SoundMindsAI · SoundMindsAI · May 29, 2026 · May 29, 2026 · gemini-code-assist · May 29, 2026
@@ -6,7 +6,7 @@ _Top-level index across MVP1 → GA v1+ as of **2026-05-29**. Click a release na
 
 | Release | Theme | Progress | Status |
 |---|---|---|---|
-| [MVP1 / v0.1](MVP1_DASHBOARD.md) | The Loop | 94 / 95 scoped done · 4 remaining | **In progress** |
+| [MVP1 / v0.1](MVP1_DASHBOARD.md) | The Loop | 94 / 95 scoped done · 3 remaining | **In progress** |
 | [MVP2 / v0.2](MVP2_DASHBOARD.md) | Three-Engine + Real Signals | 1 / 1 scoped done · 6 remaining | **In progress** |
 | MVP3 / v0.3 | Observable | — | **Not yet scoped** |
 | [GA v1 / v1.0](GA_DASHBOARD.md) | Production-ready | 1 item(s) queued | **Held / queued** |

@@ -21,19 +21,19 @@ Implementation in progress — resume to finish
 | Metric | Value |
 |---|---|
 | Scoped items done | **94 / 95** (99%) — feat_/infra_/chore_/epic_ past idea stage |
-| Pending work | **4** items (every not-done feat/infra/chore/bug across all priorities) |
+| Pending work | **3** items (every not-done feat/infra/chore/bug across all priorities) |
 | → P0 — do next | **0** unblocking / paying daily cost |
 | → P1 | **0** high-value, ready when P0 clears |
-| → P2 (default) | 3 important to file, not blocking |
+| → P2 (default) | 2 important to file, not blocking |
 | → Backlog | 1 captured for record, not planned |
-| Open bugs | 2 |
-| Legacy "Path to MVP1" | 4 items — scoped-not-done + bugs + chore-ideas only (excludes feat/infra ideas) |
+| Open bugs | 1 |
+| Legacy "Path to MVP1" | 3 items — scoped-not-done + bugs + chore-ideas only (excludes feat/infra ideas) |
 | Backlog ideas | 0 idea-only feat/infra (not yet scoped into MVP1) |
 | In flight | 1 feature(s) actively shipping |
 
 ## Pipeline
 
-### Done (127)
+### Done (128)
 
 | Feature | Type | One-liner | Depends on | Status |
 |---|---|---|---|---|
@@ -133,6 +133,7 @@ Implementation in progress — resume to finish
 | [chore_tutorial_polish](implemented_features/2026_05_12_chore_tutorial_polish/feature_spec.md) | Chore | The release tag `v0.1.0` is pushed with: a worked tutorial at `docs/08_guides/tutorial-first-study.md`, sample data (50-query set + sample ES index of ~1,000 docs from the Amazon ESCI subset), README  | `feat_chat_agent` `feat_digest_proposal` `feat_github_pr_worker` `feat_github_webhook` `feat_llm_judgments` `feat_proposals_ui` `feat_studies_ui` `feat_study_lifecycle` `infra_adapter_elastic` `infra_foundation` `infra_optuna_eval` | [PR #64](https://github.com/SoundMindsAI/relyloop/pull/64) merged 2026-05-12 |
 | [bug_auto_followup_completed_parent_stop_chain_race](implemented_features/2026_05_26_bug_auto_followup_completed_parent_stop_chain_race/idea.md) | Bug | The cycle-3 C3-1 cascade-cancel design tolerates terminal parents (cascade traverses through `completed` intermediates to reach in-flight descendants). But the FR-1 digest trigger fires `enqueue_follo | — | Complete |
 | [bug_capability_check_test_isolation](implemented_features/2026_05_12_bug_capability_check_test_isolation/idea.md) | Bug | Complete | — | Complete |
+| [bug_ceiling_badge_assumes_maximize_direction](implemented_features/2026_05_29_bug_ceiling_badge_assumes_maximize_direction/idea.md) | Bug | The `CEILING` badge in [`studies-table.column-config.tsx:METRIC_CEILING_THRESHOLD`](../ui/src/components/studies/studies-table.column-config.tsx) flags rows where `best_metric >= 0.99`. The threshold  | — | Complete |
 | [bug_clone_e2e_seed_template_params_mismatch](implemented_features/2026_05_26_bug_clone_e2e_seed_template_params_mismatch/idea.md) | Bug | Two helpers produce inconsistent fixture state: | — | Complete |
 | [bug_contract_test_stub_missing_target_filter_kwarg](implemented_features/2026_05_23_bug_contract_test_stub_missing_target_filter_kwarg/idea.md) | Bug | `backend/tests/contract/test_error_codes.py::TestErrorCodes::test_targets_forbidden` and `::test_targets_unreachable_via_adapter` both define an inline `_Stub` class whose `list_targets` method has th | — | Complete |
 | [bug_cursor_decode_value_validation](implemented_features/2026_05_17_bug_cursor_decode_value_validation/idea.md) | Bug | `backend/app/db/repo/_sort.py:decode_cursor()` performs a `json.loads(base64.urlsafe_b64decode(raw))` round-trip and then takes `decoded[0]` + `str(decoded[1])` without validating the payload shape or | — | Complete |
@@ -179,13 +180,12 @@ _None._
 
 _None._
 
-### Idea (3)
+### Idea (2)
 
 | # | Priority | Feature | Type | One-liner | Depends on | Status |
 |---|---|---|---|---|---|---|
-| 1 | P2 | [bug_ceiling_badge_assumes_maximize_direction](planned_features/01_mvp1/bug_ceiling_badge_assumes_maximize_direction/idea.md) | Bug | The `CEILING` badge in [`studies-table.column-config.tsx:METRIC_CEILING_THRESHOLD`](ui/src/components/studies/studies-table.column-config.tsx) flags rows where `best_metric >= 0.99`. The threshold che | — | — |
-| 2 | P2 | [bug_smoke_studies_data_table_search_flake](planned_features/01_mvp1/bug_smoke_studies_data_table_search_flake/idea.md) | Bug | [`ui/tests/e2e/studies-data-table.spec.ts:20-40`](../../ui/tests/e2e/studies-data-table.spec.ts#L20-L40): | — | Idea — surfaced during PR #273 CI watch. |
-| 3 | Backlog | [chore_demo_reseed_stale_recovery_atomic_cas](planned_features/01_mvp1/chore_demo_reseed_stale_recovery_atomic_cas/idea.md) | Chore | PR #299 added stale-status auto-recovery to the demo-reseed POST handler ([`_test.py`](../backend/app/api/v1/_test.py)): when the Redis status is `running` but `started_at` is older than `DEMO_RESEED_ | — | Idea — captured during PR #299 GPT-5.5 final review (finding #2, adjudicated non-regression) |
+| 1 | P2 | [bug_smoke_studies_data_table_search_flake](planned_features/01_mvp1/bug_smoke_studies_data_table_search_flake/idea.md) | Bug | [`ui/tests/e2e/studies-data-table.spec.ts:20-40`](../../ui/tests/e2e/studies-data-table.spec.ts#L20-L40): | — | Idea — surfaced during PR #273 CI watch. |
+| 2 | Backlog | [chore_demo_reseed_stale_recovery_atomic_cas](planned_features/01_mvp1/chore_demo_reseed_stale_recovery_atomic_cas/idea.md) | Chore | PR #299 added stale-status auto-recovery to the demo-reseed POST handler ([`_test.py`](../backend/app/api/v1/_test.py)): when the Redis status is `running` but `started_at` is older than `DEMO_RESEED_ | — | Idea — captured during PR #299 GPT-5.5 final review (finding #2, adjudicated non-regression) |
 
 ## Dependency graph
 

@@ -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">94 / 95 scoped done · 4 remaining</div>
+  <div class="progress">94 / 95 scoped done · 3 remaining</div>
   <span class="state-pill in_progress">In progress</span>
 </div>
 

@@ -403,12 +403,12 @@ <h2>MVP1 Progress</h2>
     </div>
     <div class="kpi warn">
       <div class="label">Pending work</div>
-      <div class="value">4</div>
+      <div class="value">3</div>
       <div class="sub">every not-done feat/infra/chore/bug across all priorities</div>
     </div>
     <div class="kpi bug">
       <div class="label">Open bugs</div>
-      <div class="value">2</div>
+      <div class="value">1</div>
       <div class="sub">tracked bug_* idea files</div>
     </div>
     <div class="kpi ">
@@ -425,7 +425,7 @@ <h2>MVP1 Progress</h2>
     </div>
     <div class="kpi">
       <div class="label">P2 (default)</div>
-      <div class="value">3</div>
+      <div class="value">2</div>
       <div class="sub">important to file, not blocking</div>
     </div>
     <div class="kpi">
@@ -435,7 +435,7 @@ <h2>MVP1 Progress</h2>
     </div>
     <div class="kpi">
       <div class="label">Legacy "Path to MVP1"</div>
-      <div class="value">4</div>
+      <div class="value">3</div>
       <div class="sub">scoped not-done + bugs + chore-ideas only (excludes feat/infra ideas)</div>
     </div>
   </div>
@@ -463,20 +463,7 @@ <h2>Pipeline</h2>
   </div>
   <div class="kanban">
 <div class="col idea">
-  <h3>Idea <span class="count">3</span></h3>
-
-<div class="card bug" data-prefix="bug" data-priority="P2">
-  <div class="name"><a href="../../docs/00_overview/planned_features/01_mvp1/bug_ceiling_badge_assumes_maximize_direction">Ceiling Badge Assumes Maximize Direction</a></div>
-  <div class="meta">
-    <span class="badge bug">Bug</span>
-    <span class="badge priority" data-priority="P2">P2</span>
-
-  </div>
-  <div class="one-liner">The `CEILING` badge in [`studies-table.column-config.tsx:METRIC_CEILING_THRESHOLD`](ui/src/components/studies/studies-table.column-config.tsx) flags rows where `best_metric &gt;= 0.99`. The threshold che</div>
-
-
-</div>
-
+  <h3>Idea <span class="count">2</span></h3>
 
 <div class="card bug" data-prefix="bug" data-priority="P2">
   <div class="name"><a href="../../docs/00_overview/planned_features/01_mvp1/bug_smoke_studies_data_table_search_flake">Smoke Studies Data Table Search Flake</a></div>
@@ -533,7 +520,7 @@ <h3>Implementing <span class="count">1</span></h3>
 </div>
 
 <div class="col done">
-  <h3>Done <span class="count">127</span></h3>
+  <h3>Done <span class="count">128</span></h3>
 
 <div class="card feat" data-prefix="feat" data-priority="P2">
   <div class="name"><a href="../../docs/00_overview/implemented_features/2026_05_21_feat_agent_propose_search_space/feature_spec.md">Agent Propose Search Space</a></div>
@@ -1783,6 +1770,19 @@ <h3>Done <span class="count">127</span></h3>
 </div>
 
 
+<div class="card bug" data-prefix="bug" data-priority="P2">
+  <div class="name"><a href="../../docs/00_overview/implemented_features/2026_05_29_bug_ceiling_badge_assumes_maximize_direction">Ceiling Badge Assumes Maximize Direction</a></div>
+  <div class="meta">
+    <span class="badge bug">Bug</span>
+
+    <span>merged 2026-05-29</span>
+  </div>
+  <div class="one-liner">The `CEILING` badge in [`studies-table.column-config.tsx:METRIC_CEILING_THRESHOLD`](../ui/src/components/studies/studies-table.column-config.tsx) flags rows where `best_metric &gt;= 0.99`. The threshold </div>
+
+
+</div>
+
+
 <div class="card bug" data-prefix="bug" data-priority="P2">
   <div class="name"><a href="../../docs/00_overview/implemented_features/2026_05_26_bug_clone_e2e_seed_template_params_mismatch">Clone E2E Seed Template Params Mismatch</a></div>
   <div class="meta">

@@ -20,11 +20,11 @@ MVP1 (v0.1) **shipped** — all six differentiators live (Bayesian/TPE optimizer
 
 Detail + reasoning for each is in [`state_history.md`](state_history.md).
 
-- **2026-05-29** — `chore_state_md_size_compression` (PR #303 + finalization #304). Split `state.md` (360 KB → 9.3 KB snapshot) from new `state_history.md` (append-only narrative, root); added `state-md-size-guard` pre-commit hook (60 KB cap) + CLAUDE.md snapshot-vs-history convention. **This is the first merge recorded under the new convention.**
+- **2026-05-29** — `bug_ceiling_badge_assumes_maximize_direction` (PR #305 + finalization #306). Studies-list CEILING badge (best_metric ≥ 0.99) mislabeled minimize studies (0.99 is a bad score there). Preflight found it had gone latent→live (feat_study_baseline_trial made `direction=minimize` creatable). Added `direction` to StudySummary (defaults maximize) + gated the badge on `direction !== 'minimize'` (rolling-deploy-safe per Gemini). 7 tests.
+- **2026-05-29** — `chore_state_md_size_compression` (PR #303 + finalization #304). Split `state.md` (360 KB → 9.3 KB snapshot) from new `state_history.md` (append-only narrative, root); added `state-md-size-guard` pre-commit hook (60 KB cap) + CLAUDE.md snapshot-vs-history convention. **First merge under the new convention.**
 - **2026-05-29** — `chore_e2e_api_base_url_construction` (PR #301 + finalization #302). Swept 28 `${API_BASE}<path>` concats across 10 e2e specs to `new URL(...)`; aligned dashboard-reseed's API_BASE env var; URLSearchParams for a query. Mechanical, zero behavior change.
 - **2026-05-29** — `bug_demo_reseed_button_silent_enqueue_failure` (PR #299 + finalization #300). Top-level `except Exception` barrier in `run_demo_reseed` + `reseed_status_is_stale()` POST auto-recovery so a worker init crash flips Redis to `failed` instead of stuck-`running`. 14 unit tests.
 - **2026-05-29** — `bug_smoke_seed_es_unavailable_shards_race` (PR #297 + finalization #298). Retry bulk-index on `unavailable_shards_exception` in the ES seed path; fixed the chronic smoke-gate red badge.
-- **2026-05-28** — `infra_pr_yml_drop_push_main_trigger` (PR #295 + finalization #296). Dropped the `push: branches: [main]` trigger from `pr.yml`; rewrote release-checklist `gh` queries.
 
 ## In flight
 

@@ -32,6 +32,8 @@
 
 > Note: this section is significantly behind the actual commit history — PRs #258–#295 are not represented here (the state.md size-compression chore at [`chore_state_md_size_compression`](docs/00_overview/implemented_features/2026_05_29_chore_state_md_size_compression/idea.md) tracks the broader cleanup). New entries land at the top; backfill is out of scope for any single feature PR.
 
+- **2026-05-29 — `bug_ceiling_badge_assumes_maximize_direction` merged into `main` as PR #305.** The studies-list "Ceiling" badge flagged any `best_metric >= 0.99` — correct for maximize objectives (NDCG/MAP/MRR/Precision/Recall pinned at their upper bound) but wrong for minimize objectives, where 0.99 is a *bad* score and the badge claimed the opposite. **Preflight finding — latent→live:** the idea (2026-05-27) deferred on the premise that no minimize study could be created, but `feat_study_baseline_trial` had since added `direction: Literal["maximize","minimize"] = "maximize"` to `ObjectiveSpec` (schemas.py:579), so a `direction=minimize` study is creatable via the API today — the bug actively mislabels it. **Fix (smallest option):** add `direction: ObjectiveDirection = "maximize"` to `StudySummary` (`_summary()` reads `row.objective.get("direction","maximize")` so pre-feature rows default correctly), gate the badge on `direction !== 'minimize'` so a minimize study shows no false badge. `types.ts` hand-edited (CI treats committed types as source of truth). **Cross-model review:** GPT-5.5 final CLEAN; Gemini 1 Medium accepted (use `!== 'minimize'` not `=== 'maximize'` so an absent/undefined direction during a rolling deploy where the FE ships ahead of the BE still defaults to maximize — matches the backend default; added a regression test for the undefined case). 7 regression tests (4 FE column-config + 3 BE `_summary`), FE minimize-case verified failing on `main`. **No migration** (`direction` already lives in the existing `objective` JSONB); additive, backward-compatible, code-only.
+
 - **2026-05-29 — `chore_state_md_size_compression` merged into `main` as PR #303.** Split `state.md` into a one-page snapshot (360 KB / 490 lines → 9.3 KB / 98 lines) + this `state_history.md` (append-only narrative + archived chained execution context, repo root). Per operator decisions (AskUserQuestion): history file at root, last 5 merges retained as one-liners in `state.md`, blocking pre-commit hook added now. Ships `scripts/check-state-md-size.sh` + `.pre-commit-config.yaml` `state-md-size-guard` hook (60 KB cap; portable POSIX `wc -c`; failure message says "move to history, don't raise the cap"; verified PASS@9.3 KB / FAIL@61 KB) + `CLAUDE.md` §"Active Work" + §"Compressed Context First" snapshot-vs-history convention (new merge entries land here, `state.md` keeps only the last 5 one-liners). **Cross-model review:** GPT-5.5 1 Low accepted (error-message section reference points to both CLAUDE.md sections); Gemini 1 Medium accepted (dropped BSD/GNU `stat` branching for single POSIX `wc -c`). The note above (#258–#295 backfill gap) is intentionally left — backfill remains out of scope. **No migration; no production code change** beyond a doc-hygiene script + pre-commit hook. **This is the first feature recorded under the new snapshot-vs-history convention.**
 
 - **2026-05-29 — `bug_smoke_seed_es_unavailable_shards_race` merged into `main` as PR #297.** Fixes the chronic smoke-gate red badge that's been blocking every PR since 2026-05-23. **Actual root cause** (found via the smoke-logs artifact from PR #297 run `26612512222`, third CI iteration): ES's disk-watermark allocation gate. GHA runners boot with ~6.8 GB free out of ~71 GB (~9.5%), tripping ES's default 90% high-watermark threshold; ES then refuses to allocate the products-index primary shard — `cluster_status: red`, `active_primary_shards: 0`, `initializing_shards: 0`. The watermark behavior is by-design on multi-node clusters (relocate shards off a full node) but harmful on the single-node dev/CI setup. Fix: `cluster.routing.allocation.disk.threshold_enabled=false` added to the `elasticsearch` service env in `docker-compose.yml`. **Secondary hardening shipped in the same PR** as defense-in-depth for cold-start timing variance: (a) `_cluster/health/<index>?wait_for_status=yellow&timeout=10m` probe between index create and bulk (synchronizes with ES allocation state machine instead of guessing); (b) `_bulk_with_retry` + `_first_bulk_error` helpers extracted as pure functions; (c) retry on `unavailable_shards_exception` only (3 attempts × 2s) — mapping bugs / type mismatches still fail loudly on attempt 1; (d) `RETRYABLE_BULK_ERROR_TYPES` frozenset for source-of-truth allowlist; (e) 408 + `timed_out: true` handled the same as 200 + `timed_out: true` (the documented ES wait_for_status timeout response). 11 unit tests in `backend/tests/unit/scripts/test_seed_es_retry.py` mock httpx to exercise the retry contract: succeed-after-N-transients, exhaust-retries-returns-False, non-retryable-fails-immediately, happy-path-one-call, plus 4 helper tests and 2 constant-pinning tests. **CI**: 9/9 green — first all-green PR run on this branch since the chronic flake began on 2026-05-23 (6 days). **Process reflection**: the iterative trip (3 CI cycles, ~30 minutes of agent + GHA time) on retry budget tuning + health probe design was wasted; the right move would have been to pull the `smoke-logs` artifact on the first failure and read the ES container output before tuning the script. The disk-watermark WARN was right there in the container logs from the start. Lesson: when a fix's expected effect doesn't match the observed CI behavior, the symptom is misleading you about the layer — look at the container/system logs before iterating on the application layer. **Cross-model review**: Gemini Code Assist (1 Medium finding rejected with cited counter-evidence — defensive isinstance ladders on `_first_bulk_error` for hypothetical-only ES response shapes); final GPT-5.5 returned `[]` empty findings (convergence cycle 1). **Tangential discovery captured + filed inline**: [`bug_seed_meaningful_demos_silent_bulk_errors`](docs/00_overview/planned_features/00_unsure/bug_seed_meaningful_demos_silent_bulk_errors/idea.md) — sibling script `scripts/seed_meaningful_demos.py:917-935` has the same race + worse, silently `resp.read()`s the bulk response body without parsing `payload["errors"]`. Affects local `make seed-demo` operators (CI doesn't exercise that path anymore per `chore_drop_demo_seed_from_ci`). Pick up after this fix establishes the retry pattern. **No migration; no Alembic head movement; no production code change** beyond a script + Docker config.