Common Signal

Common Signal is a worker-centred public-interest observatory for examining how AI changes journalism, public relations, communications, and the wider media industries. Its central questions concern work, voice, power, professional judgement, editorial independence, source plurality, and who carries the risk when communicative production is automated.

The project is informed by concerns associated with the National Union of Journalists (NUJ) and the wider trade-union movement. It is not presently an NUJ service, an endorsed NUJ project, or a substitute for collective bargaining, legal advice, or worker-led research.

Important

The included records are synthetic fixtures. Common Signal is not yet a deployed evidence service, does not collect live workplace or public-expression data, and cannot support claims about real workers, organisations, unions, or populations.

Table of Contents

• What it is for
• Current implementation
• Quick start
• AI-Assisted Installation
• Architecture
• Licensing

What it is for

Common Signal is intended to help workers, union representatives, researchers, editors, communications leaders, and public-interest organisations move from a vague claim such as “AI saved time” to inspectable questions:

• Where did checking, correction, coordination, and emotional labour move?

• Who gained or lost discretion over tools, pace, ranking, and refusal?

• Was a deployment disclosed, consulted on, or collectively negotiated?

• Did automation affect editorial or professional independence?

• Did production volume rise while source or framing plurality narrowed?

• Who receives credit, liability, surveillance, and opportunities to learn?

• What evidence supports the answer, and what inference is prohibited?

It deliberately keeps unlike evidence classes distinct. Sampled public expressions, workplace research, search interest, news corpora, incidents, and surveys cannot be averaged into a single authoritative “sentiment” or “impact” score.

Current implementation

The first implementation slice contains:

• a Rust canonical domain and query model in crates/common-signal-core;

• explicit occupational, employment, labour-impact, evidence, consent, and collective-governance fields;

• staged collective consultation and negotiation records rather than a yes/no consultation flag;

• executable default-deny prohibited-use checks and a versioned machine-readable policy;

• separate journalism, PR, communications, and cross-media duty/risk profiles;

• a public governance-authority page and JSON endpoint derived directly from the canonical machine status;

• a stable C ABI consumed by Zig;

• a project-local sixteen-slot Zig adapter boundary in adapters/hexadeca-zig, with an operational loopback-only REST transport and GraphQL/gRPC projections implemented at the contract level;

• thirteen additional protocol slots that fail explicitly as unsupported;

• the original self-contained design prototype and a critical product review in docs/prototypes/ and docs/PRODUCT-REVIEW.adoc.

No persistent store, live connector, authentication system, GraphQL transport, gRPC transport, or production deployment has yet been implemented.

Quick start

cargo build --workspace
cargo test --workspace
zig test adapters/hexadeca-zig/src/adapter.zig target/debug/libcommon_signal_core.a -lc -ldl -lpthread -lm -lunwind
just build-adapter
just run

The Rust build emits the static library consumed by the Zig test. The Zig suite checks both pure projection behavior and calls across the real FFI boundary. just run starts the Zig transport on 127.0.0.1:8080; the experimental server refuses routable interfaces by construction.

With the server running, the current authority boundary is available at http://127.0.0.1:8080/governance and as canonical JSON at http://127.0.0.1:8080/v1/governance. It currently reports draft-not-consulted, zero sessions/reviews/ratifications, and a closed live connector gate.

AI-Assisted Installation

AI assistance is optional. An agent should read 0-AI-MANIFEST.a2ml first, preserve the synthetic-data and worker-governance boundaries, and follow docs/AI_INSTALLATION_GUIDE.adoc. No installation step requires sending source, workplace information, union information, or credentials to an AI service.

Architecture

evidence sources (future)
          |
          v
consent + provenance + worker-governance boundary
          |
          v
Rust canonical domain and query algebra
          |
          v  stable C ABI
Zig sixteen-slot adapter registry and wire I/O
          |
          +-- REST       loopback transport implemented
          +-- GraphQL    projection only; no transport
          +-- gRPC       projection only; no transport
          `-- 13 slots   explicit UnsupportedProtocol

The Rust core owns meaning and invariants. Zig owns low-overhead protocol and FFI boundaries. If Elixir is added later for ingestion or durable orchestration, native integration should use supervised SNIFs rather than ordinary NIFs so a native failure does not crash the BEAM.

Worker-governance and prohibited-use details are in docs/WORKER-GOVERNANCE.adoc. The ready-to-use—but not yet conducted—worker and union review process is in docs/governance/CO-DESIGN-PROTOCOL.adoc. Versioned schemas are under schemas/v1/, and the policy synchronised with the executable Rust rules is policies/prohibited-uses.json.

Licensing

Code, configuration, schemas, and scripts are licensed under MPL-2.0. Prose documentation and the design prototype are licensed under CC-BY-SA-4.0. The authoritative texts are in LICENSE and LICENSES/ and per-file SPDX headers state the applicable licence.