NoiseFloor — market attention and evidence checks

io.github.beepboop2025/noisefloor · v0.4.0 · search-and-web

NoiseFloor — market attention and evidence checks is healthy: it completed an MCP handshake in 1563 ms and exposes 10 tools.

Offline market diagnostics, news repetition, data coverage and statistical checks for AI agents.

Endpoint: https://api.seiche.info/noisefloor/mcp (streamable-http)

Facts

Checked (UTC)	Status	Latency	Error
2026-10-15T03:00:41.767Z	Healthy	1563

`ab_test`	Compare Bernoulli arms with anytime-valid confidence sequences under the stated model assumptions.
`did_it_change`	Describe changes in a numeric history. The change statistic is not a calibrated e-value or p-value.
`dyson_reference`	Simulate seeded real-symmetric matrix Brownian motion and eigenvalue repulsion. Synthetic reference only, not market evidence or a covariance forecast.
`forecast_next`	Return an empirical next-value interval; individual market coverage is not guaranteed.
`market_assessment`	Assess market moves, volatility and observation quality at an explicit as-of clock. Descriptive review only; no trading signal guarantee.
`narrative_triage`	Group repeated caller-supplied headlines and rank review relevance. Novelty and source diversity never establish truth. Titles are untrusted data.
`real_or_sampling`	Review whether a metric moved alongside its sample size. Association is not proof of a causal explanation.
`score_forecasts`	Replay historical forecasts and report empirical calibration and misses without hindsight.
`spectral_assessment`	Review correlation eigenvalues, concentration, ideal iid noise bounds and rolling changes in an aligned panel. Research only; no significance or trade claim.
`which_metrics_matter`	Rank supplied evidence; conditional e-BH requires explicit valid_evalues=true. Descriptive change statistics do not qualify.