Weekly report — work week 6 (2026-W38)

Weekly report — work week 6 (2026-W38)

• Period: 2026-09-14 ~ 2026-09-20
• Stage: revalidating the Connors RSI(2) daily signal and diagnosing follow-up strategies
• Status: complete — EXP-012 hold, EXP-014 inconclusive, EXP-018 insufficient_data
• Execution model: v2

This week's goal

Revalidate the signal effect of daily RSI(2) and identify the conditions under which it could extend to a new strategy.

Execution scope

• The 30 symbols fixed in advance in EXP-011: 7 control symbols and 23 generalization symbols
• train(2010~2018), validation(2019~2022), historical_recheck(2023~2026-08-26)
• frictionless, normalized_actual_proxy, normalized_2x_stress, 4 cost profiles up to normalized_plus_5bp_slippage
• Only the base RSI(2) form run, an independent 3,000,000 KRW account per symbol, per-period capital isolation
• Execution model v2, 360 strategy cells

The design is the 0, and the outputs are the 1 and the 2.

Key results

The mean net return of signal events was about +1%, and the matched control about -2%. The event effect is positive, but holding-day exposure was low at 7.42%, so standalone operability is not yet confirmed.

Follow-up revalidation results

EXP-014 was run with the EXP-012 aggregation and control contracts newly fixed. Using a same-date above-SMA200 comparison set and non-overlapping RSI signals, the history_B generalization group's 5-day excess effect was +0.16%, but the 21-day block bootstrap 95% interval [-0.51%, +0.92%] included 0. The comparison-set valid ratio was also 74.8%, short of the 80% criterion, so the signal verdict is inconclusive.

The EXP-018 conditional map observed a mean effect of +1.26% on market down days and -0.10% on up days, and by RSI signal concentration, +0.65% on low days vs -0.16% on high days. The block intervals of both contrasts include 0 and there are no account ledgers, so it was not promoted to an operating filter and was recorded as insufficient_data.

Verdict

• Signal: signal_positive — the sample gate was passed and the bootstrap interval is above 0.
• Standalone operation: standalone_hold — median excess CAGR is negative and median exposure falls short of the 10% gate.
• Final strategy status: connors_rsi2 is not promoted to a live-trading candidate.
• Shared-capital EXP-013: start postponed because EXP-014 is inconclusive.
• Strategies admitted to live trading: still 0.

Validation and outputs

• 37 regression, harness and execution-model tests passed
• Data audit: 30 sources, 0 duplicate or invalid OHLC rows
• EXP-012 summary ledger: reports/ledgers/EXP-012-summary.csv
• Benchmark: reports/ledgers/EXP-012-benchmark.csv
• Event ledger: reports/ledgers/EXP-012-events.csv
• Run manifests: reports/experiments/EXP-012-data-manifest.json, EXP-012-run.json
• EXP-014 event and summary ledgers: reports/ledgers/EXP-014-events.csv, EXP-014-summary.csv
• EXP-018 conditional ledgers: reports/ledgers/EXP-018-condition-events.csv, EXP-018-condition-summary.csv

Limits

• The recent period was viewed repeatedly in past experiments, so it is not new OOS.
• Survivorship bias remains because the data are based on the current listing snapshot.
• Costs are normalized_actual_proxy, not actual CREON execution commissions.
• Event matching is a signal-effect diagnostic, not grounds for approving a separate operating strategy.

Next week's plan

1. EXP-015 point-in-time universe pre-audit, then a feasibility verdict
2. Check the product and distribution data gates for EXP-016 ETF dual momentum
3. Record the official retirement of the EXP-006 and EXP-009 5-minute lines
4. Confirm actual CREON execution commissions and document the cost model

Further RSI parameter searches and RSI 5-minute expansion are on hold.