What differs from a native baseline before the result is known?
Scientific attitude
Do not protect the framework. Try to break it.
A clean story is not evidence. A familiar analogy is not a discovery.
Every serious claim needs a test card.
What present carrier or constraint can be changed?
What happens when the alleged retained consequence is removed?
What comparison isolates the claimed operation?
What time, energy, memory, compute, and verification cost is included?
What result forces demotion, revision, or retraction?
Correspondence discipline
When two domains appear structurally similar, report the status before the story.
identical valueformal equivalencestructural correspondencewitnessed simulationshared signature onlyunsupportedincomparable
A valid comparison specifies the mapping, preserved properties, broken properties, evidence, uncertainty, and error. A shared vocabulary or function signature is not mathematical equivalence.
Evidence classes
Supported by bounded completed implementation, accepted formal work, or an experimentally specified result that stays inside its evidence.
Coherent enough to formalize and test; empirical promotion is still required.
Useful enough to guide experiments but underdetermined.
Language the program must not use without new evidence.
Immediate framework-level failures
- CF only renames established causation without new variables or discriminating tests.
- Cross-domain mappings collapse under native mathematics or require incompatible objects.
- Matched native methods outperform CF-guided methods after all verification costs are counted.
- Claims depend on history that is not instantiated in any causally relevant present state.
- The framework cannot return negative, unsupported, or incomparable results.
Immediate engineering program
MFM and Formative Field kill conditions
- Removing executable formation leaves held-out recruitment and reasoning unchanged.
- Bare-host policy explains the apparent learning without acquired formation.
- Raw transcript, ordinary RAG, RETE or production rules, procedural memory, temporal knowledge graphs, pub-sub, or content-addressable retrieval match it at equal quality and lifecycle cost.
- The proposed field scans or scores most retained state on each event.
- Reversal does not produce the predicted negative transfer from prior formation.
- Recovery formation does not restore performance on a second reversal or held-out recurrence.
- A single-susceptibility transplant does not transfer the predicted narrow reasoning difference, or ablation does not remove it.
- Reuse helps exact repeats but not structurally related held-out tasks.
- Replay produces rumination or interference without correction or generalization benefit.
Positive evidence requires causal mediation, not a pretty graph, a renamed retrieval system, or a benchmark score alone.
Distributed embodiment program
Embodiment and energy kill conditions
- The reachability index adds no decision value beyond ordinary capability discovery and service routing.
- Transferred formation fails compatibility checks or causes unbounded fleet-wide negative transfer.
- Partition and rejoin cannot preserve authority, provenance, contradiction, and local divergence.
- Energy-aware gating adds overhead without improving verified completion, safety, or lifecycle resource use.
- A proposed generation or storage placement loses to conventional infrastructure under the machine's actual load profile.
Identity across embodiments remains an open criterion question; engineering reachability does not decide it.
Go to the record
Inspect artifacts, not confidence.
Use the experiments, source documents, formal pages, public code, and scientific-neighbor analysis to evaluate individual branches.