Canon
Document refinery and canonicalization — turns raw material into structured, schema-conformant knowledge.
In the constellation
Canon highlighted in the live map — hover or tap any node to explore.
Why it exists
Knowledge arrives messy. Years of documents, code, and notes pile up duplicated, inconsistently structured, and shaped a dozen different ways — too disordered to query, too large to reorganize by hand. Every attempt to clean it up stalls because the work is enormous and the rules live in someone’s head.
Reorganizing a corpus by hand doesn’t scale, and a blind find-and-replace breaks more than it fixes. You need structure imposed reliably, at scale, without quietly corrupting what was already right.
What Canon is
Canon is a document refinery. It takes raw or messy material and drives it toward a clean, structured form — the right folders, consistent metadata, deduplicated content — defined by a declarative template you choose, not a shape baked into the tool. It both refines raw documents from scratch and re-conforms an existing corpus to the standard.
Most of the work is deterministic structuring: matching folder rules, normalizing metadata, planning the moves. AI assists only where genuine judgment is needed. And Canon never moves bytes itself — it hands a precise plan to the integrity engine, which applies it atomically so not a single reference breaks.
What it does
Pluggable structure
The target shape is data, not a hardcoded schema — supply a template and Canon conforms any corpus to it. Refine raw material or re-align an existing tree with the same template.
Audit before you change
Run a read-only diff of any corpus against the standard to see exactly where it drifts — before a single file moves.
Reference-safe restructuring
Canon decides the shape and emits a plan; the integrity engine executes it atomically, rewriting every pointer so links never dangle.
Honest gap reports
When a case needs human judgment, Canon doesn’t guess — it emits a gap report naming exactly what it couldn’t decide, one file per case.
The line between judgment and machinery
AccelMars draws one hard line through every product: what an AI decides, and what runs deterministically. Canon sits on the boundary — and keeps the two honest.
Where AI assists
- Promoting a file into its own folder
- Inferring type for ambiguous content
- Extracting structure from raw prose
What stays deterministic
- Template matching & metadata normalization
- Planning the structural moves
- Atomic, reference-safe execution
Canon is structuring first, judgment second — the bulk of the work is deterministic, and the few cases that need a human are surfaced as explicit gap reports rather than silent guesses.
Live — the refinery and the audit / align / template surface both run today. The unified Platform API is the next step.
Open source (Apache-2.0). The template format is a public spec and the mechanical engine is on GitHub — author your own templates, run it on your own corpus. The AccelMars-tuned judgment layer is the closed companion.