Skip to content

Welsh translation pending

A Welsh-language version of this page is being prepared during the Tuesday tutor session (week of 2026-05-19). Until then, the Welsh-language version of the site shows the placeholder.

The framework

Awen Weave is built from five named parts. Each has a specific job; together they form the loom on which knowledge is woven.

Llys — the interaction and decision layer

In Welsh, Llys means court. The Llys is where judgement happens visibly. Curators interact with the system here. Decisions about what to accept into the canonical record, what to reject, what to flag as contested — all of these happen in the Llys.

The Llys principle: no decision is invisible. Every acceptance of a claim, every rejection of a proposal, every override of an auto-derivation is recorded with the curator's identity, the timestamp, the data state at the moment of decision, and the reasoning. Audit reconstruction is always possible.

Craidd — the place-based trust core

In Welsh, Craidd means core. The Craidd is where canonical knowledge lives. It is curated; it is provenance-bound; it has explicit boundaries.

Three load-bearing decisions shape how Craidd holds knowledge:

Records are claims, not values. Every statement in Craidd is a claim made by someone, citing something, at a particular time, with a particular confidence. Multiple competing claims about the same subject coexist; a canonical view is materialised from them, but it never silences the competing claims. Contradictions are visible, not hidden.

Sources are themselves entities. A claim cites a source; the source is itself a recorded entity in Craidd with its own provenance and visibility rules. There is no opaque "trust me" anywhere; every layer of trust is examined.

The place is the trust anchor. Each Craidd is hosted physically in its place — the Dolgellau Town Dataset lives on a Raspberry Pi at Arloesi Dolgellau, in Dolgellau. The relationship between the data and the place it describes is not abstract. The trust is grounded.

IDRIS — reasoning and orchestration

IDRIS is named for Cadair Idris, the mountain south of Dolgellau. The reasoning layer sits above Craidd and Prawf; it answers questions, synthesises across sources, sequences operations. It is the layer that brings together what Craidd holds and what Prawf records into useful answers.

IDRIS is bounded: it reasons over what is in Craidd; it does not invent. Where Craidd is silent, IDRIS is silent. Where Craidd is contested, IDRIS surfaces the contradiction rather than choosing one side.

Prawf — the obligation and proof layer

In Welsh, Prawf means proof or test. Prawf records the evidence of process, not the correctness of outcome. Every decision the system makes — every accepted claim, every superseded claim, every retired entity, every co-signed field correction — generates a Prawf entry.

Prawf is append-only and hash-chained. Each entry includes the previous entry's hash and its own hash, so the chain is tamper-evident. Reconstruction of state at any past moment is possible by walking the chain. The principle: what was done is honoured by being recorded.

Craffter — pattern-level observation

In Welsh, Craffter means perceptiveness — the quality of one who is observant. Craffter notices patterns across the data: trends, anomalies, clusters of claims that fit together in unexpected ways. It is the machine-learning layer where one exists.

Craffter has one inviolable boundary: it is advisory, not authoritative. It never writes to Craidd; it never writes to Prawf. It surfaces observations to humans, who decide whether the observation rises to a claim worth recording. The principle: pattern recognition supports human judgement; it does not replace it.

Why these five

The five layers map cleanly to the questions any honest knowledge system must answer:

  • How is judgement made? — Llys.
  • What is known? — Craidd.
  • How is what is known reasoned about? — IDRIS.
  • What is the record of what was done and why? — Prawf.
  • What patterns are observable, and what do they suggest? — Craffter.

Conflating these questions — letting reasoning rewrite the record, or letting pattern observation silently update canonical state — is what produces opaque AI systems whose judgements cannot be traced or overridden. Awen Weave separates the questions so each can be answered honestly.

How the pattern applies

This five-layer architecture is the pattern. Every instance of Awen Weave — Dolgellau Town Dataset today, others tomorrow — implements the same five layers, tuned to its domain.