**Most important metric: `close` (daily closing price).** The dataset is one row per stock per trading day, and `close` is the standard price field for price trends since 2012. It is also the field the dataset's decision use cases (overweight/underweight, AI-surge price change) depend on. The other price columns (`open`, `high`, `low`) describe the same day's range, and `volume` is a separate activity measure. **Highest closing records (step 1 result):** - The top five closes are all **SK Hynix (000660.KS)**, between **2025-07-10 and 2025-07-16**. - The single highest close is **300,000** on **2025-07-14**. The day's range was 292,500 to 302,500 on volume of 2,936,976. - The next closes are 298,500 (07-15), 297,000 (07-10), 296,000 (07-16) and 294,500 (07-11). The peak is a cluster of one week in mid-July 2025, not a single spike. **Lowest closing records (step 1 result):** - The five lowest closes are all **BrainChip (BRN.AX)**, between **2014-12-05 and 2014-12-11**. - The lowest close is about **0.009265** (2014-12-05 and every day after it in this window). - On 2014-12-08 through 2014-12-11 the open, high, low and close are identical at 0.009265 and **volume is 0**. These look like stale or filler prices for days with no trading, not real price discovery. **Caveats:** - **Prices are not comparable across companies.** Each stock is quoted in its local currency and share scale. SK Hynix's 300,000 is in Korean won. BrainChip is on the Australian exchange and its price is tiny. The highest and lowest records therefore reflect currency and price scale, not which company is "best" or "worst". For cross-company comparison, use returns or percentage change, or normalise each series to a base date. - **Zero-volume days.** BrainChip's lowest rows show how non-trading days can distort extremes. Filtering out `volume = 0` would give a cleaner "lowest" record. I can rerun this excluding zero-volume rows, or rank by percentage change to get a comparable extreme across companies.