tpt-sensetel

Sensing with purpose — an edge-native intelligence layer that decides what data is worth capturing, compressing, transmitting, and storing.

tpt-sensetel replaces "record everything at maximum fidelity" with purpose-driven, semantic compression:

• Capture at minimal fidelity by default
• Run edge AI to detect meaningful events
• Increase fidelity only when semantics demand it
• Transmit metadata first, raw media on demand
• Formally verify the dispatch logic; empirically measure detection accuracy — and never blur the two

tpt-sensetel is the brain that makes capture decisions. tpt-kinetix is the muscle that compresses and transmits. tpt-telos is the guarantor that the decision-making logic is mathematically correct — not that every real-world event is detected.

Two guarantee tiers, kept deliberately separate

• Formally verified (tier 1): dispatch-logic properties proven at compile time or enforced at runtime — e.g. "if a trigger fires, recording starts within budget" and "fidelity never drops below the critical floor during an active event." These hold regardless of what any model outputs.
• Empirically measured (tier 2): detection accuracy (recall/precision on a labeled validation set), tracked per model version in model cards and monitored in production. Never described as "verified" or "guaranteed."

This isn't a hedge — it's the difference between an honest safety story and a liability.

Workspace

Crate	Role	Phase
tpt-sensetel-core	Core types, errors, decisions, fidelity	1
tpt-sensetel-verify	Formal verification of dispatch logic	1
tpt-sensetel-triggers	Event detection engine (motion, composition)	1–2
tpt-sensetel-policy	Policy engine and DSL	1
tpt-sensetel-metadata	Semantic extraction	3
tpt-sensetel-adaptive	Dynamic fidelity control	3
tpt-sensetel-edge	On-device pipeline	4
tpt-sensetel-api	Control plane (HTTP/gRPC/WebSocket)	5

Phase 1 is in progress; later crates are scaffolded with their design lead documented in spec.txt. Track progress in todo.md.

Getting started

cargo build --workspace
cargo test --workspace
cargo clippy --workspace --all-targets
cargo fmt --all -- --check

Documentation

• spec.txt — full design specification, module contracts, and implementation roadmap
• CONTRIBUTING.md — how to contribute
• The distinction between formally verified dispatch logic and empirically measured detection accuracy is the single most important idea in this project. Read it in spec.txt before writing a single unit test.

License

Dual licensed under either of MIT OR Apache-2.0, at your option. © TPT Solutions.