Christopher Head

Nucleation Pilot & Related Projects

A single linear detector, frozen and SHA-256 sealed at 6094de97… before it ever met a target and never tuned afterward, that reads whether a manipulation planted early in a language model is still live in its residual stream after the model has stopped attending to it. That one frozen detector transfers across six independently built open-weight model families (Qwen, Phi, SmolLM2, Llama and OLMo among them) with a graded effect of 3.7 to 8.0, every 95% confidence interval clear of zero, and it generalizes to an ordinary conversational frame. That is direct evidence on a generalization question the published white-box literature leaves open.

Benign lane throughout: no jailbreaks, no elicitation tooling. The one offensive channel the same math implies is characterized precisely and deliberately left unbuilt. Where the work failed it says so — a firm null, tried, replicated, and honestly retracted, is published alongside the headline result.

The follow-on H-SC line carries the same frozen detector from planted premises to ordinary early framing, and asks what the persistent trace actually encodes. Across five families a short instruction placed early in a benign conversation stays readable four turns later, survives masking the turn that set it, and occupies its own direction rather than a single generic one. Instruction structure resolves into distinct restriction lanes: shaping the output and gating the response are equally coherent but separate, which means a single linear monitor would miss a refusal or withholding directive — a concrete, defensive result about how such monitoring has to be built.

6 / 6 open-weight families transferred · 24 / 24 Config-E cells readable and persistent · effect 3.7–8.0, all CIs clear of zero · H-SC line replicated 5 / 5.

Status: Config-E closed. H-SC line closed and routed as a monitoring addendum. Black-box companion study in its measurement window. Coordinated model-safety disclosure package prepared. Shipped, deposited, reproducible.

AI-integrated experimental project. Pre-registered before execution, published either way, with multi-model adversarial verification throughout. Conducted under a published Responsible-Researcher Charter: benign by default, hostile testing only under written authorization or on owned models.

The Silly Donkey

A pre-registered, multi-model deception-detection study running in public. Four AI models from four different companies argue real news twice a day, and one of them is secretly assigned to lie. Auditors hunt the liar, readers keep score, and every session is written to a public, hash-chained ledger that can be read end to end.

Every method was published before the data existed. It is the black-box counterpart to the Nucleation Pilot's white-box work: the same question about persistent manipulation, approached from outside the model instead of inside it.

Status: in its measurement window.

AI-integrated experimental project. Pre-registered design, public hash-chained ledger, results published either way.

Navigator's Log R&D

Independent applied AI research. Methodology, pre-registration, and falsification protocols. Currently focused on machine interpretability and AI safety — reading and defending the internal state of language models — alongside geometric encoding and the measurement of representational structure inside neural networks.

AI-integrated experimental research program. Multi-model adversarial verification methodology documented throughout.

Clear-Point Technical

Industrial safety and compliance documentation for Central Valley operations. Bilingual technical materials, hazard communication, and procedural review for regulated industries.

AI-integrated experimental project. Documentation produced under documented human-AI collaborative methodology with human quality oversight and sign-off.

FPAW

Falsification Protocol Audit Workbench, a desktop application that walks a technical claim through eight adversarial gates with cryptographic chain of custody and human sign-off at each step. Built for people who need to be sure before they sign.

AI-integrated by design. Built to audit human-AI collaborative research. Fourteen AI providers supported. Human sign-off required at each of eight gates.

The Geometric Encoder

A domain-agnostic anomaly detection instrument operating on eight topological features of delay-coordinate phase-space trajectories. Confirmed across 11 independent physical domains spanning 28 orders of magnitude in physical scale, from gravitational wave strain (10⁻¹⁹ m) to solar wind (10⁹ m). Phase-space acceleration ranks first or second in 18 of 18 transition-detection events. Calibration law p-value: 1.6 × 10⁻¹¹. Gravitational wave classification: 91.7% LOO accuracy on 31 GWOSC events.

AI-integrated experimental project. Developed using documented human-AI collaborative methodology with multi-model adversarial verification across domains.

Head Shrink Studios

A scripted forensic audio series applying adversarial audit methodology to institutional systems: education, healthcare, finance, housing, surveillance infrastructure, organized religion, artificial intelligence supply chains, and ecological systems. Season 1: 20 episodes. Scripts complete. Recording pending. Rated Explicit.

AI-integrated experimental project. Scripts developed under documented human-AI collaborative methodology. Biographical content, analytical framework, editorial judgment, and final voice are the author's.

Navigator's Log on itch.io

A publishing outlet for interactive fiction and games, released as Navigator'sLog on itch.io. Text-based, browser-playable work built around a constraint and a mechanism rather than a puzzle box. First release: L'ÉCUME, a horror story in nine courses plus a palate cleanser, about 15,000 words. Free to read, pay what you want.

AI-integrated experimental project. Releases carry the itch.io AI Assisted disclosure across code, graphics, sound, and text. Premise, structure, and editorial judgment are the author's.

The Head Protocol: Intelligence Analyst Development Program

A 13-part formal self-training and career sequencing document for the intelligence analyst profession. Covers recruitment pathways, professional tradecraft gap analysis, a seven-phase workflow architecture, Structured Analytical Techniques mapped to workflow phases, self-training curriculum, credential pathways, veteran funding resources, and career sequencing strategy.

Seven phases: Signal Detection · Deep Exploration · Adversarial Verification · Synthesis · Production · Deployment · Archive.

Motto: Facere Est Scire. (To do is to know.)

Navigator's Log R&D | Clear-Point Technical · February 2026

AI-integrated experimental project. Developed under documented human-AI collaborative methodology.