Created: 2026-07-01 Last revised: 2026-07-01 Status: Canonical Owner: Sailesh (human-gated) Supersedes: — (first version) Purpose: The shared operating rules by which DTA concepts are captured, matured, tagged and governed in Fabric — the common understanding that both ChatGPT and Claude must follow.
Fabric is the authoritative knowledge base. The pipeline is:
Fabric (authoritative KB) → ChatGPT (authoring / synthesis) + Claude (challenge / cross-check / governance) → GitHub / Ultim8 (executable architecture, prototype) → Craft / website / PowerPoint / LinkedIn (communication channels).
ChatGPT is the primary authoring layer.
Claude is the challenge, cross-check and governance layer — not a second author.
Both assistants search Fabric first before creating anything new.
Every canonical concept carries this block. Status and Confidence are orthogonal — one is editorial standing, the other is epistemic.
Field | Values / rule |
|---|---|
Canonical name | One name per concept. The note title and its internal H1 must match. |
Created | ISO 8601 (YYYY-MM-DD) |
Last revised | ISO 8601 (YYYY-MM-DD) |
Status | Idea → Hypothesis → Research-backed → Canonical → Deprecated (editorial standing: "is this our official line?") |
Confidence | Low / Medium / High (epistemic: "how sure are we it is right?") |
Evidence strength | Weak / Moderate / Strong — with linked evidence items, never a raw count |
Supersedes / Superseded by | Links to prior/newer concept versions |
Related concepts | Links (the knowledge-graph edges) |
Provenance | Distinguish original DTA thinking from external supporting evidence |
A concept may legitimately be Canonical + Medium confidence (publicly committed while evidence is still maturing).
Tags carry discovery only. Status, confidence, dates and provenance live in the metadata block above — never duplicated as tags. Use prefixed facets (Fabric substring search filters these cleanly):
domain: — payments, banking, architecture, …
theme: — market-positioning, transformation-intelligence, knowledge-intelligence, …
channel: — website, presentation, prototype, observatory, …
type: — concept, research, evidence, draft
Reuse before you create. Search existing tags first; do not mint near-duplicates (the store is already at ~80 tags).
Canonical promotion is human-gated. Only Sailesh sets Status: Canonical. Assistants may mark Research-backed and flag "proposed canonical" — they never self-promote.
Search-first, supersede-don't-duplicate. Before creating a concept, search Fabric. If it exists, revise and link via Supersedes — do not create a parallel note.
One canonical name per concept. Fabric title = internal H1 = the name used in conversation. (This charter exists partly because one note recently carried three different names.)
ISO dates everywhere. YYYY-MM-DD in titles, prefixes and metadata — no ambiguous DD/MM vs MM/DD.
Drift detection on revision. When newer thinking diverges from a canonical concept, either update canonical (human-gated) or log it as a new Hypothesis linked by Supersedes — never silently overwrite.
This charter is itself a canonical concept and follows its own rules. Changes bump the version (v1.0 → v1.1 …), update Last revised, and record Supersedes.