Workflow — how the two DTG directories work together

2026-07-17. The two roots are related, not separated: one is the workbench for direct working-group
work, the other is the research root that draws from it and feeds it. This doc is the sorting.

The two roles

~/dtgwg-zkp-tf-mage — THE WORKBENCH (Mitch's direct ToIP WG work).
The upstream clone of trustoverip/dtgwg-zkp-tf, maintained by Mitch as co-chair. Everything in it is
TF-deliverable-shaped:

• the tracked tree carries only changes Mitch intends to PR (currently: the 2-line README edit);
• the runtimes/ lab (git-excluded) is the evidence bench — reference implementations + property
  tests that back TF positions with runnable behaviour (rt 01 9/9, rt 07 11/11), plus the TF-facing
  registers (CRED-SPEC-COHERENCE, CRED-SPEC-OPPORTUNITIES, STRAWMAN-COHERENCE-EDITS E1–E7);
• its exit is upstream: PR / issue to ToIP, by Mitch, never automatic.

~/dtgwg-cred-spec-main_mage — THE RESEARCH ROOT (exploration of the ideas).
No upstream discipline; iterate freely. It holds the vendored cred-spec (untouched, others' work), the
integration map, the decision-document working draft, and explorations/ — the ideas built out across
the board. It draws from the workbench (cites the runtimes, the coherence map, the register) and
feeds two directions: back into the workbench when an idea matures into TF work, and outward into
the agentprivacy corpus.

The loop

                    ToIP upstream (dtgwg-zkp-tf · dtgwg-cred-spec)
                         ▲                        │
                 (3) PR/issue — Mitch only        │ new spec material
                         │                        ▼
              ┌──────────┴───────────┐   vendored into
              │  dtgwg-zkp-tf-mage   │   the research root
              │  WORKBENCH           │
              │  direct WG work      │
              └──────────┬───────────┘
                 ▲       │
     (2) promote │       │ (1) draw from
                 │       ▼
              ┌──────────┴───────────────┐
              │ dtgwg-cred-spec-main_mage │
              │ RESEARCH ROOT             │
              │ explorations of the ideas │
              └──────────┬───────────────┘
                         │ (4) integrate
                         ▼
        agentprivacy corpus (master · spellweb · docs · guide)

(1) Draw from. Explorations cite the workbench freely — runtimes, coherence map, E-register. The
workbench is the source of evidence; the research root is where its meaning is stretched.

(2) Promote. An exploration graduates to the workbench when it stops being an idea and becomes TF
work — concretely, when it needs one of the workbench's two lanes:

• needs runnable evidence → build the runtime / extend property tests in runtimes/ (the design doc
  stays in explorations/ as the source; the lab gets the code and a NOTES pointer back);
• needs TF review → the document moves into the workbench as a PR-shaped tracked change at its
  upstream path, and the research-root copy becomes frozen lineage (or a pointer).

(3) Upstream. Only from the workbench, only by Mitch, only via PR/issue after TF discussion. The
research root never touches upstream directly — the workbench is the airlock.

(4) Integrate. Explorations also project outward into the corpus via Integration Map
(master /model, skills, spellweb KG, docs canon via the pipeline, guide federation). That lane rides
the existing display/re-sync trains and Mitch's G-M gate — independent of the upstream lane.

Placement rules (what lives where)

Artefact class	Home	Why
Reference implementations, property tests	workbench runtimes/	evidence bench; future conformance fixtures
Predicate↔anchor registers, E1–E7 candidate edits	workbench runtimes/	TF-facing, PR-adjacent
PR-shaped spec-text changes	workbench tracked tree	the only thing that leaves the machine
Vendored upstream specs	research root (dtgwg-cred-spec-main/)	others' work, read-only
Exploration/design docs (O-items, VWC seat, …)	research root explorations/	free iteration
TF deliverable drafts while being written	research root	drafting is exploration
TF deliverable drafts once review is requested	promote → workbench (then upstream path)	it has become direct WG work
Corpus integration maps / status	research root	the outward lane's ledger

Current dispositions under these rules

• Decision — Overview — stays here while it iterates (v0.1.0-draft). Its own
  §31 names its upstream home (docs/implementation-guide/boundaries/… in the TF repo); when Mitch calls
  it review-ready, it promotes to the workbench as a tracked, PR-shaped addition at that path — the first
  full pass through lane (2)→(3).
• explorations/ — home is correct. Each doc's "build plan" M-items that create code (O2 M2/M3,
  VWC M1/M2, O3 M3, O7 M3) execute in the workbench lab per lane (2); the doc itself stays here.
• CRED-SPEC-COHERENCE / OPPORTUNITIES / STRAWMAN-COHERENCE-EDITS — stay in the workbench lab: they
  are TF work product (the register keeps a pointer here; the explorations cite them — lane (1) both ways).
• Versioning: the research root is fast-and-loose (no git today; give it its own local git later if
  the lineage starts mattering). The workbench's tracked tree stays clean enough that git status always
  answers "what am I about to propose to the TF?"