For treasury researchers, risk teams and agent builders

Bring funding evidence into your daily research.

Add a source-linked evidence packet to the workbook, research job or agent you already use. Keep the original response, its dates and its gaps. Start with one recurring question.

Open the verified funding review Run a research capture Try public research in your browser Scope an integration pilot

Free public research. No account, API key, LLM or broker connection required. Example runner 1.0.0 uses the separately released Financial Evidence 0.1.5 package.

One recurring job

Choose the evidence your workflow needs.

Morning funding review

For: treasury and macro researchers.

Retain Seiche's public money-market context with source-reported clocks. Review funding inputs and missing evidence before writing the desk note.

--workflow funding

Method and coverage

Bank review context

For: counterparty and credit research teams.

Retain LiquiLens's public board alongside the separate funding context. This job gives board-level context. Use the institution recipe below for a named bank's filing review.

--workflow bank-context

Named institution review recipe

Liquidity research context

For: execution researchers and agent builders.

Retain Undertow's public summary and separate Seiche funding context. Position-specific depth and exit estimates use Undertow's dedicated research tools.

--workflow liquidity-context

Position-size research tools

For China research, Palimpsest supplies separately scoped information-state and economic evidence. For offline verification, use Evidence Carrier with an actual supported carrier. The packets produced here are ordinary research captures.

From first result to a repeatable job

Keep one research run.

Download and inspect evidence_job.py. Use Python 3.10 or newer and Git. Install the exact source revision of Financial Evidence 0.1.5 in a project environment:

python3 -m venv .venv
.venv/bin/python -m pip install 'git+https://github.com/beepboop2025/financial-evidence-skills.git@f12a4889180f1a6d1b76324ebe62659c97865eb5'
.venv/bin/python evidence_job.py --workflow funding --output captures/first

On Windows, use .venv\Scripts\python.exe in place of .venv/bin/python. The package is pinned; review the downloaded runner before executing it. This example does not change your existing agent configuration.

File	Use it for
`packet.json`	The complete returned documents, including observations, units, limitations and failed-source records.
`sources.csv`	A spreadsheet index of source URLs, retrieval times, source-reported clocks and transport failures. Full financial observations remain in the JSON.
`run.json`	Workflow identity, exit code and hashes of both saved files. Written last; a directory without this file is an incomplete local write.

The runner refuses to overwrite a capture. On a later research day, run the same job into captures/second. Each source is requested once, with a 1 MiB response limit and a 15-second socket timeout. There are no automatic retries. A scheduler should also set a total job deadline.

Read the result: exit 0 means all retrievals succeeded; 1 means partial retrieval; 2 means none succeeded; 3 means a local job error. Successful retrieval can contain stale, restricted or unavailable evidence. Source validity, freshness and Carrier verification remain unassessed. A hash records file integrity, not a signature or trading permission.

Operator checks must add --verification. The runner marks those requests synthetic and records that fact locally. Legacy upstream logs may still lack reliable attribution; do not count these checks as users.

Use your own scheduler

Download the GitHub Actions template. Commit the reviewed runner at scripts/evidence_job.py and the template at .github/workflows/research-evidence.yml in your own repository. Start it manually. The weekday schedule is commented out until you choose to enable it.

The template retains artifacts for 14 days, including partial results, and fails the job when retrieval fails. Prefer a private repository for research captures and follow upstream data rights. GitHub quotas, billing and scheduler delays apply. Local execution has been checked; a customer-owned GitHub schedule needs its own acceptance run.

Use your existing tools

One evidence layer, several delivery routes.

Where you work	Connection	First useful result
Agent framework or MCP client	`https://liquilens.in/mcp/financial-evidence`	List topics, route `money-market`, then fetch that topic. Keep source-reported metadata and complete documents.
Excel or SQL notebook	Power Query · DuckDB views	Read a funding document beside your existing licensed data. The CSV above is an alternative for a captured run.
OpenBB	Existing router extension	Call `obb.financial_evidence.routes()` before fetching a selected topic.
Hermes, OpenClaw or n8n	Agent kit and tested setup guides	Funding, exact covered-bank evidence or position-size exit research through dedicated tools.

These are available connection routes. They do not imply vendor endorsement, marketplace acceptance or access to Bloomberg, Reuters or LSEG content.

Prove a useful dependency

Test one workflow over ten research days.

Start with a free self-serve evaluation. Agree one recurring research question, its coverage and the reviewer who will judge it. Compare the existing process with the evidence-assisted process using the same question and sources.

First result: retain a packet and have a reviewer confirm that it answers a real research task within the published scope.
Repeat use: run the same workflow on at least three distinct research days and record whether its output was used.
Outcome: compare review time and omissions against the manual baseline. Record failures and cases where coverage was insufficient.

Download the private pilot scorecard. Keep it in your environment. No usage telemetry or organization identifier is sent by this example. Share only an agreed, redacted outcome if you want integration help.

Discuss an integration pilot Report a public integration blocker

Use the private email route for confidential requirements. Public issues should contain only redacted technical examples. Dedicated capacity, support terms and any commercial agreement require separate scoping; this page promises no SLA or automatic paid enrollment.