Plan, apply and roll back changes to a production ontology, with a blast radius report and a proof an auditor can re-check.
Rust ai-nativedescription-logicsformal-verificationinfrastructure-as-codeisabelle-languageknowledge-graphleanlean4mcpmcp-serverontologyowlowl2reasoningrustshaclsparqltheorem-proving
docs: move the caught-defects record to the rules it belongs with, and give them an index
FABIOTESS2fcba2d080ad
diff --git a/README.md b/README.mdindex eec5b1ee..9364bf87 100644--- a/README.md+++ b/README.md@@ -64,18 +64,10 @@ flowchart LRC -->|forged| X["refused, exit 1"]```-**What this discipline has caught, in one week of running it against this engine.** Five-description-logic false cleans, each an inconsistent ontology reported consistent with full-confidence. A rule that could conclude a triple no serialiser can write, reachable from ordinary-OWL, which left the store non-deterministic: three runs of one input kept 40, 9 and 24 inferences.-And two independently verified kernels disagreeing on the same certificates, always in the safe-direction, tracing to a gap in the format that neither proof could see: it did not say what a-repeated binding key meant, so one kernel refused the shape and the other answered from whatever-its lookup happened to do. A key bound twice to different values is satisfied by no substitution-at all, so there were never two readings, only two ways of discarding half the certificate. Closed-by [decision 0008](docs/decisions/0008-a-binding-is-data-and-evidence-admits-one-reading.md), and-the differential now reports zero divergent rows over a corpus four times deeper in the property-its induction rests on. Every one of those had passed every test that existed before.+The discipline is not free and it has earned its keep: in one week against this engine it caught five+description-logic false cleans, a rule that could conclude a triple no serialiser can write, and two+independently verified kernels disagreeing on the same certificates. Every one had passed every test+that existed before. [What the rules are, and what each has caught](docs/decisions/).## What is actually proveddiff --git a/docs/decisions/README.md b/docs/decisions/README.mdnew file mode 100644index 00000000..846d50a2--- /dev/null+++ b/docs/decisions/README.md@@ -0,0 +1,44 @@+# The rules, and what they have caught++One file per rule. Each names the failure it exists to prevent, because a rule whose cost is visible+and whose benefit is not gets dropped the first time it is inconvenient.++| | Rule | The failure it prevents |+| --- | --- | --- |+| [0001](0001-an-inference-is-not-an-assertion.md) | An inference is not an assertion | A materialised conclusion becoming the next run's premise, so the engine cites itself |+| [0002](0002-an-inference-carries-a-certificate.md) | An inference carries a certificate, and the certificate has a proof | Trusting the engine's report of its own work |+| [0003](0003-a-rule-is-data-and-an-assumption-is-not-a-fact.md) | A rule is data, and an assumption is not a fact | A user-supplied rule table earning the word reserved for the checked one |+| [0005](0005-a-prover-is-an-oracle-and-a-translation-is-a-theorem.md) | A prover is an oracle, and a translation is a theorem | Reading a theorem prover's confident answer as evidence |+| [0006](0006-a-model-is-a-certificate-and-a-refutation-is-not.md) | A model is a certificate, and a refutation is not | Treating an exhausted bounded search as unsatisfiability |+| [0007](0007-a-slice-preserves-a-conclusion-or-it-does-not.md) | A slice preserves a conclusion, or it does not | A retrieval slice quietly dropping the conclusion it was asked about |+| [0008](0008-a-binding-is-data-and-evidence-admits-one-reading.md) | A binding is data, and evidence admits one reading | A certificate format admitting two readings, so two checkers disagree |++There is no 0004. The decision now numbered 0006 was drafted as 0004 on a branch that never merged,+so the number never reached `main`. It is left as a hole rather than reused, because a reused number+makes an old citation silently come to mean something else.++## What this discipline has caught++In one week of running it against this engine.++**Five description-logic false cleans.** Each an inconsistent ontology reported consistent, with full+confidence.++**A rule that could conclude a triple no serialiser can write**, reachable from ordinary OWL, which+left the store non-deterministic: three runs of one input kept 40, 9 and 24 inferences.++**Two independently verified kernels disagreeing on the same certificates**, always in the safe+direction, tracing to a gap in the format that neither proof could see. It did not say what a+repeated binding key meant, so one kernel refused the shape and the other answered from whatever its+lookup happened to do. The sharper fact came out of the second implementation rather than the first:+a key bound twice to different values is satisfied by no substitution at all, so there were never two+readings, only two ways of discarding half the certificate. Closed by+[decision 0008](0008-a-binding-is-data-and-evidence-admits-one-reading.md), and the differential now+reports zero divergent rows over a corpus four times deeper in the property its induction rests on.++Every one of those had passed every test that existed before.++A single formalisation, however careful, cannot see a defect in the format it formalises. That is the+argument for the second kernel, and it is the only argument for it that survived contact with the+work: two proof assistants agreeing does not mean what a casual reader assumes, and the value came+from independence of the specification reading rather than from quantity.