• workshop_dense_100#slide 11

Taken together, the papers support a two-tier reading in which the full RAIDT concept space is organised by eight named branches. At the macro level there are three construct families: governance artefacts, influence methods as governance interventions, and outcome constructs. At the governance-outcome level, the third family is unpacked into five pillar branches: Responsibility, Auditability, Interpretability, Dependability, and Traceability. Read in this way, RAIDT has eight organising branches without losing the papers' own architecture.

This matters because the framework is not merely a list of values. Governance artefacts specify what evidence must exist for one configured use: prompt records, configuration digests, retrieval snapshots, hashes, timestamps, tool-use traces, and review logs. Influence methods explain how configured interventions alter both behaviour and the evidentiary trail. The five pillars then express what governance quality looks like when that evidence is inspected. The run-level evidence pack is the carrier that binds these branches together around the run as the unit of governance, while the score profile makes their relationships visible through anchors 1=missing / 3=partial / 5=audit-ready.

For supervisory purposes, this eight-branch reading is useful because it prevents category errors. It stops reviewers from treating the entire theory as only evidence logging, only RAG, only alignment, or only one pillar. Instead, it clarifies that RAIDT spans what is recorded, how behaviour is shaped, and how governance readiness is judged.

In the public-service eligibility vignette, the evidence branch includes the prompt, the policy retrieval snapshot, identifiers, hashes, and the review log. The intervention branch includes retrieval augmentation as a governed design choice. The five pillar branches then ask whether that specific run was used responsibly, can be audited, is interpretable for the caseworker, behaves dependably across repeats, and remains traceable to the policy text that informed it.

This is why an eligibility assistant cannot be assessed only by saying 'it used retrieval'. Retrieval belongs to one branch. The full RAIDT concept space requires the organisation to connect that branch to the run-level evidence pack and then score the outcome profile across all five pillars.

• 12-RAIDT_DSR_Theory_M_v8
• 11-RAIDT_Academic_Logic_M_v11