iotwerk_device_auth

Canonicalizes device_id, nonce and message, computes an HMAC-SHA256 tag with caller-supplied ephemeral key material, and optionally compares a claimed tag in constant time. It does not provision devices, store keys or replace asymmetric device identity; do not submit long-lived production secrets.

Applies a transparent heuristic: cycle fade scales with equivalent full cycles and depth-of-discharge exponent; calendar fade scales with square-root time and a temperature factor. Coefficients are caller supplied or declared defaults. This is not cell-chemistry validation, warranty advice or a safety prognosis.

Computes cohort median, median absolute deviation and robust z-scores for one numeric reading per device. A caller-selected threshold determines flags. It detects cross-sectional outliers only and does not diagnose root cause or learn temporal behavior.

Computes free-space loss at one metre, applies a caller-supplied path-loss exponent over distance, then reports estimated receive power and margin against receiver sensitivity. This planning estimate does not model terrain, Fresnel obstruction, interference, antenna mismatch or regulatory duty cycle.

Canonicalizes firmware hash, version, device model and minimum bootloader version, then produces a SHA-256 manifest digest and HMAC-SHA256 release tag using caller-supplied ephemeral key material. This is a symmetric integrity tag, not a vendor PKI signature or code-signing service.

Fits reference = slope × observed + offset with ordinary least squares over paired caller-supplied readings. It reports residual diagnostics and does not claim traceability, uncertainty accreditation or suitability beyond the supplied calibration range.

Groups caller-supplied scalar field changes by field name, selects the highest version, and reports a conflict when equal-version values differ. It does not connect to a twin platform, merge nested documents or resolve semantic conflicts.

Computes a deterministic shortest path over caller-supplied directed or bidirectional links with non-negative costs. It does not measure live radio conditions, reserve capacity or push routes to devices.

Filters caller-supplied bids by CPU, memory and latency constraints, computes workload cost from quoted unit prices, and ranks feasible candidates by total cost then latency. It does not contact providers, attest hardware or place a bid.

Computes the minimum Euclidean distance from each circular obstacle center to every path segment, then subtracts obstacle radius and robot radius. It detects planar static clearance only; it does not model robot shape beyond a circle, dynamics, uncertainty or moving obstacles.