Launched this week
Tidewatch
Scores obligations so agents know what to work on next.
2 followers
Scores obligations so agents know what to work on next.
2 followers
Tidewatch is a Python scoring engine that tells agents and humans which obligation to work on next. Six math-driven factors — deadline decay, materiality, dependency fanout, completion state, timing sensitivity, violation history — produce a single pressure score per task. Unlike binary overdue/not-overdue systems, pressure accumulates continuously as deadlines approach. Zero dependencies, 659 tests, 97% coverage. Scores 500K items in <5 seconds. Apache-2.0 + commercial dual license.
A bit more on the technical design for those interested:
The key idea is deferred scalarization. Most scoring systems immediately collapse multiple factors into a single number. Tidewatch preserves all six dimensions in a component space — the product collapse only happens when a consumer explicitly requests a scalar. Until then, you can do Pareto comparison across obligations, inspect individual factors, or apply custom weighting for your domain.
Why this matters for agents: an LLM-based agent dispatching work can ask "show me everything in the red zone where dependency fanout is the dominant factor" — meaning tasks whose delay will cascade. That query is impossible against a pre-collapsed score.
On the honest trade-off with EDF: Earliest Deadline First is theoretically optimal for pure deadline minimization. Tidewatch pays ~8.6% more missed deadlines at N=200 compared to EDF. That is the cost of encoding materiality and dependency structure alongside time pressure. The return is that the right deadlines get met — a filed legal brief matters more than an updated README, even if the README is due sooner. The evaluation (Monte Carlo, 200 trials, seed=42, LogNormal durations) is fully reproducible from the benchmarks/ directory.
Integration is lightweight — Tidewatch is a scoring engine, not a platform. Feed it obligations from JIRA, ServiceNow, GitHub Issues, PagerDuty, or a custom DB; it returns ranked scores. See examples/ in the repo for ready-to-use patterns including a zero-dep REST API server.