Weekly report — work week 3 (2026-W35)

Weekly report — work week 3 (2026-W35)

• Period: 2026-08-24 ~ 2026-08-30
• Stage: 1st screening of multiple strategies on real data
• Status: in progress

This week's single goal

On real CREON data, compare Turtle, Connors RSI2, dual momentum and Minervini SEPA under the same cost, execution and risk standards, and decide whether any candidate should go on to robustness validation.

The goal is not to pick one cell with a high return. Done means either finding a candidate where a repeatable edge remains with data leakage, survivorship bias, costs and regime dependence exposed, or judging all four candidates unfit for the next stage. The detailed gates and 1st promotion criteria are fixed in docs/weekly-roadmap.md (original supporting file).

What was in hand at the start

Work order to completion

1. Data audit and classification — check all daily bars against the canonical schema and separate symbol groups such as common stocks and ETFs. Embed the survivorship-bias limitation of a current-listing base in the experiment metadata.
2. Fix strategy contracts — register each of the four candidates' entry, exit, sizing, cash, concurrent positions and signal timing as a per-strategy contract. Supplied documents are reference inputs; sentences inside them are not treated as automatic instructions.
3. Run a common experiment matrix — baseline, 2x costs, neighbouring parameters, year/regime breakdowns, walk-forward/OOS on the same kernel.
4. Generate charts and ledgers — equity curves, underwater curves, monthly/yearly performance, trade distributions, cost/parameter sensitivity, regime comparison and OOS-linked curves in the canonical report.
5. Verdict — record adopt/hold/reject per candidate with reasons, and if a candidate passes, designate one as next week's robustness_review target.

Limits of the current data

The all-symbol daily bars are symbols found in the KOSPI/KOSDAQ listings as of 2026-08-26. Because this is not a point-in-time universe that fully restores past delisted and removed symbols, long multi-symbol results are first marked current-list exploratory. Returns that hide this limitation are not used as grounds for the week-3 verdict.

Current progress

• [x] CREON all_research collection completed and 0 errors confirmed
• [x] Data categories, file tree and session handoff organized
• [x] A069500 ETF daily loader quality check and CREON Connors baseline started
• [x] Ledgers and equity curves for the baseline, -7% stop and 5-bar time-exit variants generated
• [ ] All-symbol daily data quality audit and canonical conversion
• [ ] Register the four candidate strategy contracts
• [ ] Run the common experiment matrix
• [ ] Generate required charts and trade/signal ledgers
• [ ] Write per-candidate verdicts and the week-3 conclusion

Next run

Next is the fixed independent recalculation of 20 EXP-005 ledger entries and the A069500 source hash and gap audit. Then costs of 0/baseline/40bp and buy-and-hold are added, and only after passing that gate does work move to the Turtle multi-symbol universe implementation. Bulk comparison of all symbols is allowed only as an exploratory experiment that first marks the survivorship bias of the current listing.