---
description: >-
  Borrowed from software engineering: ADRs, GitHub Spec Kit's spec-driven development, and supply-chain provenance — each re-aimed from code to intent for SaaS delivery.
---

# Borrowed from software engineering

Rosetta did not invent its instruments. It borrowed some of software
engineering's best ideas — deliberately, with attribution — and then
customized them, because SaaS delivery breaks the assumption those
ideas grew up with.

The assumption is that **source is canonical**. In classical software
engineering, the code is the truth: check it out, build it, and you
can reproduce the system anywhere. A SaaS platform like Salesforce
does not work that way — the metadata is alive. You cannot take an
org, cut a source artifact, and faithfully reproduce that org
somewhere else; configuration, data, and history are entangled in a
running system. So every engineering discipline that anchors itself
to source needs a new anchor before it works here. Rosetta's answer:
anchor to **intent and decisions** instead of code.

## Architecture Decision Records — from memory to guardrail

The ADR is a software-engineering classic (Michael Nygard, 2011): a
short record in the code repository explaining a consequential
decision, so future engineers inherit the *why*. Rosetta keeps the
idea and changes its job. Here, ADRs are first-class nodes in the
knowledge graph, attached to the very patterns they govern —
queryable in the middle of a live conversation, by people and by
agents. In the governed lane they are not written
ad hoc: a decision earns its ADR through recorded review
[gates](governance.md). And they are
*enforced*, not just remembered — when an implementing agent proposes
to improvise around a governed decision, the ADR is what tells it no.
In a code repo, an ADR explains the past. In Rosetta, it polices the
present.

## The Spec Manifest — the Spec Kit idea, re-aimed at SaaS

The [Spec Manifest](spec-manifest.md) descends openly from the
**spec-driven development** movement — GitHub's Spec Kit most
directly, alongside kin like AWS Kiro and Tessl. That movement's core
moves are all here: the specification, not the code, is the artifact
that directs the work; a project **constitution** states the
non-negotiable invariants every task must honor; and executable units
of work derive from the spec rather than from someone's recollection
of a meeting.

The customization is where it gets interesting. Spec Kit compiles a
spec into code for a system being built from scratch; Rosetta's
manifest compiles from a [digital twin](digital-twin.md) — an
evidence-bearing model of a *customer's existing, living org* — into
governed configuration of a platform nobody rebuilds from source. Its
constitution binds probabilistic implementers, human and AI, so every
clause carries provenance and its acceptance criteria come from
evidence. Same engineering shape, different target: not "spec →
code" but "intent → governed operator environment."

## The governance rubric — supply-chain discipline for knowledge

The rubric that gates everything — three-class
[provenance](governance.md), recorded decision traces, a maturity
ladder with explicit promotion gates — borrows from the software
**supply-chain and governance-as-data** tradition: W3C-style
provenance records, attestation chains, policy as inspectable data
rather than tribal habit. The customization: the supply chain being
secured is not a package registry but a body of *delivery knowledge*.
A pattern's journey from proposed to field-validated is tracked the
way a build artifact's journey is tracked — so when an auditor asks
where a recommendation came from, the answer is a record, not a
recollection.

The pattern across all three is the same: keep the engineering
instrument, move its object from code to intent, and add governance
where determinism was lost. That is what "customized for SaaS"
means here — and why the working papers describe their
lineage in detail rather than claiming invention
([the research](../research/papers.md)).

---

*Terms used across these pages are collected in the plain-English
[glossary](glossary.md).*