[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AP9E9MTXA3CS87A1FEDBP5":3,"skill-same:01M7AP9E9MTXA3CS87A1FEDBP5":64},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":33,"ownerDescription":34,"updatedAt":35,"install":36,"resources":49},"01M7AP9E9MTXA3CS87A1FEDBP5","pandects",null,"Check a credit protocol against executable laws: conservation, accrual and withdrawal claims, each one a Solidity component with a broken specimen it is proven to catch. Use when someone asks which invariants a lending or credit system should hold, wants properties for a fuzzing campaign, or hands over a protocol and asks what could be checked mechanically. Do not use it to generate a harness for one repository; that is fizz. Never report a campaign under an engine that did not run.","wildcat-finance\u002Fskills","wildcat-finance","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F113041915?v=4","01M7AP0V1G8X0SWHP6C5V90RM5","The Wildcat Protocol","01M7AP0V24Y50BGVCQ4DV496VD",13,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fpandects\u002Fskills\u002Fpandects\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: pandects\ndescription: >\n  Check a credit protocol against executable laws: conservation, accrual and\n  withdrawal claims, each one a Solidity component with a broken specimen it is\n  proven to catch. Use when someone asks which invariants a lending or credit\n  system should hold, wants properties for a fuzzing campaign, or hands over a\n  protocol and asks what could be checked mechanically. Do not use it to\n  generate a harness for one repository; that is fizz. Never report a campaign\n  under an engine that did not run.\nmetadata:\n  version: \"1.1.0\"\n---\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"..\u002F..\u002Fassets\u002Fcharacters\u002Fpandects.png\" width=\"1200\">\n\u003C\u002Fp>\n\n# Pandects\n\n## Frontier\n\nPandects owns its own executable-law frontier, not Hexaemeron's delivery or\nSolidity frontier. Its version, held target, next job, and maturity\nstate live in [EVOLUTION.md](EVOLUTION.md). Do not recommend or run\nanother frontier pass after that ledger becomes mature.\n\n\u003C!-- marketplace-context:start -->\n## Where this sits\n\nPandects supplies executable credit laws, each paired with a deliberately broken specimen it is proved to catch and a replayable counterexample.\n\n**Current frontier.** The search-record runner records only the Foundry campaign, so Echidna and Medusa results survive as audit prose rather than as records.\n\u003C!-- marketplace-context:end -->\n\nFizz or another engine can search these laws, Janus can apply the relevant\nones around a hook transition, and Hermes must preserve them when changing gas.\nPandects states and tests the law; it does not generate a target-specific fuzz\nharness, audit the whole protocol, or turn one bounded campaign into security\nsign-off.\n\nSynkrisis compares validated run observations. Its shipped comparison cannot\npromote repeated campaign outcomes into a law or security verdict, and a\nPandects campaign result is not a run observation merely because it exists.\n\nA fuzzer searches a state space. It cannot decide which economic facts must\nsurvive that search. Generic token property libraries cover balances, approvals\nand round trips; credit adds time, accrual, queues, changing terms,\ndelinquency, partial repayment and claims that are neither liquid nor reducible\nto a token balance. Every audit rediscovers those facts in prose, implements a\nsubset in a local harness, and the harness dies with the engagement.\n\nThis is the corpus that survives it.\n\n`$SKILL_DIR` is the directory holding this file. The tool lives at\n`$SKILL_DIR\u002F..\u002F..\u002Fscripts\u002Fpandects.py`; resolve it from where you loaded this\nskill.\n\n## Day to day\n\n**Security and audit.** A credit protocol arrives and the invariants have to be\nworked out before anything can be fuzzed. The corpus states them, says which\naccounting model each one assumes, and ships the broken contract each is proven\nto catch, so a law that holds against your target is a law you can see working.\n\n**Developers.** A change touches accrual or the withdrawal queue and the\nquestion is what it might have broken. Point the corpus at the build and read\nwhich laws stopped holding, with the quantities that were compared.\n\n## What a law is\n\nSix parts. Anything with fewer is not a law, and the checker refuses it:\n\n1. **It executes.** A Solidity component, not a sentence.\n2. **It catches something.** A deliberately broken contract it is proven to\n   fail against. A law that has never failed may be a tautology; a property\n   asserting `seen() >= 0` survives a campaign of five hundred calls without\n   complaint, and nothing in that run distinguishes it from a real law.\n3. **It has been reduced.** A minimal counterexample, replayable without a\n   fuzzer.\n4. **It says where it applies.** The accounting model, the assumptions and the\n   observables it requires.\n5. **Its bounds are justified.** Exact, or a tolerance naming the arithmetic\n   that produces it. An epsilon chosen because it made a test pass is refused.\n6. **It judges rather than reverts.** A law returns `(bool held, string\n   detail)`.\n\nThat last one is the decision worth knowing about. A campaign runs with\n`fail_on_revert = false`, because a credit system reverts constantly and\ncorrectly. Under that setting a revert carries no verdict, so a law using\n`require` to mean \"violated\" reports nothing and is counted as silence. Silence\nread as assent is what this exists to prevent.\n\n## Two shapes\n\nConservation is a fact about one state, so a `Law` reads one target. Accrual is\nnot: debt falling without payment cannot be violated by any single state,\nbecause the violation is in the transition. A `PairLaw` judges two\n`Observation`s instead, each a snapshot of everything a target reports.\n\nThree pair laws compare a system with its own past. One compares two systems\nstarted identically and advanced over the same span by different routes, which\nis the only way to see interest compounding when it should not. A law says\nwhich kind of pair it means in its applicability, and handed one it cannot\njudge it either holds or refuses on a single rule: a pair that spans real time\nis a state of the world, and a pair nobody could have meant is a mistake by\nwhoever built it.\n\n## The commands\n\n```bash\npython3 scripts\u002Fpandects.py laws\npython3 scripts\u002Fpandects.py check\npython3 scripts\u002Fpandects.py run --out search-record.json\npython3 scripts\u002Fpandects.py render\n```\n\n`laws` lists the catalogue with each law's applicability. `check` refuses a law\nmissing any of its six parts, naming the part rather than the file. `run`\nsearches and writes down how it searched. `render` writes `docs\u002Fcatalogue.md`\nout of the catalogue, which is the only thing that should ever write it.\n\nExit codes: 0 success, 1 a check failed, 2 usage or validation error.\n\n```bash\nforge test\n```\n\nruns every law against every specimen and asserts the diagonal: each law fails\nthe specimen written for it and holds against the others. The corpus takes no\nexternal Solidity dependency: there is no `lib\u002F`, no submodule and nothing to\nfetch.\n\nThe catalogue holds ten laws in three families. Conservation: value is\nconserved across held assets, debt, claims and fees; assets reserved never\nexceed the claims recorded; and reserved plus borrowable never exceed what is\nheld. Accrual: debt falls only against assets arriving, rises at rest only\nagainst assets leaving, and the same span costs the same however it is cut up.\nWithdrawal claims: a recorded claim is never written down smaller, nobody is\npaid ahead of somebody who has waited longer, and what a system declares\npayable it has set aside; pooled lender claims cover what is still owed on open\nwithdrawal batches.\n\nNine are exact. `path-independent` carries the corpus's only tolerance, and the bound\nis derived rather than chosen: linear accrual on principal truncates once per\nstep, so `n` small steps differ from one long step by at most `n - 1` units.\n\nTo point another engine at it, `src\u002Fcampaigns\u002FSpecimens.sol` carries a harness\nper specimen declaring both `echidna_` and `property_` prefixes, since Echidna\nlooks for the first and Medusa defaults to the second. Every entry point\nsnapshots the system on the way in, which is what lets a pair law read a past.\n`explain()` returns each law's reason for the current state, so a failing\nsequence gives up its reason rather than making somebody derive it from a call\ntrace.\n\nOne harness passes everything and the contract underneath it is broken.\n`CompoundsPerStepCampaign` cannot see its own defect, because path independence\ncompares two systems advanced by different routes and a campaign drives one\nalong one route. It is the plainest illustration here of what a clean campaign\nis worth on its own.\n\n## Pointing it at a protocol\n\n`adapters\u002F` is where the corpus meets somebody else's system. Two shapes, and\nthe difference is a limit rather than a convenience.\n\n`CorpusObserver` holds any address and offers the six one-state laws. Nothing\nroutes calls through it, so it offers no pair law: one that never sees a call\nhas no past to compare with, and would report holding forever.\n\n`CorpusDriver` takes the calls. Extend it, write the protocol's entry points,\nput `records` on every one that changes state, and the three succession laws\nrun as well. Nothing catches a forgotten `records` -- it looks like three laws\nholding quietly -- so `recordedCalls` is public and `successionExercised` reads\nit. Zero recorded calls means those laws searched nothing, whatever they\nreported.\n\nPath independence is in neither, because it compares two systems and an adapter\nfronts one. `PathIndependenceProbe` takes both.\n\n## Saying how it was searched\n\nA campaign result without its settings is an anecdote, which is the same\nrefusal the corpus already makes of a tolerance that does not name its\narithmetic. `run` writes a record naming the engine, the argv, the determinism\nclass, the configuration read out of `foundry.toml` rather than restated, the\nsequence length and a digest of the corpus searched. It is shaped as an\n`ariadne` command entry, so a result drops into a release statement without\ntranslation.\n\nAn engine that did not run is absent from the record, not present and empty.\nWhat nobody could read is absent rather than null. And a campaign killed by the\ntimeout is neither passed nor failed: it ran, searched less than it was asked\nto, and says so.\n\n## What a law may read\n\n`ICreditObservables`. It names economic roles: the asset, total assets held,\ntotal debt, total lender claims, reserved and borrowable assets, accrued fees,\nand the time the observation describes. A target implements it, or a thin\nadapter does, and that adapter is the only place a protocol's own names appear.\n\nA system with a withdrawal queue also implements\n`IWithdrawalQueueObservables`: the claim count, each claim's owed and paid\namounts, and the bound through which the system declares claims payable. Kept\nseparate because a system with no queue would implement three members and mean\nnone of them, and an observable that means nothing reports zero, which reads\nlike an answer. A target that lacks it reverts on the read, and a revert is no\nverdict rather than a false one.\n\nThis is enforced rather than requested. A property written against one\nprotocol's function names is a property about one codebase, and the corpus\nexists because the same economic facts hold across codebases that share nothing\nelse.\n\n## Pointed at a real design\n\n`integrations\u002Fwildcat\u002F` is the corpus against a codebase it was not written\nfor: a reduced model of a Wildcat market with batched withdrawals, a reserve\nthe borrower may not touch, delinquency and penalty accrual.\n\nSeven laws apply flatly. Three do not, and the three are what the integration is\nfor.\n\n`queue-order-preserved` applies at batch granularity and says nothing per\nlender, because a batch is paid pro rata. `path-independent` holds while the\nmarket is solvent and stops once penalty accrual runs, because the grace timer\nadvances when the market is poked. And `recorded-claim-never-shrinks` does not\nhold over a batch that is still open: a batch accumulates while open, so the\namount owed on it rises, and the law says a recorded claim keeps its amount.\n\nThat third one was found by Echidna against the shipped adapter, after the\napplicability notes had already claimed the law held. Reading found two\nconditions; running the corpus against a real design found the third, in a\ndocument that had already got it wrong.\n\nNone of the three is a yes or a no.\n`integrations\u002Fwildcat\u002FAPPLICABILITY.md` carries all of them with their reasons.\n\n## The documents\n\n`docs\u002Fcatalogue.md` renders every law for a reader, and it is a rendering\nrather than a second source: `pandects render` writes it, and a test compares\nthe committed bytes with what the renderer produces. `docs\u002Fwriting-a-law.md`\nhas the six parts in the order somebody adding a law meets them.\n`docs\u002Fapplicability.md` states the rules once -- what a law may read, the two\nshapes, what a pair law does with a pair it cannot judge, and what a revert\nmeans.\n\n## What this never does\n\n- Replace an engine. Echidna, Medusa and Foundry search; the corpus says what\n  to look for.\n- Generate a harness for one repository. That is `fizz`, which can select and\n  adapt these laws and stays useful where no public law describes the protocol.\n- Report a campaign under an engine that did not run. A result carries its\n  engine, configuration, seed, sequence length and corpus digest, or it is an\n  anecdote.\n- Present Wildcat-shaped law as universal law. Every law states the accounting\n  model it assumes, and a law with no applicability contract is not in the\n  corpus.\n- Decide whether a law is true. The checker decides whether something is a law\n  at all, which is a lower bar and the one worth enforcing mechanically.\n\n## Promise Machine contract\n\n### pandects-law-contract\n\n- Promise: A successful `pandects.py check` establishes that every discovered catalogue entry has the six required executable-law parts.\n- Evidence: The law registry, Solidity component and broken-specimen references, reduced counterexample, applicability, justified bound, judgement interface and zero-exit checker diagnostics.\n- Evidence classes: checked\n- Boundary: The check establishes that an entry has the law contract; it does not establish that the law is true, applies to a target or catches its specimen.\n- Authorises: Treating the entry as structurally eligible for the separate specimen, rendering and campaign checks.\n- Consequence: 1\n- Refuses: Admitting an entry missing executable judgement, a broken specimen, reduction, applicability, observables or justified bounds.\n- Recovery: Supply the named missing part and rerun `check` before any dependent operation.\n- Exceptions: none\n\n### pandects-catalogue-render\n\n- Promise: A successful `pandects.py render` followed by the catalogue-byte regression check reproduces `docs\u002Fcatalogue.md` from the checked authored registry.\n- Evidence: The checked law registry, deterministic render output, committed catalogue bytes and passing byte-comparison test.\n- Evidence classes: checked, recomputed\n- Boundary: Rendering establishes agreement with the registry; it does not establish law truth, applicability to a target or specimen coverage.\n- Authorises: Publication of the regenerated catalogue as a derived view of the checked law records.\n- Consequence: 1\n- Refuses: Hand-editing the rendered catalogue, publishing drifted bytes or using rendering as evidence that a law holds.\n- Recovery: Correct the authored registry, rerun `check` and `render`, then rerun the exact catalogue comparison.\n- Exceptions: none\n\n### pandects-broken-specimen\n\n- Promise: A green `forge test` establishes the shipped diagonal relation: each law fails its named broken specimen and holds against the other shipped specimens under the deterministic harness.\n- Evidence: The exact law and specimen Solidity, reduced replay, compiled harness and complete passing Foundry test result.\n- Evidence classes: checked, recomputed\n- Boundary: The diagonal establishes the named fixture relation only; it is not a formal proof, a verdict on another protocol or evidence that a search engine explored a target.\n- Authorises: Presenting the law as demonstrated to catch its shipped defect and using the specimen as a regression oracle.\n- Consequence: 1\n- Refuses: Calling a law proven to catch a defect when its diagonal test did not run, failed or was replaced by a tautological campaign result.\n- Recovery: Reduce the failing relation, correct the law or specimen without weakening the oracle and rerun the full diagonal.\n- Exceptions: none\n\n### pandects-search-record\n\n- Promise: A successful `pandects.py run` records the engine that actually ran, its argv, configuration, determinism, sequence length, disposition and corpus digest.\n- Evidence: The executed engine command, captured configuration, declared per-law exercise map, readable test and invariant-suite summaries, decoded positive succession call count, timeout or result status and generated search-record JSON.\n- Evidence classes: measured, recorded, checked\n- Boundary: The record establishes what the named campaign searched and reported; it does not decide law truth, cover an absent engine, prove universal safety or make a non-applicable law fit a target.\n- Authorises: Reporting the bounded campaign result or attaching its command record to an Ariadne release statement.\n- Consequence: 3\n- Refuses: Inventing an engine result, treating timeout as pass or fail, hiding zero succession calls, or presenting target-specific applicability as universal credit law.\n- Recovery: Correct the adapter or applicability record, run the intended engine with explicit settings and preserve a new search record.\n- Exceptions: none\n","\u003Cp align=\"center\">\n  \u003Cimg src=\"..\u002F..\u002Fassets\u002Fcharacters\u002Fpandects.png\" width=\"1200\">\n\u003C\u002Fp>\n\n# Pandects\n\n## Frontier\n\nPandects owns its own executable-law frontier, not Hexaemeron's delivery or\nSolidity frontier. Its version, held target, next job, and maturity\nstate live in [EVOLUTION.md](EVOLUTION.md). Do not recommend or run\nanother frontier pass after that ledger becomes mature.\n\n\u003C!-- marketplace-context:start -->\n## Where this sits\n\nPandects supplies executable credit laws, each paired with a deliberately broken specimen it is proved to catch and a replayable counterexample.\n\n**Current frontier.** The search-record runner records only the Foundry campaign, so Echidna and Medusa results survive as audit prose rather than as records.\n\u003C!-- marketplace-context:end -->\n\nFizz or another engine can search these laws, Janus can apply the relevant\nones around a hook transition, and Hermes must preserve them when changing gas.\nPandects states and tests the law; it does not generate a target-specific fuzz\nharness, audit the whole protocol, or turn one bounded campaign into security\nsign-off.\n\nSynkrisis compares validated run observations. Its shipped comparison cannot\npromote repeated campaign outcomes into a law or security verdict, and a\nPandects campaign result is not a run observation merely because it exists.\n\nA fuzzer searches a state space. It cannot decide which economic facts must\nsurvive that search. Generic token property libraries cover balances, approvals\nand round trips; credit adds time, accrual, queues, changing terms,\ndelinquency, partial repayment and claims that are neither liquid nor reducible\nto a token balance. Every audit rediscovers those facts in prose, implements a\nsubset in a local harness, and the harness dies with the engagement.\n\nThis is the corpus that survives it.\n\n`$SKILL_DIR` is the directory holding this file. The tool lives at\n`$SKILL_DIR\u002F..\u002F..\u002Fscripts\u002Fpandects.py`; resolve it from where you loaded this\nskill.\n\n## Day to day\n\n**Security and audit.** A credit protocol arrives and the invariants have to be\nworked out before anything can be fuzzed. The corpus states them, says which\naccounting model each one assumes, and ships the broken contract each is proven\nto catch, so a law that holds against your target is a law you can see working.\n\n**Developers.** A change touches accrual or the withdrawal queue and the\nquestion is what it might have broken. Point the corpus at the build and read\nwhich laws stopped holding, with the quantities that were compared.\n\n## What a law is\n\nSix parts. Anything with fewer is not a law, and the checker refuses it:\n\n1. **It executes.** A Solidity component, not a sentence.\n2. **It catches something.** A deliberately broken contract it is proven to\n   fail against. A law that has never failed may be a tautology; a property\n   asserting `seen() >= 0` survives a campaign of five hundred calls without\n   complaint, and nothing in that run distinguishes it from a real law.\n3. **It has been reduced.** A minimal counterexample, replayable without a\n   fuzzer.\n4. **It says where it applies.** The accounting model, the assumptions and the\n   observables it requires.\n5. **Its bounds are justified.** Exact, or a tolerance naming the arithmetic\n   that produces it. An epsilon chosen because it made a test pass is refused.\n6. **It judges rather than reverts.** A law returns `(bool held, string\n   detail)`.\n\nThat last one is the decision worth knowing about. A campaign runs with\n`fail_on_revert = false`, because a credit system reverts constantly and\ncorrectly. Under that setting a revert carries no verdict, so a law using\n`require` to mean \"violated\" reports nothing and is counted as silence. Silence\nread as assent is what this exists to prevent.\n\n## Two shapes\n\nConservation is a fact about one state, so a `Law` reads one target. Accrual is\nnot: debt falling without payment cannot be violated by any single state,\nbecause the violation is in the transition. A `PairLaw` judges two\n`Observation`s instead, each a snapshot of everything a target reports.\n\nThree pair laws compare a system with its own past. One compares two systems\nstarted identically and advanced over the same span by different routes, which\nis the only way to see interest compounding when it should not. A law says\nwhich kind of pair it means in its applicability, and handed one it cannot\njudge it either holds or refuses on a single rule: a pair that spans real time\nis a state of the world, and a pair nobody could have meant is a mistake by\nwhoever built it.\n\n## The commands\n\n```bash\npython3 scripts\u002Fpandects.py laws\npython3 scripts\u002Fpandects.py check\npython3 scripts\u002Fpandects.py run --out search-record.json\npython3 scripts\u002Fpandects.py render\n```\n\n`laws` lists the catalogue with each law's applicability. `check` refuses a law\nmissing any of its six parts, naming the part rather than the file. `run`\nsearches and writes down how it searched. `render` writes `docs\u002Fcatalogue.md`\nout of the catalogue, which is the only thing that should ever write it.\n\nExit codes: 0 success, 1 a check failed, 2 usage or validation error.\n\n```bash\nforge test\n```\n\nruns every law against every specimen and asserts the diagonal: each law fails\nthe specimen written for it and holds against the others. The corpus takes no\nexternal Solidity dependency: there is no `lib\u002F`, no submodule and nothing to\nfetch.\n\nThe catalogue holds ten laws in three families. Conservation: value is\nconserved across held assets, debt, claims and fees; assets reserved never\nexceed the claims recorded; and reserved plus borrowable never exceed what is\nheld. Accrual: debt falls only against assets arriving, rises at rest only\nagainst assets leaving, and the same span costs the same however it is cut up.\nWithdrawal claims: a recorded claim is never written down smaller, nobody is\npaid ahead of somebody who has waited longer, and what a system declares\npayable it has set aside; pooled lender claims cover what is still owed on open\nwithdrawal batches.\n\nNine are exact. `path-independent` carries the corpus's only tolerance, and the bound\nis derived rather than chosen: linear accrual on principal truncates once per\nstep, so `n` small steps differ from one long step by at most `n - 1` units.\n\nTo point another engine at it, `src\u002Fcampaigns\u002FSpecimens.sol` carries a harness\nper specimen declaring both `echidna_` and `property_` prefixes, since Echidna\nlooks for the first and Medusa defaults to the second. Every entry point\nsnapshots the system on the way in, which is what lets a pair law read a past.\n`explain()` returns each law's reason for the current state, so a failing\nsequence gives up its reason rather than making somebody derive it from a call\ntrace.\n\nOne harness passes everything and the contract underneath it is broken.\n`CompoundsPerStepCampaign` cannot see its own defect, because path independence\ncompares two systems advanced by different routes and a campaign drives one\nalong one route. It is the plainest illustration here of what a clean campaign\nis worth on its own.\n\n## Pointing it at a protocol\n\n`adapters\u002F` is where the corpus meets somebody else's system. Two shapes, and\nthe difference is a limit rather than a convenience.\n\n`CorpusObserver` holds any address and offers the six one-state laws. Nothing\nroutes calls through it, so it offers no pair law: one that never sees a call\nhas no past to compare with, and would report holding forever.\n\n`CorpusDriver` takes the calls. Extend it, write the protocol's entry points,\nput `records` on every one that changes state, and the three succession laws\nrun as well. Nothing catches a forgotten `records` -- it looks like three laws\nholding quietly -- so `recordedCalls` is public and `successionExercised` reads\nit. Zero recorded calls means those laws searched nothing, whatever they\nreported.\n\nPath independence is in neither, because it compares two systems and an adapter\nfronts one. `PathIndependenceProbe` takes both.\n\n## Saying how it was searched\n\nA campaign result without its settings is an anecdote, which is the same\nrefusal the corpus already makes of a tolerance that does not name its\narithmetic. `run` writes a record naming the engine, the argv, the determinism\nclass, the configuration read out of `foundry.toml` rather than restated, the\nsequence length and a digest of the corpus searched. It is shaped as an\n`ariadne` command entry, so a result drops into a release statement without\ntranslation.\n\nAn engine that did not run is absent from the record, not present and empty.\nWhat nobody could read is absent rather than null. And a campaign killed by the\ntimeout is neither passed nor failed: it ran, searched less than it was asked\nto, and says so.\n\n## What a law may read\n\n`ICreditObservables`. It names economic roles: the asset, total assets held,\ntotal debt, total lender claims, reserved and borrowable assets, accrued fees,\nand the time the observation describes. A target implements it, or a thin\nadapter does, and that adapter is the only place a protocol's own names appear.\n\nA system with a withdrawal queue also implements\n`IWithdrawalQueueObservables`: the claim count, each claim's owed and paid\namounts, and the bound through which the system declares claims payable. Kept\nseparate because a system with no queue would implement three members and mean\nnone of them, and an observable that means nothing reports zero, which reads\nlike an answer. A target that lacks it reverts on the read, and a revert is no\nverdict rather than a false one.\n\nThis is enforced rather than requested. A property written against one\nprotocol's function names is a property about one codebase, and the corpus\nexists because the same economic facts hold across codebases that share nothing\nelse.\n\n## Pointed at a real design\n\n`integrations\u002Fwildcat\u002F` is the corpus against a codebase it was not written\nfor: a reduced model of a Wildcat market with batched withdrawals, a reserve\nthe borrower may not touch, delinquency and penalty accrual.\n\nSeven laws apply flatly. Three do not, and the three are what the integration is\nfor.\n\n`queue-order-preserved` applies at batch granularity and says nothing per\nlender, because a batch is paid pro rata. `path-independent` holds while the\nmarket is solvent and stops once penalty accrual runs, because the grace timer\nadvances when the market is poked. And `recorded-claim-never-shrinks` does not\nhold over a batch that is still open: a batch accumulates while open, so the\namount owed on it rises, and the law says a recorded claim keeps its amount.\n\nThat third one was found by Echidna against the shipped adapter, after the\napplicability notes had already claimed the law held. Reading found two\nconditions; running the corpus against a real design found the third, in a\ndocument that had already got it wrong.\n\nNone of the three is a yes or a no.\n`integrations\u002Fwildcat\u002FAPPLICABILITY.md` carries all of them with their reasons.\n\n## The documents\n\n`docs\u002Fcatalogue.md` renders every law for a reader, and it is a rendering\nrather than a second source: `pandects render` writes it, and a test compares\nthe committed bytes with what the renderer produces. `docs\u002Fwriting-a-law.md`\nhas the six parts in the order somebody adding a law meets them.\n`docs\u002Fapplicability.md` states the rules once -- what a law may read, the two\nshapes, what a pair law does with a pair it cannot judge, and what a revert\nmeans.\n\n## What this never does\n\n- Replace an engine. Echidna, Medusa and Foundry search; the corpus says what\n  to look for.\n- Generate a harness for one repository. That is `fizz`, which can select and\n  adapt these laws and stays useful where no public law describes the protocol.\n- Report a campaign under an engine that did not run. A result carries its\n  engine, configuration, seed, sequence length and corpus digest, or it is an\n  anecdote.\n- Present Wildcat-shaped law as universal law. Every law states the accounting\n  model it assumes, and a law with no applicability contract is not in the\n  corpus.\n- Decide whether a law is true. The checker decides whether something is a law\n  at all, which is a lower bar and the one worth enforcing mechanically.\n\n## Promise Machine contract\n\n### pandects-law-contract\n\n- Promise: A successful `pandects.py check` establishes that every discovered catalogue entry has the six required executable-law parts.\n- Evidence: The law registry, Solidity component and broken-specimen references, reduced counterexample, applicability, justified bound, judgement interface and zero-exit checker diagnostics.\n- Evidence classes: checked\n- Boundary: The check establishes that an entry has the law contract; it does not establish that the law is true, applies to a target or catches its specimen.\n- Authorises: Treating the entry as structurally eligible for the separate specimen, rendering and campaign checks.\n- Consequence: 1\n- Refuses: Admitting an entry missing executable judgement, a broken specimen, reduction, applicability, observables or justified bounds.\n- Recovery: Supply the named missing part and rerun `check` before any dependent operation.\n- Exceptions: none\n\n### pandects-catalogue-render\n\n- Promise: A successful `pandects.py render` followed by the catalogue-byte regression check reproduces `docs\u002Fcatalogue.md` from the checked authored registry.\n- Evidence: The checked law registry, deterministic render output, committed catalogue bytes and passing byte-comparison test.\n- Evidence classes: checked, recomputed\n- Boundary: Rendering establishes agreement with the registry; it does not establish law truth, applicability to a target or specimen coverage.\n- Authorises: Publication of the regenerated catalogue as a derived view of the checked law records.\n- Consequence: 1\n- Refuses: Hand-editing the rendered catalogue, publishing drifted bytes or using rendering as evidence that a law holds.\n- Recovery: Correct the authored registry, rerun `check` and `render`, then rerun the exact catalogue comparison.\n- Exceptions: none\n\n### pandects-broken-specimen\n\n- Promise: A green `forge test` establishes the shipped diagonal relation: each law fails its named broken specimen and holds against the other shipped specimens under the deterministic harness.\n- Evidence: The exact law and specimen Solidity, reduced replay, compiled harness and complete passing Foundry test result.\n- Evidence classes: checked, recomputed\n- Boundary: The diagonal establishes the named fixture relation only; it is not a formal proof, a verdict on another protocol or evidence that a search engine explored a target.\n- Authorises: Presenting the law as demonstrated to catch its shipped defect and using the specimen as a regression oracle.\n- Consequence: 1\n- Refuses: Calling a law proven to catch a defect when its diagonal test did not run, failed or was replaced by a tautological campaign result.\n- Recovery: Reduce the failing relation, correct the law or specimen without weakening the oracle and rerun the full diagonal.\n- Exceptions: none\n\n### pandects-search-record\n\n- Promise: A successful `pandects.py run` records the engine that actually ran, its argv, configuration, determinism, sequence length, disposition and corpus digest.\n- Evidence: The executed engine command, captured configuration, declared per-law exercise map, readable test and invariant-suite summaries, decoded positive succession call count, timeout or result status and generated search-record JSON.\n- Evidence classes: measured, recorded, checked\n- Boundary: The record establishes what the named campaign searched and reported; it does not decide law truth, cover an absent engine, prove universal safety or make a non-applicable law fit a target.\n- Authorises: Reporting the bounded campaign result or attaching its command record to an Ariadne release statement.\n- Consequence: 3\n- Refuses: Inventing an engine result, treating timeout as pass or fail, hiding zero succession calls, or presenting target-specific applicability as universal credit law.\n- Recovery: Correct the adapter or applicability record, run the intended engine with explicit settings and preserve a new search record.\n- Exceptions: none\n",{"name":5,"metadata":24,"description":26},{"version":25},"1.1.0","Check a credit protocol against executable laws: conservation, accrual and withdrawal claims, each one a Solidity component with a broken specimen it is proven to catch. Use when someone asks which invariants a lending or credit system should hold, wants properties for a fuzzing campaign, or hands over a protocol and asks what could be checked mechanically. Do not use it to generate a harness for one repository; that is fizz. Never report a campaign under an engine that did not run.\n","main","6b6406ecd922c4127bc688e0b5122bbe395be0eb","plugins\u002Fpandects\u002Fskills\u002Fpandects",6,"https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills","https:\u002F\u002Fgithub.com\u002Fwildcat-finance","https:\u002F\u002Fwildcat.finance","Banking, But Worse","2026-11-07 17:14:07",{"source":8,"skillName":5,"commands":37,"prompt":47,"zipAvailable":48},[38,41,44],{"tool":39,"command":40},"npx","npx skills add wildcat-finance\u002Fskills --skill pandects",{"tool":42,"command":43},"bunx","bunx skills add wildcat-finance\u002Fskills --skill pandects",{"tool":45,"command":46},"pnpm","pnpm dlx skills add wildcat-finance\u002Fskills --skill pandects","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fpandects\u002Fskills\u002Fpandects\u002FSKILL.md",true,[50,56,60],{"path":51,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":54,"sourceHtmlUrl":55},"agents\u002Fopenai.yaml","REFERENCE","参考文档",273,"https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fpandects\u002Fskills\u002Fpandects\u002Fagents\u002Fopenai.yaml",{"path":57,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":58,"sourceHtmlUrl":59},"DEMONSTRATION.md",2468,"https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fpandects\u002Fskills\u002Fpandects\u002FDEMONSTRATION.md",{"path":61,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":62,"sourceHtmlUrl":63},"EVOLUTION.md",2357,"https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fpandects\u002Fskills\u002Fpandects\u002FEVOLUTION.md",[65,72,78,85,91,97],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7APAWXHFSKVQEJBAHKASH25","example","A deliberately unsupported evidence class fixture.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Funsupported-evidence-class\u002Fplugins\u002Fexample\u002Fskills\u002Fexample\u002FSKILL.md",[],[],{"skillNo":73,"name":67,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":19},"01M7APATBK93107VF5NMPA3RJH","A router fixture target.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Funresolved-router\u002Fplugins\u002Fexample\u002Fskills\u002Fexample\u002FSKILL.md",[],[],{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":19},"01M7APARV9C0MQB8WTB0GGAAMG","promise-machine","A deliberately unresolved router.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Funresolved-router\u002F.agents\u002Fskills\u002Fpromise-machine\u002FSKILL.md",[],[],{"skillNo":86,"name":67,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M7APAQMDG4977HXGSKMB29KV","A deliberately unclassified skill fixture.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Funclassified-skill\u002Fplugins\u002Fexample\u002Fskills\u002Fexample\u002FSKILL.md",[],[],{"skillNo":92,"name":67,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M7APAPE6DE05A24J4BYPK6WY","A deliberately unattributed exception fixture.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Funattributed-exception\u002Fplugins\u002Fexample\u002Fskills\u002Fexample\u002FSKILL.md",[],[],{"skillNo":98,"name":67,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":19},"01M7APAKYXRTABBSN7Q637CC6M","A deliberately conflated version fixture.","https:\u002F\u002Fgithub.com\u002Fwildcat-finance\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fpromise-machine\u002Fpackage-as-skill-version\u002Fplugins\u002Fexample\u002Fskills\u002Fexample\u002FSKILL.md",[],[]]