Stage 1
Stage 1 approves intent. It checks the goal is the one you meant before any plan is written.
Handover, Version 5
Turn a rough prompt into finished work.
Handover asks a few questions. You approve once. It runs the plan itself. Each run caps at $2, and it runs supervised.
How it works
Three steps. You approve. Then it runs.
-
1
Translate the rough prompt
Handover turns your rough prompt into a full plan and orchestration. It asks only the early questions needed to set the plan.
-
2
Approve in two stages
Stage 1 approves intent. Stage 2 approves the execution manifest. Two approvals catch a wrong plan before any work starts.
-
3
It runs inside the envelope
A coordinator owns the plan and runs it with full autonomy inside the approved envelope. Each run caps at $2, and it runs supervised.
The author is never the verifier, and the judge rules in a fresh context. That split keeps a weak result from passing its own test.
The gate
Nothing runs until you approve it twice.
Stage 2
Stage 2 approves the execution manifest. It checks the exact work before the run starts.
A repair that changes the manifest needs reapproval. A gap is never converted into a pass, and a dead end report comes when the predicate is unmet.
The 16 engines
Sixteen ways to think. One umbrella over them.
All 16 engines plus the Discovery umbrella were proven Nov 12 to 14, 2026. Each proof used sealed tests with 2 fresh blind trials and a fresh judge.
The core 4 loop
- Anomaly Hunting finds what breaks the model.
- Abduction names the best cause.
- Falsification tests it.
- First Principles sits underneath.
Invention is the outcome, not a 17th skill. The loop turns a broken model into a new idea.
Anomaly Hunting
ResearchFinds what breaks the model.
Abduction
ResearchNames the best cause.
Falsification
ProofTests the cause.
First Principles Questions
ResearchSits underneath the loop.
Induction
ResearchBuilds a pattern from cases.
Deduction
ProofDraws what must follow.
Analysis and Synthesis
ResearchSplits a problem, then rebuilds it.
Thought Experiments
StuckRuns a case in the mind first.
Analogy and Cross-Domain Transfer
ResearchCarries a working pattern across fields.
Inversion
StuckStarts from failure and works back.
Counterfactual Thinking
StuckAsks what changes if one fact flips.
Constraint-Based Discovery
SearchUses limits to narrow the search.
Search the Possibility Space
SearchMaps options before picking one.
Evolutionary Search
SearchKeeps what works and varies it.
Serendipity plus Prepared Mind
SearchNotices a surprise and knows what it means.
Consilience
ProofChecks that separate lines agree.
17
Discovery
The umbrella over all 16 engines. One skill holds the router, the phases, and the core 4 loop.
Proof
Router run 3 passed on Nov 14, 2026.
Both fresh blind trials passed both test situations. The run cost $0 of the $4 cap.
2fresh blind trials
2test situations passed
5 of 5repeats each
$0of the $4 cap
12 of 14spawns used
2 of 3fix cycles used
Limits
What the proof does not cover.
- The loader check never ran, because no Muse skill loader command exists in the environment where the proofs ran.
- The router proof covers two test situations, not live use of all 17 skills.
Guides
Read the manual, or inspect the schema.
Visual guide
Parts, 6 assembly steps, loop and graph, and troubleshooting. The full IKEA-style manual for Handover v5.
Field kit and graph
The schema, the kit diagram, the graph diagram, the per-unit list, and the map from each drawing to its schema path.
Star Handover on GitHub.
Version 5 is on GitHub now. A star tells the next visitor the work is worth a look.
Star Handover on GitHubgithub.com/richardsondx/handover