Why does this demo also have an underscore-style package name?
For Python users and contributors: understand an import-compatible facade that preserves access to the canonical demo without duplicating its engine.
01
Read the compatibility layer
Inspect how this package locates or re-exports the canonical implementation. The alias is supporting infrastructure, not a separate performance demonstration.
Confirm the resolved implementation path when debugging imports. Run from the repository root or use the documented launcher so relative package paths resolve consistently.
{"revision":"5b4cebb309a83a7a6749d8911d8bf96a1921e042","sources":[{"file":"demo/huxley_godel_machine_v0/__init__.py","content":"\"\"\"Python-friendly namespace for the Huxley–Gödel Machine demo.\"\"\"\n","format":"text","sha256":"18e06203e3bff24460b1836c3fa8147808b34f5b626b145ca4fbcb5730cb990f","bytes":70,"download":"/AGIJobsv0/examples/18e06203e3bff244-__init__.py","source":"https://github.com/MontrealAI/AGIJobsv0/blob/5b4cebb309a83a7a6749d8911d8bf96a1921e042/demo/huxley_godel_machine_v0/__init__.py"},{"file":"demo/Huxley-Godel-Machine-v0/README.md","content":"# Huxley–Gödel Machine · AGI Jobs\n\n**Turn intelligence into useful work, and use reviewed outcomes to guide the next investment.**\n\n[Open the research console](https://montrealai.github.io/AGIJobsv0/experiments/huxley-godel/) · [Operator runbook](RUNBOOK.md) · [Validation and limits](VALIDATION.md)\n\nAGI Jobs is designed as a scalable machine labor layer for authorized, lawful screen-based work—coordinating specialized agents to execute tasks, produce reviewable evidence, and support independent verification and settlement across a broad range of computer-based workflows.\n\nThis module makes the improvement loop inspectable: compare hierarchical exploration with a greedy baseline, enforce an experiment budget, inspect the lineage, and produce a source-bound benchmark analysis. The long-term ambition is a productive network that can reinvest demonstrated value into better research, software, infrastructure and eventually greater energy and industrial capacity. Scale follows evidence, resource availability and governance.\n\n## Start in two minutes\n\nFrom the repository root, with Python 3.12:\n\n```bash\npython demo/Huxley-Godel-Machine-v0/run_demo.py --seed 7\n```\n\nThe standard-library simulator writes reports to `demo/Huxley-Godel-Machine-v0/reports/` and its current comparison to `demo/Huxley-Godel-Machine-v0/web/artifacts/comparison.json`. Open the published console and import that comparison file. Uploaded files are processed in your browser; the page does not dispatch workers, connect wallets or send payments.\n\nFor a local, self-contained viewer, use the repository's pinned Node/npm versions (`.nvmrc`, `package.json`) and locked dependencies:\n\n```bash\nnpm ci\nnpm run demo:hgm:build\npython -m http.server 8765 --bind 127.0.0.1 --directory build/hgm\n```\n\nOpen <http://127.0.0.1:8765>. The build includes three freshly generated recordings: reference, constrained budget and owner pause. **Import a new comparison to view your own run**; the bundled reference does not update when another process runs. Stop the local viewer with Ctrl+C.\n\n## What is implemented\n\n| Surface | Working behavior | Evidence boundary |\n| -------------------- | ----------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- |\n| HGM simulator | Seeded lineage search, clade bookkeeping, thermostat, sentinel, owner limits, queued-cost reservation | Synthetic outcomes; no customer task execution |\n| Greedy baseline | Separate seeded strategy with the same budget ceiling and owner controls | Different scheduling rates and RNG stream; not a controlled efficacy trial |\n| Research console | Scenario selection, exact totals, SVG chart, lineage, logs, record import/export | Imported records have untrusted provenance |\n| Analysis deliverable | Computes committed cost and value less commitments, binds source summaries by SHA-256, exports JSON | Actual local computation on simulation inputs |\n| Candidate checker | Recomputes the expected source-bound output and rejects changed fields or approval flags | Content checks, not independent provider provenance or substantive review |\n| Screen-work planner | Ten workflow categories, USDC budget, reserved reviewer time, exportable work-order draft | A draft requires buyer-specific scope and operator admission |\n| Connected worker | Reuses the repository's admitted OpenClaw Responses adapter with exact task/job binding and persistent dispatch journal | Requires a separately commissioned runtime, credentials and admission; not run by the static website |\n| ChatGPT Work | Export a bounded task for an authorized operator-led session; import the candidate | No assumed remote Work dispatch endpoint |\n\nThe scope spans software, research, public datasets, web applications, scientific reproduction, AI evaluations, editable documents/presentations, vendor research, tooling and other authorized computer workflows. Tool availability expands what can be attempted; it does not prove that every task can be completed reliably. Use approved public, licensed or synthetic non-personal inputs. This module's connected benchmark task uses synthetic input only.\n\n## Read the numbers correctly\n\n- `gmv` is simulated gross value, in USD-equivalent units. It is not a USDC balance, customer revenue or settled work.\n- The historical `roi` field is **gross value / completed cost**, a gross multiple. Net return would be `(gmv - cost) / cost`. Undefined ratios serialize as JSON `null`.\n- `reserved_cost` is the cost of queued, unfinished work. Admission checks `cost + reserved_cost + next_cost <= max_budget` before scheduling. Reports retain these commitments at the horizon instead of pretending they completed.\n- `pending_tasks` records unfinished work. The console's value-less-commitments metric is `gmv - cost - reserved_cost`.\n- Both strategies observe the same budget ceiling and owner directives. Compare seeds, evaluation counts, elapsed steps and costs; a favorable single seed does not establish superiority.\n- The website's **$40 trillion/year** opportunity is a user-supplied planning assumption. The calculator applies eligibility and capture percentages; its result is illustrative gross work value before costs, not measured TAM, platform revenue or a forecast.\n\n## Run a bounded experiment\n\n```bash\npython demo/Huxley-Godel-Machine-v0/run_demo.py --seed 7 \\\n --set economics.max_budget=100 \\\n --output-dir /tmp/hgm-budget-study \\\n --ui-artifact /tmp/hgm-budget-study/comparison.json\n\npython demo/Huxley-Godel-Machine-v0/run_demo.py \\\n --set owner_controls.pause_all=true \\\n --output-dir /tmp/hgm-paused \\\n --ui-artifact /tmp/hgm-paused/comparison.json\n```\n\nMalformed values, non-finite numbers, unknown configuration keys, negative costs and inconsistent ranges fail before output creation. Owner pause prevents new scheduling; it does not revoke effects already dispatched in another system. The simulator is a finite experiment, not a continuously running controller.\n\nOutputs: `effective_config.json`, `summary.json`, `summary.txt`, `hgm_timeline.json`, `baseline_timeline.json`, `hgm_lineage.mmd`, `roi_comparison.svg`, `logs.md`, plus the requested comparison file. Save different experiments to different directories. The same seed and resolved configuration reproduce metrics; timestamps and destination paths vary.\n\n## Preserve the system map\n\n```mermaid\nflowchart LR\n Operators((Mission Owners)) --> demo_Huxley_Godel_Machine_v0[[Demo → Huxley Godel Machine v0]]\n demo_Huxley_Godel_Machine_v0 --> Core[[AGI Jobs v0 (v2) Core Intelligence]]\n Core --> Observability[[Unified CI / CD & Observability]]\n Core --> Governance[[Owner Control Plane]]\n```\n\nThis is the original conceptual systems map. The static console is an observer and authoring surface; integration with live orchestration requires the explicit admission process in the runbook. The original web evolution flow and Grand Operator Console diagrams are also retained. Their terms “self-modification” and “mission execution” describe simulated quality mutations and sampled outcomes here, not changes to production code.\n\n## Directory and compatibility guide\n\n| Path | Role |\n| ----------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |\n| `run_demo.py`, `simulator/runner.py`, `src/hgm_v0_demo/` | Canonical simulator and report writer |\n| `config/hgm_demo_config.json` | Canonical experiment configuration |\n| `web/`, `scripts/build_site.mjs` | Research console, content checker and offline asset build |\n| `scripts/worker.cjs`, `config/worker-profiles.example.json` | Shared admitted-worker integration; empty admissions by default |\n| `ui/` | Preserved Grand Operator Console, bundled under `legacy/` |\n| `run.py`, `hgm_demo/`, `config/hgm_config.json` | Preserved historical simulator with its own assumptions; not the canonical bounded-worker interface |\n| `scripts/demo_hgm.js`, `scripts/hgm_owner_console.py` | Guided launcher and owner override helper |\n| `tests/`, `web/tests/` | Simulation regressions, content-contract tests and browser QA |\n| `reports/` | Generated artifacts, excluded from source control |\n\n`make demo-hgm` and `python -m demo.huxley_godel_machine_v0.simulator` remain supported. The historical `run.py` path remains available for comparison. Do not mix configuration or telemetry formats between implementations.\n\n## Verify before merging\n\n```bash\nPYTEST_DISABLE_PLUGIN_AUTOLOAD=1 python -m pytest demo/Huxley-Godel-Machine-v0/tests -q\nnpm run demo:hgm:build\nnpm run demo:hgm:test\nnpm run demo:hgm:lint\nnpm run demo:hgm:qa\n```\n\nChanges land through a pull request with required checks green. The dedicated workflow exercises the simulator, task inspector and browser; the Pages workflow also checks the published HGM route. No live deployment, unrelated reviewer acceptance, buyer use, payment or production commissioning is asserted by these checks.\n","format":"text","sha256":"cab83bcc58f85ceedfc9e571c18a62f3adc83911548a9a99d7bc7ba1b9ec2a1b","bytes":10644,"download":"/AGIJobsv0/examples/cab83bcc58f85cee-README.md","source":"https://github.com/MontrealAI/AGIJobsv0/blob/5b4cebb309a83a7a6749d8911d8bf96a1921e042/demo/Huxley-Godel-Machine-v0/README.md"}]}
FROM READING TO A REPRODUCIBLE RUN
Continue to the implementation.
Supporting material
This entry supports another demo or the collection as a whole. Use the canonical experience for execution; preserve the compatibility layer or presentation asset when making changes.
Open the canonical experience below. This package exists for import compatibility; its success should be evaluated by delegation and behavioral parity, not a separate synthetic score.
The source inspector above reads bundled repository material. Local commands run separately on your computer. Recorded examples may contain historical timestamps, placeholders and simulated metrics.
MAKE IT YOUR OWN
One useful experiment.
Compare the imported module path and exported API with the canonical source. If behavior differs, inspect the shim and the environment before changing the underlying engine.
THE SYSTEM, MADE VISIBLE
Architecture & relationships
This supporting entry has no standalone architecture diagram. Follow its related experience below to inspect the canonical system.
WHEN SOMETHING DOESN’T MATCH
Troubleshooting
A command or service fails
Start at the first error, confirm the working directory and pinned toolchain, then follow the linked component guide. Optional packages and provider services have separate prerequisites.
A report looks successful but lacks evidence
Check its source revision, configuration, timestamps and underlying events. Historical fixtures and generated plans do not prove a fresh execution.
TRACE THE CHECKS
Verification & next steps
This directory has no standalone test source files. Follow the shared implementation and its behavioral tests before making correctness claims.