Version 1.8.0 — Proof, carried forward

The latest 198-page AGI ALPHA manuscript is now the versioned research specification, with its original PDF and Markdown preserved byte-for-byte, a pinned source manifest and an implementation/evidence map.

The new Compounding Lab learns a forecasting policy from one mandate, freezes it, and actually uses it on new tasks. Continuity, regime-change and no-archive scenarios expose both progress and failure. Four measured arms, full prior learning charges, explicit coordination assumptions and human review accounting support a bounded local claim. Broader manuscript promotion remains HOLD for missing strongest-agent comparisons, independent validation, scaling, α-WU calibration and delayed outcomes.

A complete 13-section Evidence Docket includes raw tasks, predictions, comparisons, capability, cost/safety ledgers, review, replay, calibration status and per-file checksums. Native CLI commands execute, review, export and replay browser or Python results. Strict imports recompute everything before accepting a document.

Proof Bloom now uses the manuscript's meaning of ECI: Evidence Contact Index. Its old executed-advantage test remains a separate required gate. Existing Chronicle events, all original demos and historical white papers remain available. The redesigned flagship experience adds a guided experiment, responsive prediction charts, explicit failure paths, keyboard access, mobile layouts and offline recovery.

Release acceptance includes cross-runtime equality, failure/ablation/cost gates, tamper and stale-review rejection, ZIP replay, manuscript checksums, real browser downloads, offline/mobile/accessibility checks, and all previous native, container, legacy demo, local model, chain and public deployment gates. The final release manifest contains the evidence from the exact published commit.

The operator API now measures actual request bytes, rejects ambiguous framing and declared-length mismatches, and bounds upload time before any state mutation. All early rejections keep response security headers. Current operating, recovery, capability and readiness guides now describe the same 1.8.0 release and the final manuscript, with tested deployment scope and research obligations stated explicitly.