Coerce ``value`` to a stripped string; ``None`` ⇒ empty.

``sim_abc123def456`` → ``SIM_ABC123``.

Poll for the dispatcher to land ``expected`` rows in the log.

Poll for the log to reach ``expected`` lines (the daemon thread writes on its own schedule). Returns the actual line count.

Check if simulation environment is alive (can receive Interview commands) Args: simulation_id: Simulation ID Re

Check whether the simulation environment is alive Checks the env_status.json file and verifies the PID is still running.

Close file handler and remove from loggers

Factory for NER extraction — a mechanical task that works fine on smaller/faster models. Uses NER_* config when set, otherwise falls back to

Generate embedding for a single text. Returns: Float vector of configured dimensions

Mark task as failed

Fetch new action records from the database with full context information Args: db_path: Database file path last_rowid: Maxim

Intelligently generate complete simulation configuration (step-by-step generation) Args: simulation_id: Simulation ID

Get the agent_id -> entity_name mapping from simulation_config This allows displaying real entity names in actions.jsonl instead of codes li

Return a short summary of belief dynamics.

Get simulation timeline (summarized by round). Scans JSONL directly — no intermediate AgentAction objects.

Attach/refresh the MCP tool catalogue block on an agent's system message.

Attach/refresh the tool-results block on an agent's system message.

Apply :func:`inject_posting_rules` to every agent in *agent_graph*. Returns the number of agents updated. Best-effort — individual agents tha

Return a locale-specific language directive ready to prepend to an LLM prompt. Non-empty output ends with "\\n\\n". Returns "" for unknown locale

All communities for this graph, largest first.

Log to the main logger

Log simulation end

Record ``(sim_id, status)``; return ``True`` only on the first call.

Normalize an EN string for comparison: lowercase, strip punctuation, collapse whitespace. Used to detect legitimate case/whitespace variants.

Read a quoted string starting at s[i] (the quote char itself). Returns (value, end_index_after_closing_quote) or None if not at a quote. Handl

Render the public-gallery cards as an RSS 2.0 XML feed. Mirrors :func:`render_atom` but emits the older format for readers that haven't moved

Run Reddit simulation Args: max_rounds: Maximum simulation rounds (optional, used to truncate long simulations)

r"""Signs up a new user with the provided username, name, and bio. This method prepares a user message comprising the user's details and

Called by Twitter/Reddit loops after each round to publish sentiment.

Append a unified observability event to {sim_dir}/events.jsonl. This is safe to call from the simulation subprocess — no Flask dependency.

Same renderer as :func:`slack_notify.belief_bar`. Re-implemented here rather than imported so a future change to the Slack bar width / glyphs

The bootstrap blob handed to the in-page poller as a JSON literal. Keeps the rendered page meaningful for users with JS disabled (the static

LLM call with retry, includes JSON fix logic

Clean JSON tool call wrappers in Agent responses and extract actual content

Resolve the client IP for rate-limit keying, tolerating proxies.

Coerce a percentage-shaped value to ``float`` in ``[0, 100]``. Returns ``None`` when the input is missing or non-numeric so the caller can de

Static scan of every ``app/api/*.py`` and ``share.py`` for route decorators. Returns fully-qualified path templates (with ``<>``).

Sort key for ``date`` — empty strings sort last (oldest).

Convert Neo4j relationship to the standard edge dict format.

Emit an llm_call observability event (best-effort).

Escape a string for inclusion inside a BibTeX field value. Handles the seven BibTeX specials plus backslash, brace, caret, and tilde. Collaps

Shape a submission failure into the structured response.

Fallback: generate interviews using the LLM directly from agent profiles when the real Wonderwall interview API is unavailable or fai

Fire the outbound webhook + each channel-native notifier for a terminal sim. Each dispatcher is opt-in (no-op when unconfigured), per-channel ded

Find rounds where significant belief shifts occurred.

Convert ``"/api/simulation/<simulation_id>"`` → ``"/api/simulation/{simulation_id}"``. Handles Flask converters like ``<int:round_num>`` and

Match the share card's percentage formatting — integer when the value rounds cleanly, one decimal otherwise. Keeps the thread's numbers aligne

Render a percentage field the way the SPA renders it ("62.0%").

Match the JS ``fmt`` helper so the static SSR markup and the polled JS updates render identical numbers for the same input.

Generate prediction markets from the simulation requirement. Creates *num_markets* well-formed YES/NO questions that agents can trade

Get the latest Interview result from the database

One tweet per dominant-stance flip. Format: ``"Round N: stance shifted to <label>\n<stance line>"``. Stays well under 280 characters even wit

Execute an Interview on a single platform Returns: Dictionary containing the result, or dictionary containing an error

Check if an action is an interview, handling both enum and string.

Check if we're talking to an Ollama server.

Build one ``llm_call`` event line matching the on-disk schema.

Return the ``PROMPTS`` dict from ``locales.<locale>.<module>``. Returns an empty dict if either the module or the dict is missing. Cached aft

Read a token file from disk; missing / corrupt → ``None``. Defensive on every branch — a corrupted token file should make ``resolve_token`` s

Load ``<sim_dir>/state.json`` defensively. Missing/corrupt → ``None``.

Lightweight stand-in for SimulationState — only the attributes the gallery helper reads.

After a round, parse CREATE_POST content for <mcp_call> tags. Dispatched results are queued in ``pending_results[agent_id]`` for injection on

Inject MCP tool catalogue + prior results into tool-enabled agents. Called once per platform per round. No-ops when bridge is None.

Lowercase + validate. Returns ``None`` when the value is empty or not in ``allowed`` so a caller passing ``?consensus=`` (empty) gets the same

Parse tool calls from LLM response Supported formats (by priority): 1. <tool_call>{"name": "tool_name", "parameters": {...}}

Resolve the on-disk Polymarket DB. Handles both the current layout (``<sim_dir>/polymarket_simulation.db``) and the older nested layout (``<si

Population std dev. ``0.0`` for an empty or singleton list. Population (not sample) so a single-delta trajectory reports ``0.0`` rather than

Synchronously POST one Bot API ``sendMessage`` call.

Return the raw, non-empty lines of the log (preserving order). Never raises — a missing or unreadable log is treated as empty so the dispatch

Parse the comma-separated ``SMTP_TO`` env var into addresses.

Read ``DISCORD_WEBHOOK_URL`` at call time. Late binding so a runtime env mutation (or a test ``monkeypatch``) takes effect without re-importi