[PIR] Epic: Perpetual Intelligence Runtime — RF sensing substrate

Repo epic — RF sensing substrate. Part of the RuV Perpetual Intelligence Runtime (PIR), a five-repo program across ruvector, LatentMesh, rvm, autogenous, and RuView.

Canonical plan

The authoritative program plan is docs/research/perpetual-intelligence-runtime/03-program-plan.md in ruvnet/ruvector — https://github.com/ruvnet/RuVector/blob/main/docs/research/perpetual-intelligence-runtime/03-program-plan.md. This epic links to it rather than duplicating it.

Program epic: ruvnet/RuVector#837 — https://github.com/ruvnet/RuVector/issues/837

Why this repo is involved

PIR's sixth invariant is "every physical action produces new evidence." RF sensing is where that evidence originates, and it feeds the acceptance harness's sensing-accuracy metric.

This work is a separately-funded stretch track, sequenced after the WP0–WP12 critical path, and it splits into two parts of very different evidentiary standing:

1. Closing ADR-178's declared gaps C and D — concrete, scoped, already-specified engineering.
2. The universal / heterogeneous-chipset CSI vocabulary — a first-party research bet with no prior art anywhere.

⚠️ There is no prior art for the universal CSI vocabulary

The program's evidence review searched specifically and repeatedly — direct phrase search, "chipset-agnostic," "shared vocabulary," "common backbone," "tokenize" + CSI, and explicit vendor-name combinations. No paper frames CSI from Realtek / Qualcomm / Nexmon / MediaTek chipsets as a shared "RF latent vocabulary" feeding one common backbone.

Good news for novelty, bad news for evidence backing. Do not present this as reproducing published work. Program risk #2 is precisely that this distinction gets lost downstream and someone writes an ADR citing a paper that was never written.

Adjacent work to build on and differentiate from — cite as motivation, never as validation:

• arXiv:2608.14694 — Survey of Wireless Foundation Models for AI-Native 6G Networks (Aug 2026); general, not chipset-specific.
• arXiv:2604.18255 — WiFo-MiSAC (Apr 2026); MoE tokenization of heterogeneous signals, but for multimodal sensing/comms.
• arXiv:2512.22143 — UniFi; irregular CSI sampling across packets and bands.
• arXiv:2605.14171 — CSI-JEPA; masked-prediction CSI foundation model.

Scope

WP	Role
WP14	Real-CSI validation + pose-semantic embedding pipeline (ADR-178 gaps C and D), optionally extending into the universal CSI vocabulary bet

A companion issue in ruvnet/ruvector covers validating ruvector-perception (3.8K LOC, currently synthetic-data-only) against real CSI captures.

Program shape

PIR is consolidation and reconciliation across sibling repos, not greenfield design. Most mechanisms it needs already exist somewhere in the org, often in more than one place with no integration between them — evaluation-before-promotion exists four times over, and the cross-repo witness/promotion contract is already written as ruvnet/ruflo ADR-322C. Read each work package as "which existing implementation do we keep, and what wires them" before "what do we build."

Note: ruvnet/autogenous ADR-400/401 cite a "metaharness ADR-322" — that is a misattribution; ADR-322 is ruvnet/ruflo's, and MetaHarness's series tops out at 250.

Prior-art framing for the wider program

PIR extends ruvnet/autogenous ADR-401 (Accepted, 2026-08-16) and ruvnet/LatentMesh ADR-009 (Proposed, 2026-08-18) and closes their declared gaps — it does not re-derive them. Name the repo when citing any ADR number; ruvector's ADR set has ~15+ duplicate numbers currently under remediation.

Child issues

• #1657 — WP14 Real-CSI validation + pose-semantic embedding pipeline (ADR-178 gaps C & D) stretch

Related

• ruvnet/RuVector#852 — companion issue validating ruvector-perception (3.8K LOC, synthetic-only) against real CSI captures

Start with docs/research/perpetual-intelligence-runtime/03-program-plan.md in ruvnet/ruvector, then read child issue #1657 and the linked ADR-178 gaps C and D. The stated target is WP14: real-CSI validation plus a pose-semantic embedding pipeline; any universal CSI vocabulary work remains an explicitly unsupported research bet rather than a validated implementation.