Skip to content

Project notice

1.13.0 — Business 3 Enterprise Studio

Business 3 now produces a useful, reproducible enterprise allocation in the browser, Python CLI, installed wheel, container and notebook. Its five editable examples expose capital, staff capacity, independent review time, an AGIALPHA job budget, downside constraints, dependencies, exclusions and evidence thresholds. Exact enumeration finds the best feasible portfolio and retains negative stress results. Eleven transparent analysis roles cover nine sectors, constraints and replay.

Use the checksum-verifying installer for matching operator release assets.

Decisions and portable evidence

Seven exported files retain the scenario, complete dossier, decision brief, selected-project CSV, unsubmitted job specifications, seed draft and checksums. Python and JavaScript use the same integer rounding, tie breaking and canonical commitments. Verification recomputes the result rather than trusting a supplied hash. Existing output is never silently overwritten.

The job export is accepted by the existing ascension-compile and ascension-check commands. The protocol reference retains Nova-Seed lineage, MARK risk gating and funding, Sovereign treasury restrictions, ENS/stake admission, reputation-weighted auctions, validator decisions, refunds, slashing and a 1% AGIALPHA payout burn. Planning itself creates no transaction or approval.

Reliable launch and recovery

The console-module no-op is fixed. The default planner imports no optional SDK and requires no key or network. The container uses the correct build context, runs without root or network, and retains output on the host. The helper supports readable help without invoking Docker and forwards the selected case. The notebook uses a tagged checkout, runs the same calculation, independently verifies the dossier and exports the same files. An existing checkout supports offline notebook use.

Browser calculations and evidence preparation run in a cancellable worker. Edited input invalidates prior exports. Imported dossiers are recomputed; ambiguous JSON and oversized files are rejected. Saving drafts is explicit, and clearing one draft leaves unrelated workspaces intact. Both canonical and mirrored routes support offline recalculation after their initial load.

Preserved research and behavior changes

Breaking changes: the maintained CLI now defaults to the finite enterprise planner. Use --legacy-loop before the historical loop options to run the preserved Ω-Lattice experiment. Optional model commentary and ADK/A2A research adapters require explicit opt-in. Parent environment settings are restored, clients close once after the loop, and missing requested inference fails visibly. The research verifier rejects absent evidence; the model object retains research proposals without claiming that trained weights changed.

Original flowcharts, paths, replay charts, PDF/PPTX and manuscripts remain. The original narrative and notebook are available as clearly labeled research archives. Unsupported historical performance, physical, proof and deployment claims are not presented as verified current capabilities.

Required release evidence

  • Exact Python/browser comparison across 41 scenarios and all seven artifact bytes for five cases.
  • Python 3.11–3.13 runtime, notebook and installed Business 3 execution; all 15 finite catalog launches from the wheel, including the complete backend matrix on Python 3.11/3.12.
  • Actual container execution with no network, a read-only root and retained independently verified output.
  • Canonical and mirrored browser routes, downloads, Ascension job compilation, tamper rejection, worker cancellation, local draft recovery, mobile layout, WCAG checks and offline recalculation.
  • Public acceptance bound to the exact release commit, asset bytes and recomputed cases before publication.

Existing full regression, strict typing, repository hooks, dependency audits, model and contract acceptance, documentation, preservation and release checksum gates remain required. Exact outcomes are included in the release validation archive. The release scope remains bounded: constructed scenarios and supplied assumptions, a maintained private operator runtime, preserved research integrations and an undeployed local-EVM reference. This release does not establish general AGI or predictive superiority.