All release walkthroughs

LIQUILENS · PLUGIN REVIEW EVIDENCE

Seiche Market Atlas Corpus

Five public research workflows, shown with actual captured MCP requests and results.

90 seconds · edited browser recording · no audio · no model API calls for this recording.

Inspect the interactive capture viewer · Case transcript · Recording manifest · Underlying service captures

What this demonstrates

The video shows direct calls to the public MCP service and preserves returned source dates, restrictions and unavailable states. It is not a native plugin-installation recording or a new model-routing test. Negative model cases remain unexecuted in this capture.

The product skill and negative model-routing cases are review expectations, not newly executed model tests. No trading, enrollment, collection, private-record access or execution was performed. NoiseFloor examples use synthetic inputs; all other values retain their service-provided dates and restrictions.

Cases shown

  1. Corpus discovery

    Summarize registered market corpora and evidence classes so I can choose a research source.

    Interpretation boundary: Describe returned registrations and evidence classes. Do not infer observation availability, freshness or redistribution permission.

  2. Dataset rights inspection

    List up to ten versioned datasets, choose one returned dataset and inspect rights, publication state and product eligibility using its exact ID.

    Interpretation boundary: Use list limit 10, copy exact dataset_id into inspect_dataset and retain restrictions. If no dataset is returned, stop without fabricating an ID.

  3. Normalized market coverage

    Show normalized Seiche money-market coverage and identify exact market and source-adapter partitions for bounded queries.

    Interpretation boundary: Report returned coverage and exact identifiers. Distinguish source-adapter IDs from publisher labels and retain gaps or restrictions.

  4. Permitted bounded observations

    Discover a Seiche market and exact source-adapter partition, then request no more than twenty permitted canonical observations. Preserve source clocks, units and restrictions.

    Interpretation boundary: Copy market and source identifiers from discovery and use limit 20. Respect redistribution validation; report denial or missing coverage without reconstructing withheld values.

  5. Corpus service health

    Check the corpus service's reported health and explain unavailable storage or service state separately from dataset freshness and licensing.

    Interpretation boundary: Report health state and limitations. A healthy service does not establish current data, complete coverage or redistribution authorization.

Endpoint: https://agents.seiche.info/products/seiche-market-corpus/mcp

Video SHA-256: 91e81c501054472d66782c1bed9a2f1e22e82bc8aa3b72684e3f69f8c3f3e41d

Draft review evidence. This page is not an OpenAI approval or publication receipt.