docs/00_overview/implemented_features/2026_05_24_feat_digest_executable_followups/feature_spec.md`](../../../00_overview/implemented_features/2026_05_24_feat_digest_executable_followups/feature_spec.md) — Tier-A substrate this spec extends
 - [`docs/01_architecture/llm-orchestration.md`](../../../01_architecture/llm-orchestration.md)
 - [`docs/01_architecture/data-model.md`](../../../01_architecture/data-model.md)
-- Sibling (in-flight backlog): [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md) — Tier C `edit_template`
+- Sibling (in-flight backlog): [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md) — Tier C `edit_template`
 
 ---
 
 ## 1) Purpose
 
 - **Problem:** Tier A (shipped 2026-05-24 as PR #225) lets the LLM suggest `narrow` / `widen` / `text` followups within the **same query template**. But the LLM sometimes recognizes that a **different template entirely** is the better fit — e.g., parameter-importance is highly skewed (some declared params are dead weight), or winning trials cluster around a sub-set of params that map cleanly onto a different template's `declared_params`. Today the operator has to notice this themselves; the LLM has no structured way to say "try template X instead." The "Run this followup" substrate (`backend/app/domain/study/followups.py`, `ui/src/components/proposals/suggested-followups-panel.tsx`, the `?action=run_followup` modal prefill at `ui/src/app/proposals/[id]/page.tsx:120-184`) is in place — only the `swap_template` variant + its UI surface is missing.
 - **Outcome:** The LLM emits a fourth `kind: "swap_template"` variant carrying `{rationale, template_id, search_space}` where `template_id` references a different `query_templates.id` than the parent study used. The proposal-detail UI renders the variant as an actionable card with a side-by-side `declared_params` comparison (parent template vs proposed swap target) before the operator commits. The "Run this followup" button pre-fills `template_id = <swap_target>` (not the parent's template) plus the LLM-proposed `search_space`, with disjoint params filled from the existing heuristic at `backend/app/domain/study/search_space_defaults.py`. Lineage (`studies.parent_proposal_id` + `parent_proposal_followup_index`) is reused unchanged — the cross-template hop is explicit in the data because the child study's `template_id` differs from the parent's.
-- **Non-goal:** Auto-running swap-template followups without operator click (already covered for the deterministic narrow-around-winner case by `feat_auto_followup_studies`; cross-template swaps are a much larger trust surface and explicitly stay operator-mediated). LLM-driven template **edits** (Tier C — different surface, tracked at sibling [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md)). Side-by-side rendering of the **query body** itself (Jinja2 source) — out, only `declared_params` are compared. Auto-discovery of the swap-target template by the worker (the LLM picks; we don't fall back to a similarity search).
+- **Non-goal:** Auto-running swap-template followups without operator click (already covered for the deterministic narrow-around-winner case by `feat_auto_followup_studies`; cross-template swaps are a much larger trust surface and explicitly stay operator-mediated). LLM-driven template **edits** (Tier C — different surface, tracked at sibling [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md)). Side-by-side rendering of the **query body** itself (Jinja2 source) — out, only `declared_params` are compared. Auto-discovery of the swap-target template by the worker (the LLM picks; we don't fall back to a similarity search).
 
 ## 2) Current state audit
 
@@ -93,7 +93,7 @@
 
 ### Out of scope
 
-- **Tier C — `kind: "edit_template"` followups.** Operator-only today; LLM-suggested template edits are a much larger trust/validation surface and unrelated to this spec's lane. Tracked at sibling backlog folder [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md).
+- **Tier C — `kind: "edit_template"` followups.** Operator-only today; LLM-suggested template edits are a much larger trust/validation surface and unrelated to this spec's lane. Tracked at sibling backlog folder [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md).
 - **Auto-running swap-template followups without operator click.** Out — operator review is the entire trust mechanism for cross-template hops.
 - **Side-by-side rendering of the template's Jinja2 body.** Out — only `declared_params` are compared. The Jinja source is large, hard to diff usefully without a syntax-aware viewer, and most operators making the call don't need it; if they do, the existing template detail page at `/templates/[id]` is one click away.
 - **Auto-discovery of the swap-target template.** The LLM picks; we don't fall back to a similarity search or compute the swap target server-side. (Reason: the LLM has the full study-outcome context including parameter-importance distribution + winning-trial cluster; a deterministic similarity search would have to re-derive a much weaker proxy for "which template fits these winning params better.")
@@ -713,7 +713,7 @@ Tooltip placement uses the existing `<InfoTooltip glossaryKey="...">` primitive
 - [ ] Documentation updates across docs/01–05 are merged (§15).
 - [ ] Rollout gates from §16 are satisfied.
 - [ ] Cross-model review (GPT-5.5) on this spec and the forthcoming implementation plan completed and adjudicated.
-- [x] Deferred-phase tracking: N/A (single-phase delivery). Tier C `edit_template` is tracked at sibling [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md).
+- [x] Deferred-phase tracking: N/A (single-phase delivery). Tier C `edit_template` is tracked at sibling [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md).
 - [ ] No open questions remain in §19.
 
 ## 19) Open questions and decision log

@@ -48,4 +48,4 @@ Phase 1 of `feat_digest_executable_followups` handles `narrow` / `widen` / `text
 - **Builds on [`feat_digest_executable_followups`](../../../00_overview/implemented_features/2026_05_24_feat_digest_executable_followups/idea.md) Phase 1 substrate** — discriminated-union schema, JSONB column, lineage columns, and "Run this followup" UI scaffolding all already landed.
 - **Reuses [`backend/app/domain/study/search_space_defaults.py`](../../../../backend/app/domain/study/search_space_defaults.py)** from `feat_agent_propose_search_space` (shipped 2026-05-21) for the disjoint-set heuristic bounds.
 - **Reuses `feat_create_study_search_space_builder` row primitives** (shipped 2026-05-20) for the cross-template comparison (when feasible).
-- **Adjacent backlog item:** [`../backlog_feat_digest_template_edit_followups/idea.md`](../backlog_feat_digest_template_edit_followups/idea.md) — the Tier C `edit_template` extension, prefixed `backlog_` because its template-editor UI prerequisite doesn't exist. Promotes out of `backlog_` once this feature ships AND the editor lands.
+- **Adjacent backlog item:** [`../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md) — the Tier C `edit_template` extension, prefixed `backlog_` because its template-editor UI prerequisite doesn't exist. Promotes out of `backlog_` once this feature ships AND the editor lands.
@@ -19,7 +19,7 @@
 - Tier-A patterns are the structural template — story shapes (Domain → Worker/Prompts → API → Frontend → E2E), test-layer choice, and DoD style mirror the shipped Tier-A plan one-to-one.
 - Fail-loud tests: assert explicit status, shape, error codes, and structlog reason codes.
 - Keep increments narrow enough to verify independently — domain helper → discriminated-union widening → LLM schema/prompts → worker remap → API response widening → frontend card + prefill → E2E.
-- **Single-phase delivery.** No deferred phases — Tier C (`edit_template`) lives at sibling [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md) and is not gated by this work.
+- **Single-phase delivery.** No deferred phases — Tier C (`edit_template`) lives at sibling [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md) and is not gated by this work.
 - **No new migration.** Tier-A's JSONB column + lineage columns + CHECK constraint + BEFORE DELETE trigger apply unchanged (spec §3, FR-13).
 
 ## 1) Scope traceability (FR → epics/stories)
@@ -51,7 +51,7 @@
 
 **Spec error-code coverage vs plan:** Spec §8.5 introduces **zero** new error codes. Worker-side validation failures downgrade in-band (no API error); `POST /api/v1/studies` flow uses existing Tier-A codes (`PROPOSAL_NOT_FOUND`, `DIGEST_NOT_FOUND`, `FOLLOWUP_INDEX_OUT_OF_RANGE`, `TEMPLATE_NOT_FOUND`, `INVALID_SEARCH_SPACE`, etc.) verbatim. Match.
 
-**Deferred phases verified:** N/A — single-phase delivery per spec §3 "Phase boundaries". Tier C (`edit_template`) lives at sibling [`backlog_feat_digest_template_edit_followups`](../backlog_feat_digest_template_edit_followups/idea.md) folder and is explicitly NOT gated by this work.
+**Deferred phases verified:** N/A — single-phase delivery per spec §3 "Phase boundaries". Tier C (`edit_template`) lives at sibling [`backlog_feat_digest_template_edit_followups`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md) folder and is explicitly NOT gated by this work.
 
 ## 2) Delivery structure
 

@@ -14,7 +14,7 @@
   - Cycle 2: 5 findings (5 accepted, 0 rejected) — 3 re-raises (stale §2/§3 prose on optional schema + deterministic worker pre-clean rule; §13/§4 diagnostic field-name drift; §6 intro sentence still narrow) and 2 net-new (4th reason code `remap_invalid_search_space` for FR-7 step 3 emission; `validation_error` truncation matches the canonical `_truncate` helper)
   - Cycle 3: 1 finding (1 accepted, 0 rejected) — net-new internal-consistency catch: empty trusted intersection is unreachable on the worker path (Pydantic min_length=1 rejects empty `SearchSpace`), so helper rejects no-trusted-intersection inputs and prompt instructs LLM to skip in that case; disjoint-only swaps explicitly out of contract
   - Total: 18 accepted, 0 rejected across 18 findings (Decision Log D-17 through D-34 enumerate the resolutions)
-- Phases: 1 total (single-phase delivery — no `phase2_idea.md`; Tier C `edit_template` tracked at sibling [`../backlog_feat_digest_template_edit_followups/`](../backlog_feat_digest_template_edit_followups/idea.md))
+- Phases: 1 total (single-phase delivery — no `phase2_idea.md`; Tier C `edit_template` tracked at sibling [`../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/`](../../../02_product/planned_features/backlog_feat_digest_template_edit_followups/idea.md))
 
 ## Plan
 - Status: Approved
@@ -28,7 +28,9 @@
 - Phases covered: single-phase delivery (Tier B only)
 
 ## Implementation
-- Status: Not started
-
-## Implementation
-- Status: Not started
+- Status: Complete — admin-merged into main as PR #232 squash `791642e0` on 2026-05-24.
+- Branch: `feature/digest-executable-followups-swap-template` (deleted post-merge).
+- PR: [#232](https://github.com/SoundMindsAI/relyloop/pull/232) — admin-merged with smoke gate red. The smoke failure was a compound cascade of 5+ pre-existing regressions from PR #188 + PR #228's admin-merge bypasses (NOT introduced by Tier B code): cleared `OPENAI_API_KEY_TEST` repo secret; missing `scripts/` COPY in Dockerfile (broke api container startup); `_wait_healthy` not gating on capability check; missing `make seed-demo` step in smoke workflow; OpenAI key rejection by capability check (root unclear). Tier B's own code is clean (3 GPT-5.5 spec cycles + 2 plan cycles + Gemini accept + final-review pass with 6 of 7 findings rejected with cited counter-evidence + 1 deferred). 5 fixes applied during the smoke cascade are bundled into this same squash; remaining issues captured as separate `bug_*` ideas (OpenAI capability + ES cluster unreachability).
+- Cross-model review: spec 3 cycles 18/18 accepted; plan 2 cycles 7 accepted + 4 rejected; Gemini 1 Medium accepted; final GPT-5.5 1 deferred + 2 rejected with counter-evidence + 4 spurious from diff-window truncation.
+- Test deltas: backend unit 1331 → 1346 (+15 — 7 template_swap + 6 followup union + 1 backcompat + 7 worker validation overlap accounted); +3 integration; +3 contract; +20 vitest; +1 Playwright E2E (gated on demo-data seed which is part of the cascade).
+- **No new migration** — Tier A's `0019_digests_suggested_followups_jsonb` + lineage columns apply unchanged. Alembic head stays at `0019`.