Voltensor: Australian (NEM) data files, columns explained

voltensor-au-test-daily.csv: the historical test, one row per test, trading day and region.
  test
    day_before_13:00        plan made at 13:00 AEST the day before, from AEMO's run labelled 13:00 (2026-04-02 to 2026-10-30, 212 days)
    same_morning_04:00      plan made at 04:00 AEST on the day, from AEMO's run labelled 04:00 (2026-04-02 to 2026-10-30, 212 days)
    replanned_every_30_min  re-planned every 30 minutes from the latest hourly AEMO run, looking up to 40 hours ahead (2026-04-02 to 2026-10-31, 213 days).
                            Different battery rules: it may cycle more than once a day, pays A$5/MWh for wear and is paid
                            30-minute average prices. Not comparable with the two one-cycle tests.
  trading_day   AEMO trading day: the 5-minute intervals ending 04:05 to 04:00 AEST the next day
  region        AEMO region code: NSW1, QLD1, SA1, TAS1, VIC1
  perfect_foresight_aud_per_mw        the best plan, known only in hindsight (the ceiling)
  last_week_average_plan_aud_per_mw   plan chosen from the average price at each half-hour over the seven most recent known days
  aemo_plan_aud_per_mw                plan chosen from AEMO's PREDISPATCH forecast
  voltensor_plan_aud_per_mw           plan chosen from (AEMO's forecast + the Voltensor model's expected price) / 2
  voltensor_minus_aemo_aud_per_mw     voltensor_plan minus aemo_plan

voltensor-au-live-daily.csv: the live forward test, one row per trading day and region.
  record        the live record the plan belongs to: 1 from 8 November 2026; a new record starts when the model design changes
  Only day-before plans made on time (within 60 minutes of their AEMO run, logged before any of the prices were known).
  The 1-8 November 2026 backfill is not included. Every fully priced on-time day is added; none is selected or removed.
  perfect_foresight / aemo_plan / voltensor_plan / voltensor_minus_aemo   as above
  typical_error_aemo_aud_per_mwh        average absolute difference between AEMO's day-before forecast and the actual 30-minute price, 48 half-hours
  typical_error_voltensor_aud_per_mwh   the same for the Voltensor model's expected price (not the plan price)
  typical_error_simple_aud_per_mwh      the same for a simple forecast: the average actual price at that half-hour over the seven
                                        trading days that ended before the plan was made (days D-8 to D-2); empty if one is missing

Both files
  Earnings are A$ per MW of battery for that trading day, paid at actual prices: gross energy arbitrage only, not expected revenue.
  Not included: frequency services (FCAS), network charges, bid and trading costs, taxes, financing, and battery wear except in the
  re-planned test.
  Battery in the one-cycle tests and the live test: 1 MW / 2 MWh, 85% round trip, one 2-hour charge block and one later 2-hour
  discharge block per trading day, paid actual 5-minute prices.
  The historical test covers months the model never saw: the design was chosen on January-March 2026, then fixed.
  Values are rounded to cents; totals can differ from the page by rounding.
  Source data: AEMO public data (NEMweb). Information, not financial advice.