docs(planning): spec + impl plan for 5 MVP2 idea-stage items

[SoundMindsAI]

What

Takes five idea-only MVP2 folders from idea → cross-model-reviewed feature spec + implementation plan (the idea-preflight → spec-gen → impl-plan-gen pipeline). Each spec and plan went through mandatory GPT-5.5 cross-model review cycles with full adjudication logs in the artifacts. No source code, no migrations — planning artifacts only. Mirrors the precedent set by PR #364 (6 spec/plan pairs landed in one docs PR).

Folders advanced to Plan stage

Folder	Summary	Plan shape	GPT-5.5 (spec / plan)
chore_studies_post_arq_spy_fixture	arq_pool_spy fixture asserting no-enqueue on studies-POST rejection paths	2 stories / 1 epic, no migration	3 cyc, 4/4 acc / 3 cyc, 4/4 acc
bug_judgment_header_omits_click_bucket	Render the click (UBI) source bucket in the judgment-list header	2 stories / 1 epic, frontend-only	2 cyc, 4/4 acc / 1 cyc, 0
bug_baseline_phase_test_isolation	Fix suite-ordering dependency in baseline-phase unit tests (lazy settings read + os.environ leak)	2 stories / 1 epic, regression test	2 cyc, 2/2 acc / 2 cyc, 1 rejected (stale)
chore_template_library_expansion	Curated 6-template library + 3 per-engine tunable-params cheatsheets	8 stories / 3 epics, no migration	7 cyc, 12/12 acc / 5 cyc, 8/8 acc
chore_ubi_reader_search_after_pagination	Exact full-traffic UBI aggregation via unified scan_all (ES/OS PIT+search_after, Solr cursorMark)	5 stories / 3 epics, SearchAdapter Protocol surface, no migration	5 cyc, 16/16 acc / 5 cyc, 14/14 acc

Notable decisions locked during planning

• arq spy fixture: opt-in (not autouse), installed after lifespan (the real Arq pool is built by LifespanManager against CI Redis — the spy replaces it), records flattened (name, *args) tuples.
• baseline test isolation: durable fix is a lazy get_settings() read in _compute_baseline_wait_s (only when no explicit trial_timeout_s); the test_main_lifespan.py os.environ leak is documented but left out of scope.
• template library: kNN + hybrid-RRF demoted from runnable templates to engine-correct reference snippets (the render context has no query vector); structural fields stay Jinja literals to satisfy the platform's declared_params == search_space equality check. Three engines only — no Lucidworks Fusion.
• UBI scan_all: new engine-neutral SearchAdapter.scan_all + close_scan returning an opaque round-tripped cursor (ES/OS one impl with internal PIT-endpoint branch; Solr cursorMark via POST /select) — keeps Absolute Rule infra(foundation): bootstrap MVP1 stack — Docker + FastAPI + /healthz + Alembic + CI (#3) #4 intact.

Not included

Implementation of any of these plans (each is /impl-execute-ready as a follow-up). Two were left untouched because they are design-ahead-gated or smaller in scope.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the MVP2 dashboards and introduces comprehensive ideas, feature specifications, implementation plans, and pipeline statuses for several planned features, chores, and bug fixes, including baseline phase test isolation, judgment header click bucket rendering, an Arq pool spy fixture, template library expansion, and UBI reader pagination. The reviewer feedback highlights several issues in the newly added specifications and implementation plans: in the template library spec, the Solr edismax_basic.j2 template uses ps without defining pf, rendering it ineffective; in the UBI reader pagination implementation plan, the code sketch directly accesses settings and a hardcoded UBI_EVENTS_INDEX constant, which would cause NameErrors; and the _chunk helper's byte-length calculation needs to account for serialization overhead.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

In Solr's edismax query parser, the ps (Phrase Slop) parameter only applies to phrase queries generated from the pf (Phrase Fields) parameter. Since pf is not listed under the baked-in literals or declared parameters for edismax_basic.j2, ps will have no effect. Consider baking in pf or adding it to the template so that ps functions correctly.

[gemini-code-assist]

In the _scan_ubi_events implementation sketch, settings.ubi_query_id_batch_size is accessed directly. However, in UbiReader, settings are typically retrieved via get_settings() or stored on self during initialization (e.g., self._settings). Accessing settings directly without importing or resolving it first will raise a NameError. Please ensure the implementation plan specifies accessing this via self._settings or get_settings().

[gemini-code-assist]

When implementing the _chunk helper to bound batches by encoded byte-length (Settings.ubi_query_id_batch_max_bytes), ensure the calculation accounts for the serialization overhead (such as commas or query syntax like {!terms f=query_id}) rather than just summing the raw string lengths of individual IDs. This is especially important for Solr where the entire query string is sent in the request body/URL.

[gemini-code-assist]

The scan_all call in the sketch uses UBI_EVENTS_INDEX as the first argument. In UbiReader, the index name is typically retrieved dynamically from settings or stored as an instance variable (e.g., self._events_index). Using a hardcoded UBI_EVENTS_INDEX constant might lead to a NameError if it is not defined at the module level. Consider updating the sketch to use self._events_index or settings.ubi_events_index for consistency and correctness.

[SoundMindsAI]

Gemini Code Assist — adjudication summary

All 4 line-level findings adjudicated (against 554982ca); fixes pushed in 56a888ab.

#	File	Finding	Verdict	Action
1	chore_template_library_expansion/feature_spec.md	edismax ps is a no-op without pf	Accept	Baked pf into edismax_basic.j2's literals so the tunable ps takes effect; updated FR-2 + Notes.
2	chore_ubi_reader_search_after_pagination/implementation_plan.md	settings.ubi_query_id_batch_size would NameError in the reader	Accept (remedy adjusted)	The sketch's bare settings.X is indeed out of scope, but UbiReader is deliberately decoupled from Settings (docstring ubi_reader.py:189-190; injects values like position_bias_prior). Fixed by constructor injection (self._ubi_query_id_batch_size), not the suggested get_settings()/self._settings — that would break the decoupling. Updated the sketch + the file-table row.
3	chore_ubi_reader_search_after_pagination/implementation_plan.md	byte-length chunk must count serialization overhead	Accept	Clarified the byte budget is measured on the fully-serialized filter fragment ({!terms f=query_id} wrapper + separators / ES terms JSON), not summed raw id lengths.
4	chore_ubi_reader_search_after_pagination/implementation_plan.md	UBI_EVENTS_INDEX would NameError	Reject (stale/incorrect)	UBI_EVENTS_INDEX is a module-level constant at ubi_reader.py:75, already used directly at ubi_reader.py:515. self._events_index does not exist; the sketch is consistent with the live codebase.

These are planning artifacts (specs + plans), not code — no source changes. Merging on the docs fast lane (heavy pr.yml is path-filtered out for docs-only changes; DCO + secrets-defense green).