Reamer Research 4.4.0 and 4.5.0 are out. 4.4.0 widens what the research engine runs on and what it reports: datasets larger than RAM, Monte Carlo resampling, an equity curve, and one plainly named function for each job. 4.5.0 widens what a strategy can do: it sees everything it holds, manages it bar by bar, and sends its orders, brackets included, to a running Reamer Server. This post covers both releases and what they mean for your code.

4.4.0: data, reports and one clean interface

Datasets larger than RAM

reamer_run_backtest_files() runs over memory-mapped .bin files, one per ticker, with an optional inclusive start and end timestamp. The operating system pages bars in as the run reaches them, so a dataset larger than RAM runs on an ordinary machine. Results are identical to reamer_run_backtest() on the same bars.

Two ways to build the files ship with the kit, and neither needs a licence:

bin/reamer-csv-build converts an OHLCV CSV. It detects the delimiter, columns and timestamp format, and prints a parse report: skipped rows, duplicate timestamps, and the first and last timestamp.
reamer_write_bin() writes validated, sorted bars from your own program.

A missing, truncated or unsorted file returns REAMER_RESEARCH_ERROR_BAR_FILE before the run starts.

Monte Carlo and the equity curve

reamer_run_monte_carlo() resamples a completed run's per-trade returns with replacement and reports final-equity, loss, ruin and maximum-drawdown percentiles. It is seeded, and bit-identical for any thread count, so a robustness check reproduces as exactly as the run it checks.

The result JSON now carries equity_curve and equity_curve_ts: realised equity after each closed trade, with timestamps. It is the curve the maximum drawdown is measured on, so a chart and the summary metric always agree.

reamer_get_summary() takes a periods_per_year argument for Sharpe and Sortino annualisation, 252 for daily bars or 52 for weekly. Zero or below infers it from the trade timestamps.

One function per job, with plain names

Every export is now reamer_*: reamer_run_backtest(), reamer_get_summary() and so on, all in one version node. Each job has one entry point, on_bar has one signature, and the summary is one 31-metric struct. A ticker's ID is its position in the array you pass. A forward-filled row keeps the last real bar's timestamp, so a strategy can read how stale each ticker is from its window.

At the end of a run, open_orders_end lists every order still pending, and order_log records it as cancelled. A ticker passed with zero bars keeps every later ticker on its own index, name and position.

4.5.0: the strategy manages its own book

The strategy sees its whole book

On every bar, on_bar now receives two things it previously had to track itself:

Position state for each ticker (ReamerPositionState): signed quantity, average entry price, the take-profit and stop-loss in force, unrealised P&L and the entry timestamp.
Every resting order (ReamerOpenOrder), sorted by order_id. A strategy can cancel one of its own orders by the ID it reads here.

The engine's own accounting is the source of these numbers, so a strategy reads the same state the fill rules act on. Python strategies get the same through data.position_state(ticker) and data.open_orders().

Four actions to manage what is open

A strategy can now return up to 64 actions per bar alongside its orders:

MODIFY_ORDER re-prices a resting order and replaces its brackets.
MODIFY_POSITION replaces a position's take-profit and stop-loss. A stop may move to breakeven or into profit, so trailing stops and break-even rules are a few lines in the strategy.
CLOSE_ALL closes every position at market.
CANCEL_ALL cancels every resting order.

Modify and cancel-all take effect from the next bar. A bar the strategy has already seen is settled against the levels it had when it saw it, which keeps the run free of look-ahead. Each action reports whether it was applied and, if not, why. A rejected action leaves the order or position as it was and the run continues. EXECUTION_SPEC.md §6b sets out the timing and validation rule by rule.

Every applied modification appears in the result JSON order_log as a Modified record, so a trade-by-trade diagnosis shows when each stop moved and to where. The result schema is now version 2.

A run you can stop

An optional on_progress callback receives steps done and steps total between bars, at most once per 256 bars and 100 ms, with a final call at completion. Returning nonzero stops the run with REAMER_RESEARCH_ERROR_CANCELLED. A dashboard can show progress on a long sweep and stop a run that is no longer needed. A run's result is the same with or without the callback.

From research to Reamer Server through the same orders

The Reamer Research library now ships a strategy relay: reamer_relay_open(), reamer_relay_send(), reamer_relay_poll(), reamer_relay_get_positions(), reamer_relay_is_connected() and reamer_relay_close().

The relay sends the same ReamerOrderRequest a strategy returns from on_bar in a backtest to a running Reamer Server over its local strategy socket, with the backtest's order IDs and submission rules. Fills come back as the same ReamerPositionState the strategy reads in research. Your logic, your order structure and your instrument names carry over. A stop and target travel with the entry order, and your broker connector places them at the venue. RESEARCH_TO_SERVER.md in the kit runs one order through the reference gate to a simulated fill on Linux and macOS.

The pre-trade gate and the broker connector in Reamer Server stay yours to build. That is where your risk model and your venue live.

Moving to ABI 7

ABI 6 and ABI 7 are both breaking changes. Rebuild strategies against the ABI 7 header:

Rename calls from reamer_research_* to reamer_*, and drop the ticker_ids argument.
Implement the single on_bar signature, with the 4.5.0 state and action parameters, and the vtable {user_data, on_bar, on_progress}. Pass NULL for on_progress if you do not need it.

Two behaviours to note:

A cancel request now takes effect after the current bar replays. An order that fills on the bar the strategy saw stays filled.
A strategy that writes no actions produces the same result JSON as 4.4.0, byte for byte, apart from abi_version and schema_version.

From ABI 7 on, interface versions are additive: a program built against ABI 7 runs unmodified against a newer library. The added state costs about 1 to 2% per on_bar call on the same machine and build flags.

The 4.5.0 kits include everything in 4.4.0 and are available now, and a 30-day trial key is a full licence on the same terms. See pricing and What is Reamer Research?

Reamer Research 4.4.0 and 4.5.0: Larger-Than-RAM Data, Monte Carlo, Full Position State and a Relay to Reamer Server