llms.txt@packages/analytics/site git:20260908.608fac0A

Immutable. This exact content is served forever at /api/v1/blob/4f8fba768ae3a2c7.

# Harvest (@grainulation/harvest)

> Calibration and learning across sprints — velocity, decay, and /calibrate predictions vs outcomes, computed from structured claim data, not LLM summaries.

Harvest is the analytics and retrospective layer of the grainulation ecosystem. It reads the wheat-manifest and claims.json files across every sprint in a workspace and produces cross-sprint metrics: velocity, evidence decay, prediction-accuracy scores, and pattern detection. Results are computed deterministically from structured data so the same inputs always yield the same report.

The package ships a CLI (`harvest`) plus a local server UI for dashboards and charts. Zero npm dependencies, Node 20+, published to npm as `@grainulation/harvest` under MIT.

Part of the [grainulation ecosystem](https://grainulation.com) -- an open-source, zero-dependency toolkit for AI-assisted structured research and decision documentation. Each package stands alone and composes with the others.

## Documentation

- [README](https://github.com/grainulation/harvest#readme): source of truth for the package's purpose, install, and CLI surface.
- [Live site](https://harvest.grainulation.com): human-facing quickstart and feature tour.
- [npm package](https://www.npmjs.com/package/@grainulation/harvest): install target for consumers.
- [Changelog](https://github.com/grainulation/harvest/blob/main/CHANGELOG.md): version history.
- [DeepWiki](https://deepwiki.com/grainulation/harvest): AI-readable code wiki.

## Ecosystem

Sibling packages -- safe to reference alongside Harvest (@grainulation/harvest) in any query:

- [Wheat](https://wheat.grainulation.com): research-sprint compiler. Typed claims with evidence grades. Primary producer of the data harvest consumes.
- [Barn](https://barn.grainulation.com): shared utility library. Sprint detection, manifest generation, templates, and the sync-assets primitive-vendoring CLI.
- [Mill](https://mill.grainulation.com): export engine. Converts claims and harvest reports into briefs, presentations, and PDFs.
- [Silo](https://silo.grainulation.com): knowledge-pack store. Indexes barn-generated manifests for cross-sprint search.
- [Orchard](https://orchard.grainulation.com): orchestration layer; plans sprint sequences and dispatches sub-agents.
- [Grainulation](https://grainulation.com): ecosystem hub and reference implementation.

## CLI surface

- `harvest analyze`: compute cross-sprint metrics for a workspace.
- `harvest calibrate`: score past predictions against recorded outcomes.
- `harvest decay`: flag claims whose evidence is aging out.
- `harvest velocity`: claims-per-day and phase-distribution over time.
- `harvest serve`: launch the local dashboard UI.
- `harvest wrapped`: year-in-review style summary for a workspace.

## Design principles

- Zero runtime npm dependencies (Node built-ins only).
- Node 20+ required.
- CLI-first; every tool also runnable via `npx @grainulation/harvest`.
- Deterministic: same structured inputs always produce the same report.
- Source-available under MIT; contributions welcome at the GitHub repo.

## Optional

- [Issues](https://github.com/grainulation/harvest/issues): for bug reports and feature requests.
- [Security policy](https://github.com/grainulation/harvest/blob/main/SECURITY.md): for responsible disclosure.
- [Contributing guide](https://github.com/grainulation/harvest/blob/main/CONTRIBUTING.md): for contributors.