GOAL-REC-TIERING-01 Review-Only Recommendation Score Tiering Gate

[RyanLu0203]

Purpose

Define and validate a review-only recommendation score tiering gate after Quant04, without producing actionable recommendations.

Dependency gate

Do not start unless all are true:

1. Issue GOAL-QUANT-RESEARCH-04 Regime-Conditional Factor Evaluation Gate #9 Quant04 is completed and merged into latest project-current.
2. Post-merge validation passes.
3. Quant04 explicitly identifies at least one factor as ready or conditionally_useful under the governing contract.
4. The User explicitly authorizes this goal after reviewing the Quant04 summary.

If any condition is false, stop and report BLOCKED. Do not self-unlock.

Base branch

Latest authoritative project-current only.

Suggested work branch

codex-max/rec-tiering01-review-only

Scope

Review-only tiering design and validation. Define score tiers, uncertainty handling, warnings, and review semantics. No actionable recommendation execution.

Hard boundaries

• No BUY/SELL/HOLD.
• No target prices.
• No actual recommendation actions.
• No actual position sizes, target weights, portfolio weights, or order quantities.
• No dashboard/frontend.
• No trading, broker, production, local-lake, factor-mining, or DQN/RL.
• No automatic unlock of GOAL-10B.4 or position work.

Required analysis

• tiering contract
• score calibration and uncertainty treatment
• regime-conditional caveats inherited from Quant04
• provider/data-quality warnings
• explicit review-only labels and non-actionability guarantees
• leakage / boundary checks
• deterministic audit and reproducible tests

Validation

Run full project validation and fresh-clone verification. Confirm no forbidden recommendation/action/position outputs are created.

Required outputs

Implementation / tests / config / docs / audit / manifest / report / handoff sufficient for review.

Final response

Goal status / Branch / Commit / Base / Validation / Tiering design summary / Warnings / Boundary confirmation / Recommended next action.

Human action required

Humans review only the standardized summary and whether the review-only contract is acceptable. No line-by-line code review expected by default.

[jiang4wqy]

Status: BLOCKED — dependency gate unmet (ready_factor_count = 0)

Context. A request came in (framed as "GOAL-REC-TIERING-V1-ALPHA-RANKING-SYSTEM") to build a factor scoring → regime-weighted aggregation → stability tiering → ranked_factors_v1.csv (status ready/watch/ignore, Top-K=10) → outputs/premarket_signal_v1.csv (composite signal strength + recommendation tag) layer on top of the now-merged Quant04 outputs (PR #14). That deliverable is this gate (recommendation score tiering), so it is tracked here rather than built ad-hoc.

Diagnosis (authoritative project-current, Quant04 merged):

• goal_rec_tiering01_recommendation_score_tiering_gate = locked_future, implemented_in_repo=false, allowed_next_action = remain_locked_until_goal_quant_research04_regime_conditional_factor_evaluation.
• ready_factor_count = 0 (21 conditionally_useful, 0 ready). Quant04 deliberately gates ready on the base Quant03 candidate_for_rec_tiering to prevent regime-mining; no factor clears that bar.
• Quant04's own allowed_next_action is remain_research_only_no_ready_factor_no_rec_tiering_request — Quant04 does not even request this gate while ready_factor_count is 0.

RecTiering's unlock condition is ready_factor_count > 0 AND explicit User approval. The > 0 half is a data/science gate, not something authorization can waive: assigning ready status or emitting a premarket_signal while 0 factors passed validation would fabricate the readiness the anti-overfitting design withheld, and would create locked recommendation/signal outputs (recommendation_outputs_created must remain false).

Hard boundaries that must hold when this gate is eventually built: research-only; no BUY/SELL/HOLD; no portfolio construction; no trading logic; no modification of Quant04 logic; ranking/tiering only over genuinely ready factors.

Proposed resolution (to reach ready_factor_count > 0 legitimately):

1. Raise real readiness upstream, not the label. Get factors to clear the existing Quant03 candidate_for_rec_tiering thresholds through legitimate research — e.g. GOAL-DATA-PANEL-02 evaluation-panel expansion (more symbols / history / coverage), or refined candidates from the alpha-refinement chain — rather than loosening thresholds to force passes.
2. Optionally, a documented methodology review of whether the Quant03 candidate thresholds are calibrated correctly — a legitimate fix if mis-calibrated, distinct from weakening them to manufacture candidates.
3. Only once ready_factor_count > 0 legitimately → with explicit User approval, unlock this gate per its existing allowed_next and implement a boundary-safe V1: factor alpha-score, regime-weighted aggregation, stability tiering over the ready set, and a research-only, non-actionable ranking artifact (no signal/recommendation output at the actionable root, no BUY/SELL/HOLD).

Not acceptable: self-unlocking this gate, assigning ready with 0 candidates, or emitting premarket_signal / recommendation outputs before a legitimate unlock.

Holding per the Codex Max protocol (dependency unmet → BLOCKED, no self-unlock) pending a User decision on either the governance-level unlock or the upstream-research path above.