compromises
Community@tiendungchs
First-principles reasoning methods for deriving ML architectures, loss functions, and system designs from forced constraints rather than memorized patterns.
Agent Skills by compromises
Showing 5 vetted skills indexed across 1 GitHub repositories.
tiendungchsderive-dont-declare
Derive explanations and design decisions through forced-question chains instead of asserting conclusions.
tiendungchsmechanism-derivation
Derive model architectures, objectives, and learning rules from first principles.
tiendungchsintuitive-block-design
Derives new neural network blocks from first principles using eight ordered design questions.
tiendungchsfirst-principles
Guides structured problem framing, building, and stress-testing for hard or ambiguous engineering problems.
tiendungchsfind-the-framing
Guides reframing of hard design and debugging problems before committing to solutions.
Frequently Asked Questions About compromises
FAQPage SchemaWhat tasks can I accomplish with compromises' skills?▼
You can derive architectures, losses, and update rules from forced constraints, design new blocks from first principles using an eight-question method, debug systems whose behavior contradicts your mental model, and reframe ambiguous problems before committing to expensive implementations.
Who should use these reasoning skills?▼
ML researchers and engineers designing model architectures or objectives, developers debugging systems that defy their mental model, and anyone reading papers whose equations feel arbitrary or choosing between competing design approaches.
When should I apply the intuitive-block-design skill?▼
Apply it before writing a block contract, registering a candidate, or opening an invention block. It walks through eight ordered questions producing artifacts—episode, cost table, forced-move list, failure mode, readout probe—that feed the block's note and known-answer test.
How does find-the-framing differ from first-principles?▼
Find-the-framing targets problems that feel harder than they should, spending effort on the framing that collapses the problem. First-principles is broader, covering new system design, unfamiliar code, stubborn bugs, and judging whether a proposed approach is sound.
What prerequisites are needed to use these skills?▼
No installation or dependencies are required; they are reasoning methods. You need a concrete problem—an architecture to design, a mechanism to explain, a bug resisting quick fixes, or a paper whose derivations feel arbitrary—and willingness to chain forced questions.