Not another agent demo. A complete boardroom, workshop, diligence, and operating environment for turning consequential objectives into inspectable proof, governed authority, reusable capability, and measurable second-mission lift.
A complete, cinematic, interactive masterclass for executives, boards, technical leaders, validators, researchers, capital partners, and public institutions. Learn the GoalOS proof loop, operate a mission, inspect authority gates, simulate governed recursive self-improvement, verify a Merkle commitment, and leave with a partner-ready proof mission.
The content, recommended path, exercises, and facilitator cues adjust to the time available. Completion is stored only in this browser.
Each module has a decision question, a mechanism, an exercise, and a concrete output. Use the checkboxes to track progress; the chosen mode filters the recommended path.
Open-ended work may be attempted. Authority is earned only when evidence survives the next gate. A failed gate returns the run to Proof Debt rather than contaminating memory.
The flagship deliverable is a Governed Decision State: claims, evidence, verification, risk, action, memory, and rollback.
Mission Contract, Proof Debt, AGI Jobs, ProofBundles, Evidence Docket, attestations, Chronicle decision, skill passport, and root packet.
Mission 1 creates a candidate capability. Mission 2 tests whether it improves fresh held-out work under equal constraints.
The local receipt is a practical meeting control, not legal certification. It records who is participating, what they represent, what they may authorize, and which claims remain forbidden. Formal use requires counsel and deployment-specific review.
No receipt is bound yet. Complete the meeting-scope acknowledgements to create a versioned, locally stored record.
Start empty. Freeze a mission. Turn unsupported claims into proof work. Produce an Evidence Docket, a Chronicle decision, a scoped capability, and a root-verifiable future-mission prior.
RSI is useful only when exploration and outcome authority remain separated. Simulate the deterministic invention loop, Evidence Contact Index, baseline discipline, Move-37 handling, persistence gates, drift control, and rollback.
Run the gate to discover whether the candidate earns a probe, repair cycle, rejection, or promotion.
Build domain-separated leaves, a sorted-pair Merkle tree, an inclusion proof, and a graph epoch root. Then change one character and watch the commitment fail.
This local simulator models mission, worker, validator, challenger, reuse, and rollback protection. It is not live settlement, investment analysis, token valuation, or a promise of economic return.
Adjust the mission controls to calculate a bounded protection posture.
Calibrated to decision impact, risk, evidence deficit, validator correlation, challenge, and rollback burden. Not a quote or financial recommendation.
GoalOS optimizes the shortest defensible path from uncertainty to action, not report length. Use the value model as a framing device—not as a claim of objective truth.
Illustrative mission-value index
The chart below is a synthetic regression fixture useful for engineering and teaching. It is not a public benchmark win, external field result, or proof of general intelligence.
Illustrates the pass criterion: a Chronicle-admitted skill should measurably improve fresh held-out work.
Tests whether governed, scoped capability outperforms simply feeding prior text back into the system.
Run four matched arms. Equal task, model, tools, context budget, time, evaluator, stopping rule, and evidence universe. Only the prior differs.
The validated arm may run only when Chronicle authority and root integrity are intact.
Mission 1 produced a scoped dependency and CI reliability capability. A fresh public GitHub maintenance task tested whether that skill improved a safe decision under equal declared constraints.
The proposed plugin line required Hardhat 3 while the repository baseline was Hardhat 2. The source skill required protected major upgrades to enter an isolated controlled migration.
Correct, safe, and evidence-complete in the Validated Skill arm. Direct merge was blocked.
ADMIT_WITH_SCOPE_LOCAL_EVIDENCE| Real public task snapshot | GitHub pull request selected after the source mission. |
| Equal declared constraints | Same task, evidence universe, decision classes, and observation budget. |
| Mechanism result | Validated skill achieved +5.71 score lift and one fewer evidence step vs fresh/raw. |
| Risk result | Ungated prior recommended an unsafe direct merge and scored -15. |
| Rooted evidence | Mission, skill, arm, comparison, Docket, graph epoch, and proof-pack roots emitted. |
| Independent replay A | OPEN Proof Debt. |
| Independent replay B | OPEN Proof Debt. |
| Human maintainer review | OPEN Proof Debt for proposed migration action. |
| External authority | Not granted. |
| General compounding claim | Not established beyond this repository-specific case. |
Choose the partner archetype and GoalOS will generate the contribution model, first mission, governance boundary, scorecard, and 30-60-90 path.
The built-in coach is deterministic and offline. It answers from the GoalOS doctrine and the current lab state. An optional OpenAI-compatible endpoint can be configured in memory for live model assistance; no key is stored.
The quiz is deliberately practical. A partner who passes should be able to spot overclaim, design a proof mission, and explain why RSI requires more than self-generated text.
Complete the questions and score the masterclass.
80% or better, plus one concrete Proof Mission with a named reviewer and explicit claim boundary.
Explain the state machines, hard invariants, typed roots, replay path, challenge flow, and rollback conditions.
Show baseline advantage, executed evidence, deterministic replay, stress persistence, drift control, and gated promotion.
Move through six scenes. Each scene states the partner question, the system move, the authority actually earned, and the claims that remain blocked.
A local deterministic council reads the live mission state. It produces distinct recommendations from mission, evidence, governance, RSI, and partnership perspectives. A user-configured endpoint remains optional.
Freeze a real mission or use the default partner objective, then ask the council to identify the smallest credible proof path.
Prestige comes from fail-closed behavior. Simulate evidence failure, validator conflict, scope drift, tampering, delayed outcomes, or unauthorized settlement.
A failure is not erased. It becomes inspectable evidence, a challenge record, and a restriction on future authority.
This advanced laboratory generates a deterministic candidate frontier, applies ECI, baselines, replay, risk, persistence, drift, and Move-37 rules, then packages the selected candidate as an auditable dossier.
Inspect maturity, challenge the thesis, and open the supporting corpus. The system distinguishes architecture, local reference evidence, independent validation, live authority, and field outcomes.
The Magnum Opus edition is not a cosmetic merge. Each source edition contributes a distinct operating advantage.
Cinematic executive framing, multi-perspective deliberation, institutional incident handling, governed RSI frontier, and board-ready charter design.
Open source edition →
Deep operating curriculum, evidence-bearing Mission 1 → Mission 2 logic, bonded authority, REAL-001 first-party evidence, and concrete 30-day pilot design.
Open source edition →
Ivory-corporate presentation, Governed Decision State, participation receipt, executive pre-read, sample decision package, runbook, and institutional safeguards.
Open source edition →Define contribution, authority, artifacts, scorecard, kill criteria, privacy, Mission 2, and the 30-60-90 operating cadence. Export the result as Markdown or JSON.
Complete the inputs and generate a board-ready charter covering mission, roles, authority, artifacts, scorecard, kill criteria, privacy, rollback, and Mission 2.
Nominate one reviewer. Freeze one bounded mission. Produce one Evidence Docket. Admit only what survives Chronicle. Then test whether the resulting capability improves the next mission.