What changed, including what we corrected and what we withdrew

The state of each part of the service, past incidents and this log are now open to anyone. The methodology gains a page with what we measured and what did not work.

Measured out of sample and net of costs it showed no advantage; it is withdrawn until it is rebuilt on a measured expectancy.

Compared with what happens on any day, its result did not hold; it no longer appears in Reports, the Ranking or the AI's answers.

It classified the near-24-hour day instead of the regular session; the guide now uses the table measured on the regular session, with a dated correction note.

The two windows shared hours, which inflated the agreement. London is now measured up to the open of the market's own session; the corrected figure sits close to that of any day.

The near-24-hour candle was being classified, which changed the split between types. Corrected in ten markets; the rest change with the next nightly update.

They used to print the rate of all cases even when the card said 'active' or 'back inside'. The figure now matches what the card shows, with its sample and interval, and the overall rate sits beside it as the base.

Weekday bias, month, day of month and overnight direction move to a separate group that shows what was stated and what was observed. Context describes the size of the move, not its direction.

Every simulated figure carries the hypothetical-results notice beside it. The in-session reading is described as what it is: data delayed by up to 15 minutes, for reference only. The portal footer shows the session the data runs through.

Profile across several timeframes, where the market stands in the auction and what followed in similar situations, measured against a control at the same distance. Navigation and the report catalogue are ordered by moment of the day.

Profile levels against any price at the same distance, the size of the range after a sequence of lows or highs, and whether the first half hour sets the day's extreme.

Several profile reports used the calendar day's high and low instead of the session's, which shifted their figures (for example, the Initial Balance's share of the day). They were recalculated with each market's own session.

Three simulators filled the order at the level even when the session opened beyond it, which cannot happen in practice. Corrected: the favourable result they showed came from that defect.

In metals, energy, bonds and currencies the page detected the gap and opening ranges on the index timetable, different from the one behind the printed figure. Detection and figure now use the same session.

The 0-100 index was calculated from the current year's reports only, not the three years the page stated. It now uses the full three years of weekly reports.

Weekday tables open with the 'All days' row and mark the cells that cannot be told apart from it. In directional reports the all-days rate leads; the weekday breakdown stays for range and structure reports.

Every rate carries its sample and 95% interval; with very few cases the figure is not shown and with few it is marked as a small sample. A badge says whether the rate was stable between two stretches of history, and the report can be recalculated for a period.

In those markets gap fill behaves differently in recent years than in earlier ones. The gap reports say so in their first note instead of blending the two stretches without warning; the cause is not established.

Positioning index per group for the 30 futures, net-position chart and the outcome of past extreme readings compared with random entries.

Next session's levels with how often the market has travelled that far, macro regimes with what followed in each future, and session reports for metals, energy, grains, currencies and crypto on their own timetable.

The Journal replays each trade over the day's candles and receives trades through a link unique to each user. The portal comes down to fifteen destinations in the order of a trader's day, and there is a new guide on when to roll contracts.

For ten of the years the calendar picked the annual revision instead of the monthly report. The dates were corrected and, with them, the retail sales row of the table in the macro data days guide, which carries its correction note.

Four pages stayed on 5 October data from the 5th to the 9th. Their files are now rebuilt every night and the download retries on its own after temporary errors. The details are in the incident log on the status page.

Pasting events into the calendar replaced the shared events file that the daily briefing also reads. Each user now has their own.

Report rate bars draw their interval and flag small samples. The backtest lab shows results after commission and slippage, with the gross figure beside them and a chronological out-of-sample split.

The time of day at which the gap fill and the Initial Balance break usually happen, and how likely they remain if they have not happened by a given time; today's expected range with its out-of-sample calibration; and the premium or discount of the next contract.

The Journal imports files from the most used execution platforms, measures how far each trade went for and against, and adds risk rules with a compliance score. Market Profile adds the volume profile and a period-by-period replay of the session.

Seasonal patterns for 29 futures adjusted at contract rolls. Each pattern is compared with every window of the same length, split into an older and a recent half, adjusted for the number of stretches looked at, and shown net of costs.

Letter profiles for 29 futures, how much volume has already moved to the next contract compared with earlier rolls, and how the indexes moved after each macro release.

Which funding firms allow automation, with the rules read on their official pages and the unconfirmed ones marked as such; and two guides with their dated table, their sample and what the figure does not say.

Each guide is a dated study: its tables no longer change when the data is updated, so text and figures never contradict each other. Nine tables went back to the snapshot their text was written against. A table is only changed together with its text and with a correction note.

The day's colour (up or down) is measured against its own open on every page; before, two screens could answer the same question differently.

Each session's data is added automatically in the early hours of the next day and every report is recalculated. If an essential step fails nothing half-built is published: the platform keeps the previous day's data.