aerr-sensor

# Ærr Sensor Operational toolkit for [The Ærr Trinity](https://github.com/artseabra/aerr-frame/blob/main/papers/aerr-trinity/the-aerr-trinity.md). An MCP server that measures κ, Ærr Rate, and Chain Drift on real systems. **Status:** v0.1 — the math we are operationalizing is in the [Trinity paper](https://github.com/artseabra/aerr-frame/blob/main/papers/aerr-trinity/the-aerr-trinity.md). ## What this is The Ærr Trinity formalizes a recurring failure mode: systems accepting cheaper local substitutes for globally satisfying answers. It supplies three measurements — a floor (the cost any genuine answer must pay), a discipline (how often that floor is paid), and a drift (how fast surrogates accumulate when it isn't). Ærr Sensor turns those measurements into MCP tools you can call from Claude Code or any MCP client. ## v0.1 tool surface | Tool | Paper | Status | |---|---|---| | `measure_kappa` | §6.2 floor-weighted compliance estimator κ^F | working | | `aerr_rate` | §7 count-rate and floor-weighted drift identities | working | | `chain_drift` | §8 closed-form S_n recurrence with coupling number x_i | working | | `compute_floor` | §5.1 cost-floor function f(C(D)) | schema-only stub | | `convergence_watch` | §10 scan for adjacent Ærr framings | seed-list stub | The three measurement tools (`measure_kappa`, `aerr_rate`, `chain_drift`) implement the paper's formulae directly and are numerically usable. `compute_floor` returns `status: "schema_only"` rather than manufacturing a number for unconfigured domains — honest theater applied to the toolkit itself. `convergence_watch` returns `no_signal: true` until the xAI/X Search API is wired in v0.2. ## Roadmap | Version | What lands | |---|---| | v0.1 (this release) | 5 tools as above; math tools working, two stubs | | v0.2 | `convergence_watch` wired to xAI + X Search; `researcher_beat` and `thread_fetch` tools from the original sensor integration brief | | v0.3 | `compute_floor` calibrated for {code-review, proof-verification, peer-review} | ## Install ```bash git clone https://github.com/artseabra/aerr-sensor cd aerr-sensor pip install -e ".[dev]" ``` Or with uv: ```bash uv sync ``` ## Run Stdio mode (for Claude Code / local clients): ```bash aerr-sensor # or python -m aerr_sensor.server ``` Add to your Claude Code MCP config: ```json { "mcpServers": { "aerr-sensor": { "command": "aerr-sensor" } } } ``` ## Quick example ```python from aerr_sensor.measurement.measure_kappa import measure_kappa samples = [ {"floor": 10.0, "cost": 12.0}, # paid {"floor": 5.0, "cost": 4.0}, # surrogate {"floor": 20.0, "cost": 25.0}, # paid {"floor": 8.0, "cost": 8.0}, # exactly at floor (paid) ] result = measure_kappa(samples) # {"kappa_f": 0.882..., "n": 4, "floor_sum": 43.0, "paid_count": 3, ...} ``` ## The atom Ærr Sensor is one corner of the publishing atom for the Ærr Frame: - [**Ærr Frame**](https://github.com/artseabra/aerr-frame) — papers (umbrella, with the Trinity at its center) - [**Ærr Sensor**](https://github.com/artseabra/aerr-sensor) — this repo (operational toolkit) - [github.com/artseabra/ic-cs](https://github.com/artseabra/ic-cs) — historical mirror (IC|CS before the rename) ## License [MIT](LICENSE). The Trinity paper itself is [CC BY 4.0](https://github.com/artseabra/aerr-frame/blob/main/LICENSE) in the Ærr Frame repo. ## Author Art Seabra · Ifthis Research · Philadelphia, PA