DOOP

Local-first training agent

Reverse engineering a wearable into a safer training agent.

A method-and-evidence project built from retained packets, real hardware failures, and explicit decision gates.

Case study GitHub
An unbranded biometric wrist strap on a graphite engineering workbench
Raw-first every interpretation can be re-derived
CRC-checked framing and payload integrity verified
Replay-safe commit-before-ACK history recovery
Data-free CI synthetic tests require no wearable

What made the result credible

The project treated every decode as a hypothesis, every failure as evidence, and every coaching value as unsafe until its inputs could support a decision.

Retain the raw stream

Store each frame before parsing so later corrections remain exact and auditable.

Test on owned hardware

Verify framing, commands, timing, and replay behavior against the actual band.

Separate compute from trust

Keep derived values for audit while withholding them from coaching when evidence is weak.

Publish the failures

Preserve negative results and two agent promotion revocations instead of smoothing them over.

A convincing number can still be unsafe.

An apparently plausible recovery scalar survived computation on a severely incomplete day. Doop stores derived values for audit but withholds them from coaching until coverage, sleep, and baseline gates all pass.

Follow the static trace

A ten-minute reviewer path

Each headline claim points to one artifact and one check a reviewer can run independently.

Start here Claim, evidence, command

The shortest path through the protocol, metric, and agent claims.

 Recorded results Hardware ledger

What was proven privately and what the public suite reproduces.

 Agent behavior Acceptance record

Visible outputs, forbidden disclosures, and both promotion revocations.

 Release index Evidence manifest

Traceable claims, commands, résumé language, and release artifacts.

The limits are part of the work.

This is a method-and-evidence repository, not a medical device or a replacement for a commercial health platform.