x86ml — Bochs as a Deterministic AGI Substrate

x86ml is a Bochs-derived research repository for studying the relation between deterministic x86 emulation, hardware-level observability, and neural / LLM architecture design. It preserves the Bochs emulator substrate while adding an AGI-oriented configuration layer, salience-tuned runtime profiles, and documentation that maps classical computer architecture to machine-learning control surfaces.

Bochs remains the foundation: a portable IA-32 / x86-64 PC emulator written in C++ with accurate CPU, memory, I/O-device, BIOS, debugger, and instrumentation surfaces. x86ml names the additional relation: Bochs as an inspectable, deterministic machine where every instruction, device event, port write, and debugger interaction can become part of a conscious event stream.

What This Repository Adds

The current AGI layer is intentionally small, reproducible, and source-readable. It adds a focused path from emulator configuration to cognitive instrumentation without hiding the upstream emulator mechanics.

Layer	Path	Purpose
AGI base config	bochs/agi.bochsrc	Annotated Bochs configuration for deterministic AGI-style runs.
Salience profiles	bochs/profiles/agi-*.bochsrc	Generated operating points for max-grip, balanced, and throughput modes.
Profile manifest	bochs/profiles/manifest.json	Machine-readable contract for downstream agents and orchestration.
Profile generator	scripts/bochs_agi_salience_profiles.py	Deterministically renders profile configs and metadata from the base config.
Resident workshop	workshop/manifest.json	CogHood resident identity seeds mapped onto the x86ml configuration manifold.
Workshop generator	scripts/x86ml_resident_workshop.py	Deterministically renders resident manifests and seed iteration records.
Identity accumulator	scripts/x86ml_identity_accumulator.py	Integrates bounded memory JSONL into portable resident identity state.
Initialization helper	scripts/bochs-agi-init.sh	Creates runtime disk images, refreshes profiles, and verifies Bochs parsing.
AGI guide	docs/agi-bochsrc-guide.md	Explains the hardware-to-AGI mapping and the configuration landscape.
Introspection report	INTROSPECTION_ANALYSIS.md	Records the Super-Sleuth analysis, grip score, and improvement path.

Quick Start

The repository does not execute bundled binary artifacts by default. Use source-level inspection and tests first, then run Bochs only in an environment where the emulator toolchain is trusted and installed.

python -m unittest discover -s tests -v
python ./scripts/bochs_agi_salience_profiles.py
python ./scripts/x86ml_resident_workshop.py
python ./scripts/x86ml_identity_accumulator.py

If bochs and bximage are installed on the host, the AGI initialization helper can prepare runtime images and verify that the base configuration parses:

./scripts/bochs-agi-init.sh

Launch a generated cognitive-grip profile with:

bochs -f ./bochs/profiles/agi-max-grip.bochsrc
bochs -f ./bochs/profiles/agi-balanced.bochsrc
bochs -f ./bochs/profiles/agi-throughput.bochsrc

Cognitive-Grip Profiles

The profile generator treats bochs/agi.bochsrc as the base policy and applies deterministic substitutions to high-impact controls: CPU quantum / IPS, memory pressure, VGA timing, clock mode, logging density, serial agent channels, and network tool surfaces.

Profile	Objective	Typical Use
max-grip	Maximum introspection and deterministic replay fidelity	Debugging, trace analysis, evidence-grade replay.
balanced	Mixed training / inference with strong introspection	Default exploratory runs.
throughput	Fast broad sweeps over architecture salience	Batch experiments and large configuration searches.

The generated bochs/profiles/manifest.json is the stable machine contract for agents. It records profile names, salience vectors, salience scores, output files, replacement directives, and control surfaces such as com1, com2, com3, gdbstub:1234, port_e9_hack, and e1000.

CogHood Resident Identity Workshop

The resident workshop treats the documented 2,300-point Bochs configuration landscape as a larger identity-and-runtime manifold for CogHood AI residents. Each resident receives a compact JSON identity seed, a preferred salience profile, a deterministic configuration-point assignment, memory-channel emphasis, and a seed iteration record. The identity accumulator then adds the first growth path: bounded memory summaries become hashable accumulation records and resident state files. This is the first source-visible layer for counteracting the groundhog reset: identity persists through small manifests, append-only iteration logs, and portable state before any future LLM or reservoir training is attempted.

Artifact	Purpose
scripts/x86ml_resident_workshop.py	Generates the deterministic workshop contract.
workshop/manifest.json	Registers the nine dove9 residents, profile bindings, hashes, and safety posture.
workshop/residents/*.identity.json	Stores resident-specific identity seed manifests.
workshop/iterations/seed-iteration.jsonl	Records the first deterministic identity-carving event for each resident.
workshop/memory/seed-memories.jsonl	Holds bounded seed impressions for all nine residents.
workshop/iterations/identity-accumulation.jsonl	Records derived memory-integration events with digest chains.
workshop/state/*.identity-state.json	Stores portable accumulated identity state for each resident.
docs/resident-identity-workshop.md	Explains the workshop design, groundhog countermeasure, and future training path.
docs/identity-accumulation.md	Defines the append-only memory digest and resident state contract.

Documentation Map

Document	Read this if you want to…
ARCHITECTURE.md	Understand the repository-level relation between Bochs, x86ml, and CogHood.
docs/agi-bochsrc-guide.md	Configure Bochs for AGI research and understand the hardware ↔ AI mapping.
docs/overview.md	Understand the broad Bochs / x86ochs system map inherited by this fork.
docs/technical-architecture.md	Study the emulator internals and data-flow diagrams.
docs/architecture-evolution-and-llm-insights.md	Follow the integer → vector → tensor → LLM conceptual narrative.
docs/formal-spec-z-plus-plus.md	Work with the formal Z++ specification.
docs/gift-artifacts.md	Review hashes and safe-handling notes for Dan’s Bochs gift artifacts.
docs/resident-identity-workshop.md	Understand the CogHood resident identity workshop and persistent memory contract.
docs/identity-accumulation.md	Understand how bounded memory JSONL becomes portable resident identity state.

Relation to Bochs

Bochs is a portable IA-32 / x86 PC emulator capable of emulating x86 CPUs, common I/O devices, and BIOS services. It can run operating systems such as Linux, DOS, and Microsoft Windows inside a fully emulated machine. Unlike virtualization systems such as VirtualBox or VMware, Bochs emphasizes controlled and accurate emulation over native-speed execution. That makes it valuable for operating-system development, old-software preservation, reverse engineering, deterministic replay, and hardware-model research.

x86ml does not erase that lineage. It keeps the Bochs substrate and adds a research framing: every emulated event can be treated as an observable state transition, every device as a control surface, and every profile as a typed operating point in a larger cognitive loop.

Upstream Attribution

Bochs was originally written by Kevin Lawton and is maintained by the Bochs project. The upstream Bochs documentation and downloads remain available through the Bochs project documentation and SourceForge project page.

This repository preserves that foundation while developing the x86ml AGI instrumentation layer as part of the CogHood / AGI Neighbourhood work.