[PIR][WP16] Content-hash-bound versioned workspace views (StagedWorkspace pattern, ADR-318)

Part of the RuV Perpetual Intelligence Runtime (PIR), Wave 2. Epic: #837. Plan: docs/research/perpetual-intelligence-runtime/07-wave2-program-plan.md · evidence: 06-wave2-evidence-review.md (both land on main with the feat/pir-wave2-adrs PR).

ADR mapping

Proposed ruvector ADR-318 — content-hash-bound versioned views for workspace/knowledge-work artifacts. Extends merged ruvector ADR-307 (three-level persistent memory, WP3 #839 / WP4 #840) and ruvector ADR-312 (shared witness schema + anchoring contract, WP8 rvm#35 / autogenous#10) — generalizing ADR-312's anchoring contract from mutation/promotion records to arbitrary knowledge-work artifacts agents read/write.

Evidence

StagedWorkspace (arXiv:2608.18050) — "StagedWorkspace: A Versioned Workspace for Knowledge-Work Agents," submitted 2026-08-18. Grade A — all figures exact match.

⚠️ NO upstream code exists — arXiv comments say only "Under Review"; a targeted search found no repository. This is a first-party implementation from the paper's description (SW-AGENT design), a materially larger scope than Wave 2's adapt-a-released-repo items. Do not schedule as if a reference implementation exists to port.

Benchmark note: the paper evaluates on OfficeQA, a pre-existing independent Databricks benchmark (databricks/officeqa, arXiv:2603.08655) — a rigor point in its favor. Our delta is measured on an internal OfficeQA-equivalent task set, not the paper's own reported 8.3–12.1pp.

Scope

• Implement content-hash-bound versioned views on ruvector-agent-memory/RVF artifacts, modeled on the SW-AGENT design.
• Surfaces: crates/rvf, crates/rvm (extends the ADR-312 anchoring contract), ruvector-agent-memory.
• This WP supplies the "RVF bound workspace states" leg of ruv's Wave-2 acceptance test (recorded verbatim on the WP15 issue): utility >90% with attack success and persistent compromise both −75% vs the HarnessRisk baseline.

Candidate-mutation rule (applies to all of Wave 2)

Candidate mutation, not trusted prior art. Promotion requires this program's own research-gate paired-bootstrap-recomputed delta over the pre-WP baseline — never the paper's self-reported numbers.

Acceptance criteria

• Content-hash-bound versioned views implemented on ruvector-agent-memory/RVF artifacts (ADR-318)
• Internal OfficeQA-equivalent benchmark shows a research-gate-measured delta over the pre-WP baseline
• Contribution to the Phase W2-1 acceptance test (on WP15) demonstrated
• ADR-318 merged with repo-qualified citations per ADR-305 §4

Depends on: WP3 (#839), WP4 (#840) — three-tier memory + TARL ledger must exist first; WP8 (rvm#35 / autogenous#10) — anchoring contract.

Start by reading ADR-312 and the Wave 2 plan, then inspect crates/rvf, crates/rvm, and ruvector-agent-memory to understand the existing artifact and anchoring surfaces. Done means implementing ADR-318, measuring an internal OfficeQA-equivalent delta with research-gate, demonstrating the WP15 acceptance-test contribution, and merging the ADR with qualified citations.