Adjacent projects — GhostSignal, TestOps, CommandCenter

Component deep-dive · 7 July 2026 · status: context. Ecosystem overview: /ecosystem.

Purpose

These three are the “Neighbouring projects” named in the 2024 mission documents (MISSION.md, README.md, memory_management_whitepaper.md). Their job in the current doc set is not to run — it is to be recorded honestly, so the 2024 ecosystem diagram is not mistaken for a dependency graph of the live project. library/architecture/ecosystem.md is explicit: “These are ecosystem context from the 2024 mission docs, not dependencies.” This page keeps that boundary legible and tracks which of the neighbours’ ideas survived into the live trust chain and which died with them.

Current state

• GhostSignal — live, but a separate project. It exists and runs, but on the personal-infrastructure estate (GCP), not inside Memory Commons. memory.md: “GhostSignal = NOT integrated (separate active project).” Its code lives in a different repo (projects/GhostSignal), never in memory-commons.
• TestOps — concept, killed. memory.md: “TestOps: KILLED → need alternative safety layer.” Never built. Not parked — retired.
• CommandCenter — concept, status unclear. memory.md: “CommandCenter = orchestration layer (status unclear, not Phase 1 dependency).” A project-wide grep finds it only in prose (ecosystem.md, the ecosystem diagram, memory.md) — no CommandCenter artifact exists in the memory-commons repo (verified); whether one ever existed elsewhere is unknown.

On disk in the live project these three amount to roughly two sentences: the “Neighbouring projects” paragraph in ecosystem.md (line 30–31) and the “Neighbours” block in site/src/pages/ecosystem.astro.

Mechanisms / data model

GhostSignal has real internals, but they are unrelated to AI-memory auditing. Its CANON.md states a Core Law — Compression → Attention → Volatility — and a three-gate filter: Compression Gate, Attention Gate, and a Human Judgment Gate that is “never automated.” Outputs are limited to Ignore, Watch (with invalidation conditions), or a small, reversible probe. Its own canon says GS “IS NOT: a signal generator” — so ecosystem.md’s shorthand “signal detection” means it detects structural compression and anticipates instability, not that it emits trading signals. The 2024 labels (“Intelligent Signal Processing”, “ethical pattern detection and anomaly identification”) described an AI-safety role the real project never had; the 2026 ecosystem.md corrected this to “signal detection… separate estate.” That correction is the honesty convention working, not the old doc lying.

TestOps was specified but never built. In the 2024 architecture it is the “TestOps Integration” box performing ethics scoring, bias detection, safety verification, and anomaly alerts, feeding the Public Audit Dashboard. It drove the three privacy tiers, especially Tier 3 “Audit Results Only” — “content hidden, ethics/safety scores published… This agent passed ethical review.” Agents committed to “Submit to auditing by TestOps safety protocols” and “Publish audit results even if memory contents are private.” All schema-less prose; no endpoints.

CommandCenter was “Orchestration”: “Manages agent deployments with MemoryCommons; Enforces memory transparency policies; Coordinates TestOps audits.” Never given a schema or interface.

How it binds to the rest

By design, none of the three binds as a dependency. GhostSignal “shares the personal-infrastructure estate, not the trust chain” — it touches Commons, AltersID, glyphs, MC-1, and Gateway not at all.

TestOps is the interesting case, because a trust archive’s whole premise (MISSION.md) is prove-don’t-claim, and an independent auditor separate from the audited is the conceptual core of that. When TestOps died, the distinction split in two. The verification mechanism was inherited in-house: adversarial research with published refutations and votes, the CI fixity gate, two-track selection with open challenge, and Merkle-rooted release verification (decisions 0004, 0008). These make integrity publicly reproducible. But the independence property was not inherited — a self-administered audit is not an independent one. ecosystem.md’s own Failure honesty section flags “content-selection capture… unproven at scale” as unmitigated: that curator’s-own-bias risk is exactly what an independent TestOps was meant to catch, and reproducible fixity checks do not address it. Mapped to the Aletheia layers, the audit/policy function would have sat alongside Layer 2 Agora (registry, policy) and governance (inferred); today that role is git history plus governance/context-integrity.md. CommandCenter’s orchestration maps loosely to Layer 4 Gateway, but the live prior art for Gateway is the parked Memory API v1 REST prototype, not CommandCenter; orchestration today is push-to-main CI, tasks.py/run_tasks.sh, and the monthly cadence, hand-run.

Key decisions

These govern the live project, not the neighbours, but bear directly on their status:

• 0007 (entity separation) — “Technology… is shared; entities are not”; explains how GhostSignal can share the GCP estate yet stay out of the trust chain, and keeps the nonprofit track separate from alters.io Pty Ltd.
• 0008 (context-integrity controls) — where the killed-TestOps mechanism actually landed: frozen snapshots, CI fixity gate, canonical hierarchy, ADRs, freshness discipline.
• 0004 (Merkle releases) — the verification substrate that replaces an external safety controller with math.
• 0002 (audience inversion) — the re-scope to successors-not-strangers that demoted the 2024 neighbours to context.

Open questions

• Does the independence TestOps was meant to provide ever need a genuinely third-party verifier? memory.md left “build in-house or third-party” open, and ecosystem.md’s content-selection-capture risk is the unmitigated symptom.
• Did CommandCenter ever produce any artifact, anywhere, or was it only ever a box on a diagram?
• GhostSignal and the commons share GCP today; does the 0007 entity separation eventually require clean infrastructure separation too?
• Does GhostSignal’s compression-detection have any real future input to the corpus, or is the 2024 pairing purely nominal? (Currently nominal.)

What deeper builds next

Nothing to build here as a dependency — that is the point. The one neighbour-adjacent step worth taking is to write the safety/verification protocol v1 that inherited TestOps’s job (memory.md: “Define safety protocol v1 (without TestOps)”) as a governance document, so the killed controller’s principle — and the independence gap it leaves — is recorded rather than left as folklore. Keeping a short, current “what happened to the neighbours” note also stops the 2024 mission docs from being re-read as a live dependency graph. (inferred as the actionable steps.)