TRANSFORMATIONAL INTELLIGENCE · A WORKING FIRST STEP
Change the representation.
Keep the purpose.
Can a summary preserve its argument better if we map the prerequisites before choosing what to keep? Run the comparison. Inspect the clauses. Leave with a useful summary and a method you can test again.
Deterministic local prototype · independent review pending · no operating authority
This experiment measures annotated structural coverage. It does not measure truth, reader understanding, general intelligence or a certified ceiling escape. No model, wallet, payment or network request is used.
The included text is an authored public fixture. Your source remains text; instructions inside it cannot change this app’s permissions. Confirmation records your classification; it does not verify publication rights or detect personal information.
Fixed-summary charter
Produce at most three exact source clauses. Keep the purpose, source, permissions, clause limit and evaluation instrument unchanged across all methods. An interactive understanding product would need a broader, separately authorized charter.
Allowed: local computation and user-triggered file export. Budget: zero model calls, zero network calls, zero currency spend. Human annotation, review and maintenance effort are not measured.
Check every clause, give its importance from 1 to 1000, and list the clause IDs it needs as premises. This is an auditable representation you supply, not an automatically verified argument. Empty prerequisites mean none are annotated.
What assumption are we challenging?
Inherited design choice: draft first, then critique. Alternative: plan with prerequisite closure before compressing. Hard constraint: unchanged three-clause summary. Hypothesis: the representation may improve structural coverage. Falsifier: no score gain, incorrect dependencies, or an unrestricted comparator does better.
The incumbent takes the first three source clauses. It remains available even if another method has a higher local structural score.
Draft, then critique
Choose the highest-weight clauses without their graph. Audit missing premises afterward and make one bounded repair pass.
Graph before compression
Enumerate bounded subsets and retain only those containing every annotated prerequisite. Select the highest supported weight. The output remains exact source clauses.
Strong unrestricted comparator
Exhaustively search every subset within the same clause budget under the same structural instrument. It can match the graph method; this is not a novel-discovery claim.
- Currency spend
- 0
- Model / network calls
- 0 / 0
- Search bound
- Human effort, migration, maintenance
- Unmeasured; required before any value or savings claim
structural-coverage/v1 sums declared importance only for selected clauses whose complete annotated prerequisites are also selected. The weights and graph are not calibrated. This common instrument cannot establish truth or independent understanding. Independent evaluation remains pending.
Keep a scoped method, not a claim of general ability.
Reuse only with public text, a reviewed acyclic clause graph, and the same charter and instrument. Recheck whenever the source, dependencies or weights change. Retire the method if annotations are unreliable or the useful task needs broader context. No spending or execution authority travels with it.
Next: save this evidence, try a fresh public source, and ask an independent reviewer to inspect the graph and usefulness. More elaborate strategies need fresh confirmation before promotion.
Continue with the native discovery runtime
This local browser experiment and the native runtime have distinct execution histories. Export an exact native request, create its campaign in an initialized Alpha runtime, inspect its binding and explicitly approve there. Native search and confirmation each have a declared 100,000-unit logical budget; currency spend remains zero. No browser choice grants native authority.
Installed runtime: alpha-agent successor-transform create transformational-intelligence-request.json. Review the returned campaign and binding before using successor-transform approve. Use successor-transform export to save the native evidence, then inspect it here. Existing runtime initialization is a separate setup step.
One portable native project
Carry the original request, native evidence, separate USDC and $AGIALPHA inspections, Computer Use history and reuse review in one Fleet project. Imports retain historical bytes and grant no authority. Adapter inspections and signed library reviews remain historical unless checked by an installed receiver.
The completed native result supplies the exact source-grounded HTML. Executing the handoff requires the existing native Computer Use permission and configured executor. Preparing or importing it does not run a browser, certify usefulness, approve payment or promote a method.