docs: record I5d standalone MMP feasibility progress

[StackOverFlow11]

What changed

Docs-only progress update in CLAUDE.md recording the merged minute-factor milestones:

• P-I5c (PR feat: I5c MMP minute-factor study (exploratory) #37) — MMP (Minute Microstructure Pressure) minute factor + EW daily aggregation; configurable intraday.score_feature; first EXPLORATORY minute factor wired end-to-end as an opt-in alpha through the event engine. Daily close-to-close unchanged.
• P-I5d (PR feat: I5d MMP minute-factor grouped backtest — standalone feasibility (incl. I5c plumbing) #39, standalone against main) — MMP 5-quantile grouped backtest of the I5c daily score intraday_mmp20_ew_0930_1450: each monthly rebalance ranks the cross-section into 5 equal-count buckets (Q1 low / Q5 high), each group runs as its own long-only equal-weight portfolio through the same BacktestEngine + IntradayTailEventModel + SimExecution(fee_rate=0.001) with I5b raw stk_limit execution feasibility ON.
  • Real run: CSI500 (000905.SH), 2021-06-01 → 2026-05-31, 59 monthly rebalances; final NAV Q1→Q5 0.9822 / 1.0423 / 1.0495 / 1.1275 / 1.1577 (monotonic, Spearman 1.0000); synthetic Q5−Q1 spread +17.15% cumulative.

Also updated: merged-PR list, quality-gate counts (577 passed / 17 configs; +I5c=18 +I5d=19), and the next-step roadmap (I5d independent generalization).

EXPLORATORY — not a performance/tradable claim

The MMP quintile monotonicity is the project's first positive intraday signal but is one overlapping 5-year window on one universe, no tuning, no robustness matrix, no learned weights. Q5 > Q1 is NOT framed as a performance claim.

Caveats preserved

• covered=892/995 — 103 minute-uncovered CSI500 constituents dropped (disclosed; the 5-year window is not shortened).
• stk_mins_live_calls=0 (cache-only minute load).
• Q5−Q1 spread is a synthetic long-only leg difference, not a separately executed dollar-neutral book.
• Execution model still lacks partial-fill / liquidity / volume-cap realism.

Superseded-PR note (#37 / #38)

• feat: I5c MMP minute-factor study (exploratory) #37 (I5c) auto-merged when feat: I5d MMP minute-factor grouped backtest — standalone feasibility (incl. I5c plumbing) #39 landed — its head feat/i5c-mmp-minute-factor became a main ancestor (the 5 I5c commits were bundled into feat: I5d MMP minute-factor grouped backtest — standalone feasibility (incl. I5c plumbing) #39). Not a manual merge.
• feat: I5d MMP quintile 5Y group backtest (exploratory, stacked on I5c) #38 (i5d stacked on i5c, base feat/i5c-mmp-minute-factor) is fully superseded by feat: I5d MMP minute-factor grouped backtest — standalone feasibility (incl. I5c plumbing) #39 — zero remaining commit/file diff vs main. Left OPEN pending cleanup; not closed (no explicit approval to close PRs).

Verification

• git diff --name-only main...HEAD → CLAUDE.md only (docs-only, no code change).
• git diff --check main...HEAD → clean.
• Secret scan on changed doc: real .config.json token value 0 hits; no FAKE_TUSHARE_TOKEN; .config.json / tushare.token appear only as pre-existing config-key / scan-target descriptions.
• On merged main: pytest 577 passed, ruff clean, 17 configs validate, run-phase0 → ic_mean=0.9600, annual_return=0.8408 (no behavior drift).

Docs-only, no code behavior change. (AGENTS.md progress history predates this — stuck at P4-2/#25 — and is a separate drift not resynced here.)