Skip to the page
RELEASES/RELEASE NOTES

Release notes

An untraded run that names its preset replays exactly on every later release. A run that used the default preset replays only on releases with the same default, so each section below names the preset that moved. A traded run replays exactly on the release that recorded it, and one recorded before 0.8.5 matches only up to its first trade. The full history is in the library's CHANGELOG.md.

0.11.0

2026-11-03

pt-v22 is the default model. It lists contracts beside the stocks: index futures with a night session, policy-rate, term-rate and oil futures, and American options on every name. An agent trades them by symbol in whole contracts, and exercises an option at a close with tf.Exercise (Futures, Options). Over 21 years pt-v22 passes 190 of the 190 long-run criteria that count toward its grade, and reports 41 more beside them, five of them outside their marks (How it is measured).

Margin is set at every close (Margin). A future is margined from a risk scan and pays its move to the day's settlement mark in cash. A name's options and its stock are margined together, as a broker's portfolio margin is: the worst loss over ten moves of the name up to 15% either way, and at least 0.375 dollars a share for each option. An order that would leave equity under the initial margin is refused with MarginError. A close under the maintenance level makes a call due by the next open. If equity isn't back at the initial margin then, or the close left the account below zero, the account's contracts and its options' stock are sold at the market.

Scenario.load("recession_2008") adds the 2007-09 recession, with the VIX following the shape of the real 2008-09 VIX. tradefloor.contracts.explain_move says why a contract's price moved, and a strategy spec can hold the front index future as a hedge. What's new has the detail.

Some changes break code written for 0.10. A run that leaves model out now gives pt-v22's market, so any run that used the default gives a different one; model="pt-v21" keeps 0.10.2's, and every preset replays as it did. On the default an agent's observation carries the contracts, so the payload is now version 3. The Rust crate's API only grows.

0.10.2

2026-10-09

A patch release for programs that host the engine, with no coefficient, preset or default changes. pt-v21 stays the default, and every known-answer digest is 0.10.1's, so a run replays exactly as it did on 0.10.1.

A snapshot restored onto an engine rebuilt on a later day's companies now continues bit for bit. From pt-v19 the VIX's anchor is derived from the roster an engine is built on, and the snapshot did not carry it, so the rebuilt engine's VIX parted from the original's at its next close. Snapshots now carry vix_anchor, and ones written before still restore.

A host can write each company's share count, for buybacks and share issues, with Engine.set_shares_outstanding. Each market cap is reset at the price standing, and everything the engine weights by capitalisation reads the new counts.

A host can raise the VIX's target without writing the VIX. Engine.set_vix_target_premium adds a premium in VIX points that fades on a half-life the host chooses, and Engine.set_vix_target_floor holds the target at or above a level until it is cleared. The engine's own reversion carries the VIX to the raised target, where a one-off write of the VIX reverts within days (Embedding the engine).

The snapshot and the state hash carry each new input only once a host has used it, so saves and hashes from a host that never calls them are unchanged.

0.10.1

2026-10-06

A library and documentation release with no coefficient or default changes. pt-v21 stays the default, and every known-answer digest the Python package checks is 0.10.0's, so a run replays exactly as it did on 0.10.0.

Building the same pt-v21 market twice is now quick. The engine keeps the last 16 builds that played the market's two years of prehistory, and a later build from the same seed, roster and model is a copy of the first, the same engine to the bit, in about 0.2 milliseconds. Engine.set_opening_cache_capacity sizes the cache, and Engine.opening_cache_info reports what it holds.

Two realism readings changed. The one-year table now reads the index's return and its falls of 3% or more over 360 seeds, as the long-run criteria do, and pt-v21 has all 19 statistics inside their bands. The two-year panel grades all 15 of its statistics, with a new band for how long high correlation lasts (How it is measured).

For Rust hosts that drive the engine themselves there's a new guide, Embedding the engine, and a close that takes the host's economic shocks, Engine::close_day_with_shocks. EngineRngState saves its ten random streams by stream id, not in the order the struct declares its fields, so a host that packed the fields in declaration order has the news and volume_idio streams swapped in every state it saved. Swap the two five-word blocks once when loading such a state, or use the new to_named_words and from_named_words.

The browser build's Sim now numbers its days as run_days does. Before 0.10.1 every day of a Sim was day 0, so on pt-v21 its earnings and dividend calendars never moved. A negative crisis_epicentre_extra is now refused.

0.10.0

2026-10-05

pt-v21 is the default model. It's pt-v20 with 104 dials moved, and it passes all 40 long-run criteria (How it is measured). Any run that used the default gives a different market, and the known-answer digests are re-based. pt-v20 and the older presets keep their values, and a run that names one replays exactly. Default market model covers what the new model adds.

This release adds populated mode. tf.Population puts background traders in the book alongside your strategy. Pass it as population= to tf.evaluate, tf.World or tf.Engine (Populated mode). Isolated mode is still the default and hasn't changed.

The MCP server can run any shipped preset, keep a market open between calls as a session, and run a population on request. On pt-v21 each asset in the observation payload also has dividend and next_earnings_in_sessions, so the payload is now version 2. Recordings made on version 1 still replay.

Some changes break code written for 0.9. Import the ten engine internals deprecated in 0.9.0 from tradefloor._core, since they're gone from the top level. MCP results and run manifests no longer have the old pretium_version key, and tradefloor_version holds the same value. A checkpoint written by 0.9 won't resume on 0.10. Building a pt-v21 engine takes about 1.5 seconds for 20 names, because the market plays two years of history before day 0. From 0.10.0 the Rust crate follows semver, and its structs and enums are #[non_exhaustive].

0.9.1

2026-10-03

A patch for the MCP server. The known-answer digests are 0.9.0's and pt-v20 is still the default, so runs replay exactly as before.

All thirteen tools have a title, annotations and a description. The description says when to use the tool over its nearest sibling, what it costs and what it returns. run_stress_scenario takes fork_day. Both markets run together up to that day and the scenario starts there, so the difference between them is the scenario's effect on one shared history (LLM adapters and MCP).

0.9.0

2026-10-03

This minor release changes the library and leaves the model alone. The known-answer digests are 0.8.8's and pt-v20 is still the default, so runs replay exactly as before. It adds API and refuses some input that 0.8.8 accepted. A patch on the 0.8 long-term support line may do neither, so the 0.8 line continues alongside it (Support policy).

Engine.restore_state refuses a snapshot with a missing field, an unknown field or a malformed value, and names the field. A refused restore leaves the engine as it was. Snapshots carry state_schema. Those written by 0.8.5 to 0.8.8 restore as before, and earlier ones are refused.

tf.fair_value(..., model=) matches the engine's valuation to the bit. The new Engine.fair_values() returns each name's fair value as the next tick starts from it. Scenario.starting_at(day) and World.apply(scenario, at=day) start a packaged scenario on a chosen day. The new switch order_flow_depth_law, 0 on every preset, divides order-flow impact by a name's depth once instead of twice (Model parameters).

Ten engine internals, such as apply_mispricing and check_rate, warn when you read them from the top level. Import them from tradefloor._core. They stay at the top level until 0.10.0 at the earliest, and tradefloor._api lists the tier of every top-level name (API index).

0.8.8

2026-10-03

A patch on the 0.8 long-term support line. The known-answer digests are 0.8.7's and pt-v20 is still the default, so runs replay exactly as before.

tradefloor is now in the MCP Registry as io.github.simoncoombes/tradefloor. A client can start it with uvx --with "tradefloor[mcp]" tradefloor mcp, and the new tradefloor mcp command runs the same server as tradefloor-mcp. Local MCP server has the setup.

0.8.7

2026-10-03

A text-only patch on the 0.8 long-term support line. The known-answer digests are 0.8.6's and pt-v20 is still the default, so runs replay exactly as before.

The documentation inside the package was rewritten for users. Each entry in Model parameters now says what the coefficient does, its units and which presets carry which value. The gap descriptions in the published envelope, the reasons score gives for a row it can't read, and several docstrings were rewritten in plain terms too. A few stale claims were corrected along the way. A test now checks every value stated in a parameter's description against the preset table.

0.8.6

2026-10-01

A patch on the 0.8 long-term support line, with no coefficient, default or trajectory changes. The known-answer digests are 0.8.5's and pt-v20 is still the default, so runs replay exactly as they did under 0.8.5.

tf.envelope.check() now says why it refuses a sector-concentrated roster on pt-v20. The four-mix roster run was repeated on pt-v20 on seeds 101 to 130, at 252 and 504 days. At 252 days every mix held every shape row the bands can grade. At 504 days the S&P-like and technology-heavy mixes put the correlation of volume with the size of a move at 0.6367 and 0.6332, over the ceiling of 0.63, and the balanced roster reads 0.6266. The refusal and the roster-concentration gap in How it is measured quote this run, which is in the library as measurements/roster-shapes-pt-v20.json. A question that passes preset="pt-v19" keeps the pt-v19 grant.

0.8.5

2026-10-01

Long-term support

0.8.5 starts the first long-term support line. The 0.8 line gets bug and security fixes for 24 months, and a fix ships in a 0.8.x patch only if it leaves every known-answer digest unchanged. The policy is the library's SUPPORT.md, and Support policy lists the digests.

pt-v20 as the default

A run that used the default won't replay against 0.8.1. Pass model="pt-v19" to keep that market. A run that names its preset replays exactly, and you can still select any shipped preset from pt-v1 on.

On pt-v20 the tape follows the model price. A stock's own news and the market's plain shocks move fair value for good, fear marks fair value down while the VIX is above 40, and agents trade in a book with depth. pt-v20 passed all 40 long-run rows registered for it on exam seeds it hadn't been run on, all 19 statistics of the one-year table, and 14 of 14 at two years (How it is measured describes the grade). The rows closest to their limits are the two-year correlation of volume with the size of a move (0.627 against a ceiling of 0.63), a macro timing rule (92% of its tolerance), and the two-year yield, which moves 3.87 bp a day against a real 5.23.

Economic data published late

On pt-v20 a turn in the business cycle reaches macro_fields["cycle"], trace rows and the LLM adapters' payload 252 sessions after it happens. That mirrors the NBER, which dates a turn about a year late. GDP growth is a quarterly figure, released 21 sessions after the quarter ends. Prices follow the true state, and code that needs it can read state_snapshot()["economy"], which a sandboxed agent can't.

No capture ratio on pt-v20

On pt-v20 a shock moves fair value for good. Knowing fair value then gives little edge, and the Oracle's P&L follows the market's month. capture_ratio returns {} with a warning that says why. versus_buy_and_hold gives each agent's P&L minus buy-and-hold's, and rank sorts on mean_excess_pnl. pt-v19 and earlier still report capture.

Orders from act

A number is a market order for that many shares. tf.Limit(quantity, price) and tf.Cancel() now work in evaluate and rank as they already did in World. In evaluate a bad entry is refused with a line in the scorecard's errors, and the rest of the dict trades. Quantities like True and "100" used to trade and are now refused. A falsy return such as [] or 0 used to pass as no trade and is now an error, so return None or {}. Agents and evaluation has the rules.

A read-only market for agents

obs.engine is a MarketView and obs.portfolio a PortfolioView in evaluate, rank, World and tca.analyse. Anything else raises tf.SandboxError. An agent that changes the market or copies the engine is scored tampered, and rank leaves it out. privileged = True gives an agent obs.hidden, and trusted_agents=True hands every agent the live engine. The scorecard marks both. The check only catches writes and copies. An agent can still read the seed and roster from the harness's frames, build a second engine and run it ahead, so run code you didn't write in a separate process.

Warm-up history

evaluate, rank and World take history_days=N, which runs the market for N days with nobody trading before the scored days. obs.history holds a daily bar per name and the published macro figures for every closed day. A 20-day breakout rule can then trade on day 1, where without a warm-up it waits until day 21. The default, history_days=0, changes nothing, and tca.analyse takes history_days too.

Margin interest

In evaluate, rank and World, a negative cash balance pays the policy rate before each close. A levered agent scores less than it did under 0.8.1. With margin_interest=False borrowing is free, and the scorecard says free-borrowing.

Explanation accuracy and its baseline

Each day the scorer asks which of ten factors moved prices most and reads the attribution after the close. It no longer counts fair_value_shift, which moves no price. On pt-v20 random_noise wins almost every day, and naming it every day scores 0.95 to 1.0. The scorecard carries explanation_baseline and explanation_edge and prints all three, and the edge is the number to quote.

New scorecard fields

The scorecard gains equity_curve, max_drawdown_pct, ruined, leverage_refusals, partial_fills, history_days, margin_interest and exposure_curve, the gross exposure after each step. Four read-only properties are computed from those curves. sharpe and volatility_pct are annualized from the daily returns with no risk-free rate subtracted, and the other two are time_in_market and avg_gross_exposure. The repr prints all four and counts errors. Agents and evaluation defines every field with its units.

Refusals in rank

When an LLM adapter refuses a single action, such as a ticker the roster doesn't have, it counts in Scorecard.rejected. AgentRecord.refused holds the count per seed, and the report gives the agent a REFUSED line with the first refusal. A seed where an agent failed at every step has no score, so it no longer counts as ahead of a falling market. The report names the benchmark it read, and on pt-v20 it says the Oracle ran and has no row.

Bar volume

Engine.bars() summed a running total, so a day bar read about two hundred times the day's volume. A bar's volume is now the shares traded in it, at every grain. A day's tick rows now add up to its five-minute bars and its day bar. If you read the tick column as a running total, take its cumulative sum per name and day. Prices and digests don't change. On pt-v20 volume_abs_return_corr moves from 0.508 to 0.596 at one year and from 0.561 to 0.627 at two, and volume_change_acf1 from -0.254 to -0.268 at one year, all still in their bands.

LLM decisions and payload

Decision schema 2 lets an action carry a limit_price, which becomes a tf.Limit, and side: "CANCEL", which becomes a tf.Cancel(). A bad action is refused on its own and the rest of the decision trades. Observation payload 1 is frozen for the 0.8.x line. portfolio.gross_exposure is renamed leverage, portfolio.open_orders lists waiting limit orders, and return_5d covers five full days. A recording carries both schema versions and a replay refuses one made under different versions, so recordings made before 0.8.5 don't replay. LLM adapters and MCP has the contract.

Fingerprint battery version 2

The battery is now seven 120-day markets, one per shipped scenario including curve_shock, so a shock on day 50 has 70 days after it. tf.battery(1) still builds the old six markets. A fingerprint only compares with one taken on the same version.

A digest for a traded run

The known-answer script now hashes one run through tf.evaluate on pt-v20, with the five reference agents and an agent that sends and cancels limit orders. The hash covers every order, fill and scorecard field, and it's checked on all five platforms with the other digests.

Fills reach the market once

Every harness passed an agent's fills to run_session as order_flow, and the session held that flow on every tick. That made one order count 65 times, and agents were marked to their own impact. Fills now arrive once, as fills, and every traded result moves. On pt-v19 the spec mean-reversion rule had beaten buy-and-hold on all 20 suite markets by a median 42 points in 60 days. With fills applied once it reads +0.5 points, ahead in 10 of 20. run_session(order_flow=...) now raises. Pass fills=, or flow_per_tick= for a standing rate. Logs, checkpoints and manifests from 0.8.x replay as they ran. Engine and data has the change.

A book with depth

Seven ModelParams variables give the book depth priced by size, consumption and refill, resting limit orders and a permanent impact per agent. All seven are 0 on every earlier preset, and pt-v20 turns them on. Engine and data says what each does to an order. A resting order fills against the model's flow only at a price inside the maker's quote for that tick. If the maker's re-quote crosses a resting order, the order trades at the maker's price and is recorded as liquidity="taker". tf.Limit and tf.Cancel compare by value, so agree no longer reports two forks that sent identical limit orders as different.

Rate indices

Universe.random(40, seed=1, bonds=True) adds UST2Y, UST10Y and IGCORP, priced off the engine's yield curve. None of them is a real security, and adding them leaves every equity price unchanged. The curve_shock scenario moves the whole curve 200 bp in one day, baselines.Balanced holds a 60/40 book, and cash_interest=True pays idle cash the policy rate. Engine and data has the pricing.

64-bit seeds

A seed can be any 64-bit integer, and every seed below 2**32 gives the market it gave before. Draw a sealed seed with secrets.randbits(64).

Recalibrated crisis scenarios

On pt-v20 recession takes the index down 44.7% at 120 sessions (2008 fell 45%) and now ends, with the cycle at trough on day 365. liquidity_crisis falls 33.9% at worst, as March 2020 did, though more slowly. Scenarios says why.

Gym resets

The gym environment draws a new market on each reset(). A reset without a seed used to replay the constructor's market. info["seed"] names the episode's seed. An action over the leverage cap is scaled down to 1.96x gross, with info["scaled"] set.

Speed

evaluate of five momentum strategies on 40 names over 20 days fell from 11.0 to 3.95 seconds of CPU, and building a pt-v20 engine from 0.70 to about 0.01 seconds.

Python 3.12 and 3.13

Python 3.12 changed sum() over floats, which could move an agent's orders away from 3.11's in the last digit. tradefloor now adds floats left to right on every version, so 3.12 and 3.13 match 3.11.

Short-lag clustering

Volatility clustering reads about a quarter of real at lag 1, so the decay-shape gap now covers abs_return_acf1 and abs_return_acf5 as well as abs_return_acf20, and tf.envelope.check() refuses a question that relies on any of them. How it is measured lists the gaps.

Examples and the MCP server

examples/08-claude-agent.py replays a committed Claude run by default and needs no key or network. With a key and TRADEFLOOR_LIVE_EXAMPLES=1 it calls Claude once per simulated day. The MCP server refuses a stress run that ends before its scenario's first event (it used to return a difference of 0.0), and it refuses strategies named after a baseline. list_scenarios gives each scenario's first and last event day, the constructors are rate_ramp and vix_shock, and pip install "tradefloor[mcp]" now installs pyarrow, which explain needs.

The model written down

The model specification states pt-v20 as equations, each with its source line, and gives every coefficient its value and how it was set. STATISTICS.md names the statistic sets behind every count the site quotes.

Rust crate API changes

The crate takes the package's version number. Cargo treats 0.8.5 as a compatible update to 0.8.1, but it isn't one. Seeds are u64, several public structs have new fields and are #[non_exhaustive], and Engine::new builds a pt-v20 market. Pin tradefloor = "=0.8.1" to stay on the old API. The CHANGELOG lists every changed signature.

Smaller fixes

An agent no longer trades with itself in the settlement book. A resting fill can sit outside the day's high and low, because a bar keeps only each tick's last print. The Scorecard repr prints sharpe=n/a (short run) for a run of fewer than 20 scored days. An act() that returns something other than a mapping is reported on an UNUSABLE line and held in AgentRecord.unusable. The scripts and output that graded pt-v20 are in the library's validation/pt-v20/.

Known gaps

The worst month of the 2020 replay is about 19% milder than the real one. With the VIX held at 65 the market is 5.1 times as volatile as with it held at 5, against 6.2 times in real markets.

0.8.1

2026-09-23

A text-only release, with no coefficient, default or trajectory changes. The README, the messages tf.envelope.check() prints, the scenario target notes and the parameter descriptions now describe pt-v19. Some of them had still quoted older presets. Re-measured on pt-v19, the decay slope of volatility clustering reads -0.515 against a real -0.436, and the memory holds to lag 20. A driven 2020-21 scenario moves prices at about a fifth of the real size, and qe_pe_boost moves nothing on pt-v16 and later.

0.8.0

2026-09-23

pt-v19 became the default. A run that used the default won't replay against 0.7.x. Pass model="pt-v18" to keep the 0.7.x market. pt-v19 meets all fifteen criteria of a long-run check (thirty 21-year histories, and 2008 and 2020 replayed with the real VIX forced in), and pt-v18 meets eight. tf.preset_record("pt-v19")["long_run"] holds every row. Over 21 years it has 1.35 bear markets a decade against a real 1.12, index volatility of 16.6% against 18.1%, and the VIX above 30 on 8.1% of sessions against 8.2%. The 2008 replay falls 41% against the real 57%.

ModelParams.from_preset refuses an override that breaks one of pt-v19's three identities, such as garch_alpha=0.07 alone, because garch_beta follows from it. ModelParams.from_preset_unchecked builds it anyway, and ModelParams.identity_breaks lists what broke.

A recorded transcript names its preset under meta["model_preset"], and replaying it against a different preset raises ReplayMiss. A manifest or checkpoint written under 0.7.x is refused, because its probe simulation runs the default preset. state_hash covers thirteen more fields, and a custom parameter set's fingerprint changes between releases.

Realism is graded on the ruled bands, each read from the longest real record for its row. envelope.certified(), envelope.score() and facts.report() take basis="shipped" for the old table. New calls: Engine.session_news(), Engine.session_tick, Scenario(vix_sets_variance=True) and hold(epicentre=...).

0.7.1

2026-09-08

Documentation only. No behavior changes, and every published digest holds.

0.7.0

2026-09-08

pt-v18 became the default. It's the first default to hold every certified row, and its index returns +5.80 percent a year where pt-v16 lost 13.64. The steady-state crisis lever reads 6.53x against real markets' 6.16x. tf.preset_names() lists the presets the engine resolves, and tf.preset_record reads pt-v1 through pt-v18 from JSON in the wheel.

Every Universe.random roster re-rolls, so pin 0.6.2 for the old draws. EDGAR rosters and rosters you build yourself don't change. A supplied macro_state now survives construction, the first central-bank meeting and OPEC decision fall inside a 252-day year, and to_instruments reads the discount rate from the model. Engine.explain decomposes a name's day down to the draws that seeded it.

0.6.2

2026-09-01

A scenario's liquidity shock now reaches the agent's volume and order cap. World(on_refusal="skip") records an unusable response and carries on. Every adapter takes prior=, a recording consulted before the provider. resample() measures an agent's own noise floor. fetch(ciks=) returns exactly those EDGAR filers, and saved files are the same bytes on Windows.

0.6.1

2026-08-31

tradefloor.integrations gains adapters for the OpenAI Agents SDK, PydanticAI and LangGraph, and one for any plain Python function. A Transcript keys each exchange by a digest of the exact input, so a recorded run replays with no framework installed and no network. The examples in examples/integrations/ run offline. LLM adapters and MCP has the contract.

0.6.0

2026-08-30

pt-v16 became the default. It ties the VIX to the market's realized volatility. tf.Scenario and six packaged scenarios arrived, with the tradefloor scenario command. tradefloor.counterfactual runs one agent in two worlds that differ by one variable, with World, agree() and compare().

0.5.0

2026-08-28

The library was renamed from pretium, which published through 0.4.3 and stays on PyPI and crates.io. A result computed under one of those versions is cited as pretium at that exact version. The rename changes no behavior, and preset names stay pt-v1 through pt-v15. pt-v15 is selectable by name, and pt-v14 remains the default.

0.4.3

2026-08-28

Restores pt-v13 and pt-v14 exactly as they were in 0.4.0 and 0.4.1, after 0.4.2 moved their dollar safe-haven gate. The gate is its own variable now, usd_crisis_vix_threshold.

0.4.2

2026-08-28

Three fixes with no trajectory change: crisis_vix_threshold now gates the dollar's safe-haven drift as well as gold, DayAdvanceOutcome carries a meeting's decision, and daily_credit_floor_gain ships at 0.0.

0.4.1

2026-08-28

No trajectory moves in this release. pt-v13 and pt-v14 now report the 60-day mispricing half-life they run, where tf.model_preset() said 68.26.

0.4.0

2026-08-28

pt-v14 became the default. Over 13 seed blocks it holds the two-year panel fully in band on 11, where pt-v12 held 3, and puts 137 of 138 roster shapes in band.

0.3.0

2026-08-27

pt-v12 became the default, the first preset in band on all 14 statistics over two years as well as one. Results recorded without naming a preset differ from here on, and model="pt-v10" gives the old market.