Research Library · Index
Third Signal Research Library
Published research from Third Signal Lab, with the claims, tensions, and open questions that connect it.
- Papers
- 07
- Signals
- 21
- Cross-paper links
- 08
- Open questions
- 04
Tags
7 of 7 papers
- No. 002Professional Identity
Professional DNA: The Case for Career Intelligence as Infrastructure
Why the next competitive layer is knowing what the labor market knows before it acts
Professional DNA reframes career navigation as market-calibrated intelligence. The core move is to stop treating resumes as truth and instead treat labor-market pricing, suppressor signals, and role adjacency as the real substrate of professional identity.
JAN 202618 minDepth: Deep3 signalsTop signalWeight 4 of 5“The resume is a lossy compression of professional identity.”
- No. 003Human-AI Collaboration
Ghost Hand: Agentic Systems That Operate Without Being Seen
Design principles for AI agents that augment human judgment invisibly
Ghost Hand defines a product philosophy for agentic systems that remove mechanical work while remaining auditable. Its central claim is that invisibility only works when the execution layer is more legible than the interface itself.
DEC 202515 minDepth: Deep3 signalsTop signalWeight 5 of 5“The best AI agent is the one you forget is running.”
- No. 004Governance
COE Cookie: Enterprise AI Governance Through Behavioral Telemetry
A lightweight protocol for observing AI adoption patterns without surveillance
COE Cookie proposes a signed telemetry receipt that captures AI value creation without storing sensitive prompts. The paper frames governance as observability with dignity rather than blanket monitoring.
NOV 202511 minDepth: Surface3 signalsTop signalWeight 4 of 5“You cannot govern what you cannot observe. But observation must not become surveillance.”
- No. 006AI Systems
Mitigating Context Rot in Agentic Software Development
The Continuous Context Protocol and the Librarian Pattern
The rapid adoption of Large Language Models (LLMs) in software engineering has fundamentally altered the development lifecycle. However, the stateless nature of conversational coding agents introduces a critical vulnerability: Context Rot. As projects scale in complexity, agents frequently hallucinate structural implementations, overwrite live data plumbing with mock placeholders, and discard established architectural decisions across isolated sessions. In this paper, we introduce the Continuous Context Protocol (CCP) and its central mechanism, the Librarian Pattern. Emulating human "shift-handover" methodologies, the CCP enforces rigorous, autonomous documentation of architectural decisions into a semantic
EXECUTION_LEDGER. Analyzing the implementation of the CCP within a complex spatial operating system (Orbital OS), we demonstrate that forcing contextual continuity drastically reduces hallucinated regressions and elevates agent operational capacity from localized task-execution to global architectural stewardship.MAR 20268 minDepth: Deep2 signalsTop signalWeight 5 of 5“Intent, not syntax, is the scarce context for a coding agent.”
- No. 007Human-AI CollaborationAwaiting review
Orbital Sonic Architecture
A layered, compile-once approach to AI-generated sound design in an agent operating system
Agent operating systems are silent by default. Operators spend whole days inside them, and nothing in the interface tells the ear where it is, what the machine is doing, or when something needs attention. This paper describes a sound architecture for Orbital, the operator surface of Third Signal Lab, that merges a system sound doctrine, a mission-radio runtime, and a generated-places engine. The architecture sorts sound into three registers (instrument, place, voice) that share six buses under one precedence rule. Voice outranks everything, alerts outrank ambience, and generated environments are compiled once, stored with provenance, and looped at no further cost. We specify the event taxonomy, the mix targets, the scene object, six contracts, and an adoption sequence, and we list the surfaces a UX/UI designer must design and the decisions that remain open.
OCT 202619 minDepth: Full immersion2 signalsTop signalWeight 5 of 5“Silence is part of the design. Not every event earns a sound.”
- No. 205Human-AI Collaboration
Raziel Lab Preview: Operator-Led Colab as Research Substrate
How Research OS turned a notebook proof into public institutional evidence before the final lab substrate existed
This case study documents the first non-mock Raziel Lab Preview run over Orbital's Colab lane. A Research OS proposal initialized a lab run, uploaded a CSV artifact, dispatched to a real downstream Swarm worker, and recovered a notebook URL plus a generated chart. The lesson is not that governance is finished. It is that preview compute becomes trustworthy the moment proposal, operator, run id, notebook, and returned artifacts remain linked.
MAR 202614 minDepth: Deep4 signalsTop signalWeight 5 of 5“Preview compute is acceptable when provenance is explicit.”
Still open04 questions
- No. 007
Which of two bus models is canonical: the Sound Bible's or the Cockpit Radio Player spec's?
openAwaiting reviewOpen paper - No. 007
Is literal place chatter compatible with the fiction-only contract?
openAwaiting reviewOpen paper - No. 007
Do ambient scenes ship last, as a luxury layer, or first, as the product?
openAwaiting reviewOpen paper - No. 007
Where does the audio transport live in a shell with no header?
openAwaiting reviewOpen paper