- Yahoo and Alpha Vantage statements, dated by the period they cover, are withheld from a run dated in the past, with a notice saying why
- statements come from SEC EDGAR by default (fundamental_data "sec_edgar,yfinance"), so a past run reads US filers as filed
- a decision without a rating label is REVIEW; a lone rating word in its prose may be one it argues against ("not a Sell"), so it is no longer read as the call
- yfinance raises its failures instead of answering them with an empty result, so a statement request that timed out is unavailable, not "no data"
- Yahoo answering that it has no such symbol (HTTP 404) stays an answer: no data
- date_window.is_historical replaces the four copies of the rule in the profile guard, Polymarket, the instrument context and the memory cutoff, which read the clock on its own
- the network guard also refuses DNS lookups and curl_cffi requests, the HTTP client yfinance uses, which never touches Python's sockets
- results, the data cache, the memory log and yfinance's time-zone cache live in a directory of the suite's own
- default_config.build_default_config gives the defaults with the environment as it is now; the overlay tests call it instead of reloading the module, whose reloaded DEFAULT_CONFIG leaked their settings into later tests
- the Ollama and API-key tests import the modules they check, which read the environment when called; two tests that only read os.environ are dropped
- the checkpoint signature folds in the run config (provider, models, endpoint, language, vendors, limits), leaving out only file locations, the checkpoint switch and the retry count, so a resume no longer carries reports that other settings produced
- the vendor profile describes the company today, so a run dated earlier is given its name marked as the current one and no sector, industry or exchange
- the language instruction exempts the lines the program reads, the "**Rating**:" line and the transaction proposal, so a decision written in another language keeps a rating that can be read
- after max_tool_rounds rounds of tool calls (default 20, TRADINGAGENTS_MAX_TOOL_ROUNDS) the analyst takes one last turn with no tools, its tool calls and results given as text, and that turn ends its graph
- the three tool-using analysts take their turns through analysts.turn.take_turn
- a limit the recursion limit cannot hold is refused when the graph is built
- backtest --analysts accepts sentiment and is checked against --asset-type before the sweep
- without --analysts, a crypto sweep runs the analysts crypto allows, not fundamentals
- the stock's own sessions set the holding window; the benchmark is valued at its last close by entry and by exit, so a coin held over a weekend is compared with SPY over the same span
- bars meet on their calendar day across time zones, and a missing or non-positive close is dropped rather than scored
- an outcome waits until the benchmark has traded through the exit
- storing a decision and settling outcomes hold a lock across the read, change and write, so runs sharing a log no longer drop each other's entries
- files.locked takes the lock on a .lock file beside the log, across threads and processes
- build_verified_market_snapshot raises NoMarketDataError for an unknown symbol or no row by the date
- get_verified_market_snapshot returns the router's no-data answer, shared as router.no_data_available
- yf_retry raises any failed request as unavailable: an unknown symbol comes back empty, so an exception is a failed request whether or not Yahoo's host answers a probe
- fundamentals, statements, insider transactions, the company profile, news and closes no longer turn a failed request into "no data"
- run_backtest takes a progress(done, total, ticker, date) callback; tradingagents backtest prints [n/total] per cell and ends with the --run-id that continues or settles the sweep
- the nine ticker-taking tools read company_of_interest from graph state, like trade_date; the model no longer supplies a symbol (it had passed indicator names such as RSI, a real ticker)
- the News Analyst prompt and the tool descriptions no longer ask for a ticker
- record_decision writes the run's state log, so CLI runs save full_states_log like propagate() does
- with both round counts from the environment, the CLI no longer says a chosen research depth was overridden
- an unknown --analysts name lists sentiment, not social
- each analyst is a graph of its own (model and tools on a private message history) that returns only its report; all start together and the research debate waits for every report
- the message-clearing nodes are gone; a checkpoint saved by the sequential layout starts fresh
- TradingAgentsGraph.stream_run streams the analysts' messages for debug mode and the CLI, whose status and timing now track the analysts side by side
- the OHLCV cache and the SEC EDGAR cache go through dataflows.files.replace_file: a reader sees the old file or the new one, and concurrent writers never share a temp file
- --ticker, --date and --analysts answer their steps, checked as the prompts check them; --save/--no-save and --show/--no-show answer the questions after the run
- with no terminal, the run stops before it starts and names every flag or TRADINGAGENTS_* variable still needed; a missing API key and an announcement no longer wait for input
- README shows an unattended run
- TradingAgentsGraph.run_settings(): version, provider, models, analysts, debate rounds, language and vendors; no endpoints, keys or paths
- complete_report.md opens with the analysis date and those settings; the saved state log carries them as run_settings
- pyproject reads the version from tradingagents/__init__.py, so a source checkout, an editable install and a wheel report the same one
- the SEC EDGAR User-Agent names tradingagents.__version__
- the Research Manager's plan and the Portfolio Manager's decision live only in investment_plan and final_trade_decision; the debate states no longer copy them as judge_decision
- the saved state log writes the Trader's plan as trader_investment_plan, its state key
- the report tree, the final display and the live CLI read the decisions from those fields
- state field descriptions name the Research Manager, the News and Fundamentals Analysts and the Portfolio Manager; comments that only repeat them are gone
- the debate-limit comments count turns per max_debate_rounds and max_risk_discuss_rounds
- date_window's docstring covers all the point-in-time rules it holds
- vendor functions and date_window take as_of_date, the date data is served as of; the model-facing tool arguments are unchanged
- build_instrument_context and resolve_instrument_context take trade_date, which is what they receive
- docstrings, comments, messages and the README section use "memory log", matching TradingMemoryLog and memory_log_path
- the write, read and settle paths are named for what they do, not by design phase
- memory/log.py holds TradingMemoryLog, still imported from tradingagents.memory
- graph/settlement.py and graph/reflection.py move to memory/; Reflector leaves tradingagents.graph's exports
- the first line opening with a rating label is the call, not a rating the text quotes later ("Consensus rating: Buy")
- "rating" must start a word, so "Operating margin: Sell-side" is not a label
- the typed rating is the state's final_rating; propagate returns it, and the memory log tag, the state log and the CLI review check read it
- the decision text is parsed only when the Portfolio Manager answered in free text
- TradingAgentsGraph.process_signal is removed
Yijia-Xiao