{"id":50391167,"url":"https://github.com/kadubon/certified-workflow-conversion","last_synced_at":"2026-05-30T18:01:49.215Z","repository":{"id":356000032,"uuid":"1230581772","full_name":"kadubon/certified-workflow-conversion","owner":"kadubon","description":"Evidence-bound workflow diagnostics and certified lower-bound reporting for long-running AI agent pipelines. Improve agent workflow throughput without changing the model.","archived":false,"fork":false,"pushed_at":"2026-05-06T06:23:33.000Z","size":130,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-06T08:34:39.019Z","etag":null,"topics":["agent-infrastructure","agent-workflows","ai-agent","ai-workflows","bottleneck-analysis","certified-throughput","evidence-contracts","evidence-ledger","fail-closed","long-running-agents","model-independent","python","workflow-optimization","workflow-verification"],"latest_commit_sha":null,"homepage":"https://kadubon.github.io/github.io/works.html","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"other","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/kadubon.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-06T06:13:12.000Z","updated_at":"2026-05-06T06:23:36.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/kadubon/certified-workflow-conversion","commit_stats":null,"previous_names":["kadubon/certified-workflow-conversion"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/kadubon/certified-workflow-conversion","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kadubon%2Fcertified-workflow-conversion","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kadubon%2Fcertified-workflow-conversion/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kadubon%2Fcertified-workflow-conversion/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kadubon%2Fcertified-workflow-conversion/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/kadubon","download_url":"https://codeload.github.com/kadubon/certified-workflow-conversion/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/kadubon%2Fcertified-workflow-conversion/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":33703065,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-05-30T02:00:06.278Z","response_time":92,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-infrastructure","agent-workflows","ai-agent","ai-workflows","bottleneck-analysis","certified-throughput","evidence-contracts","evidence-ledger","fail-closed","long-running-agents","model-independent","python","workflow-optimization","workflow-verification"],"created_at":"2026-05-30T18:01:47.951Z","updated_at":"2026-05-30T18:01:49.199Z","avatar_url":"https://github.com/kadubon.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Certified Workflow Conversion\n\nCertified Workflow Conversion is an evidence-bound, fail-closed Python toolkit\nfor diagnosing bottlenecks and reporting conservative lower-bound throughput in\nlong-running AI agent workflows.\n\nUse it for:\n\n- agent pipeline bottleneck diagnosis;\n- validation, release, and rollback capacity analysis;\n- evidence-bound certified lower-bound reporting.\n\n`certified-workflow-conversion` (`cwc`) improves AI agent pipelines without\nchanging the model. It treats an agent deployment as a workflow conversion\nnetwork: candidates become usable value only after generation, tool execution,\nvalidation, review, authorization, memory governance, release, rollback, and\nincident handling.\n\nCWC answers a narrow operational question:\n\n\u003e Given the current evidence ledger, which workflow layer is limiting\n\u003e evidence-bound certified throughput?\n\nIt does not certify factual truth, model intelligence, alignment, or production\nsafety. A certified CWC report means only that a lower-bound workflow claim is\nprocedurally admissible under declared evidence, contracts, witnesses, and\nchecks.\n\n## Why This Exists\n\nModel quality is not the only bottleneck in long-running AI systems. A stronger\nmodel can still fail to create usable output when validation queues, release\ngates, authority checks, memory governance, rollback capacity, or incident\nresponse are the binding constraints.\n\nCWC makes those constraints machine-readable:\n\n- typed evidence objects describe what is known and where it applies;\n- a monotone claim compiler rejects unsupported claims;\n- conversion networks model workflow edges and capacities;\n- hard gates block rather than become finite penalties;\n- diagnostic analyzers identify bottlenecks;\n- full-profile reports require evidence contracts and accepted verification\n  witnesses for every report-facing term.\n\n## What Is Unique\n\n- **Workflow-first, model-independent:** no model provider is required. CWC\n  analyzes the pipeline around the model.\n- **Evidence-bound, not memory-only:** raw observations are not treated as\n  certified throughput. Claims must be backed by active typed evidence.\n- **Fail-closed certification path:** missing support, inactive evidence,\n  missing TCB roots, failed hard gates, malformed witnesses, or unbound report\n  terms block the report.\n- **Report-term binding:** in `full` mode, each `EvidenceContract` exposes\n  exactly one claim-facing term, and each accepted `VerificationWitness` must\n  bind that term to deterministic numeric output and the source evidence it\n  depends on.\n- **Bottleneck investment signals:** dual prices and diagnostic scores can\n  suggest where to invest next, while remaining separate from adoption claims.\n- **Ports and adapters:** SQLite is the default local backend, not an\n  architectural assumption. Storage, analyzers, optimizers, OAWM bridges, and\n  report sinks are replaceable.\n\n## What You Can Do\n\n- Store typed evidence in a local append-oriented ledger.\n- Register conversion networks for AI workflows.\n- Compile claims against evidence, scope, dependency, expiry, and TCB checks.\n- Run lightweight diagnostic bottleneck analysis.\n- Run optional SciPy-backed full-profile lower-bound checks.\n- Import certified OAWM state as read-only evidence.\n- Build custom storage backends, analyzers, checkers, and report sinks.\n\n## Analysis Profiles\n\n| Profile | Purpose | Dependencies | Claim Strength |\n| --- | --- | --- | --- |\n| `light` | Fast local bottleneck screening | base install | diagnostic only |\n| `full` | Evidence-contract lower-bound reports | `--extra full` | procedural lower bound under supplied evidence |\n\n`light` mode is useful for engineering triage. It is not a full statistical\ncertificate, and `certified_lower_bound` requests in `light` profile fail\nclosed rather than returning a weaker pseudo-certificate.\n\n`full` mode checks active evidence, three-way reporting splits, one-term\ncontracts, accepted witnesses, confidence budgets, source/sink declarations,\nstatistical or path-law certificates, queue certificates, release accounting,\nGoodhart/open-world charges, and validation-capital root cuts.\n\nFor every report-facing term, the contract must also depend on the evidence that\nproduced the term. For example, an `edge.capacity:*` contract must depend on the\nedge support evidence, and a `statistical_lower` contract must depend on the\nstatistical certificate evidence.\n\nThe `full` lower bound is composed conservatively:\n\n```text\nfloor(max(0,\n  min(flow, statistical/path-law/report-term lower bounds)\n  - queue boundary\n  - direct cost rate\n  - Goodhart charge\n  - open-world charge\n))\n```\n\nRaw network flow is never enough by itself.\n\n## Ten-Minute Local Run\n\n```powershell\nuv sync --extra dev\nuv run python examples/coding_agent_pipeline/run_demo.py\n```\n\nThat example creates local evidence, registers a coding-agent pipeline, compiles\na claim, analyzes bottlenecks, and prints investment candidates. It uses no API\nkeys and makes no network calls.\n\nCLI workflow:\n\n```powershell\nuv run cwc init .cwc\nuv run cwc evidence add examples/coding_agent_pipeline/evidence.jsonl --state .cwc\nuv run cwc network add examples/coding_agent_pipeline/network.json --state .cwc\nuv run cwc audit --state .cwc\n```\n\nFor the full profile:\n\n```powershell\nuv sync --extra dev --extra full\nuv run python examples/full_certified_lower_bound.py\nuv run python examples/dual_price_interval.py\nuv run python examples/validation_capital_root_cut.py\n```\n\n## Minimal Python API\n\n```python\nfrom certified_workflow_conversion.core.models import (\n    ClaimRequirement,\n    ConversionNetwork,\n    ServiceEdgeProfile,\n    TypedEvidenceObject,\n)\nfrom certified_workflow_conversion.runtime.kernel import ConversionKernel\n\nkernel = ConversionKernel.open(\".cwc\")\n\nevidence = kernel.add_evidence(\n    TypedEvidenceObject.create(\n        kind=\"validation\",\n        scope=\"demo\",\n        source=\"local-test\",\n        payload={\"passed\": True},\n    )\n)\n\nedge = ServiceEdgeProfile.create(\n    name=\"validation\",\n    from_node=\"candidate\",\n    to_node=\"accepted\",\n    capacity=5,\n    evidence_ids=[evidence.evidence_id],\n)\n\nnetwork = kernel.register_network(\n    ConversionNetwork.create(\n        name=\"demo\",\n        nodes=[\"candidate\", \"accepted\"],\n        source_nodes=[\"candidate\"],\n        sink_nodes=[\"accepted\"],\n        edges=[edge],\n    )\n)\n\nclaim = kernel.compile_claim(\n    ClaimRequirement.create(\n        network_id=network.network_id,\n        target_value=4,\n        required_evidence_ids=[evidence.evidence_id],\n    )\n)\n\nreport = kernel.analyze(network.network_id, claim.claim_id)\n```\n\nFull-profile calls use the same kernel:\n\n```python\nreport = kernel.analyze(\n    network.network_id,\n    claim.claim_id,\n    mode=\"certified_lower_bound\",\n    profile=\"full\",\n)\n```\n\n## Core Concepts\n\n| Concept | Meaning |\n| --- | --- |\n| `TypedEvidenceObject` | Active, scoped evidence with dependencies, expiry, TCB requirements, and digest binding |\n| `ConversionNetwork` | Directed workflow graph whose edges represent services such as generation, validation, release, memory, or recovery |\n| `ClaimRequirement` | A requested lower-bound claim over a network and target value |\n| `CompiledClaim` | A monotone compilation result; unsupported claims remain audit-visible |\n| `EvidenceContract` | A machine-checkable contract exposing exactly one report-facing term |\n| `VerificationWitness` | Accepted checker output binding a contract to inputs, scope, checker digest, TCB, and numeric result |\n| `BottleneckReport` | Diagnostic or certified-lower-bound report with limitations and evidence ids |\n\n## Extensibility\n\nCWC uses ports and adapters:\n\n- `cwc.storage_backends`: SQLite, PostgreSQL, DuckDB, object-store, enterprise DB.\n- `cwc.analyzers`: deterministic, optimization-backed, or domain-specific analyzers.\n- `cwc.optimizers`: investment search strategies.\n- `cwc.oawm_bridges`: importers from external agent-memory systems.\n- `cwc.report_sinks`: JSON, Markdown, database, dashboard, or observability export.\n\nThe core package does not import SQLite, cloud SDKs, model providers, or OAWM.\nBackend authors can run the storage contract helpers in\n`certified_workflow_conversion.testing.contracts`.\n\n## Security And Limitations\n\n- CWC is not a sandbox, credential manager, policy engine, or external-effect\n  gateway.\n- CWC does not execute tools or release actions.\n- Certified throughput does not mean factual truth or model truthfulness.\n- `light` profile is diagnostic only.\n- `full` profile is fail-closed and evidence-bound, but only as strong as the\n  supplied evidence, contracts, witnesses, roots, and domain-specific checkers.\n- Full-profile reports currently use normalized unit throughput; richer value\n  accounting should be implemented in domain analyzers.\n- SQLite state is trusted local state in this beta.\n- Semantic validity depends on domain-specific evidence and checker plugins.\n- TCB requirements require rooted, active TCB evidence. Declaring a requirement\n  is not proof that the TCB is healthy.\n- Dual prices are local planning signals, not adoption authorization.\n- APIs and schemas may change before a stable non-beta release.\n\nFor production systems, external effects still need OS, network, identity,\nsecrets, sandboxing, audit, and recovery controls outside CWC.\n\n## Documentation\n\n- [Theory mapping](docs/theory_mapping.md)\n- [Full certification profile](docs/full_certification.md)\n- [Reporting protocol](docs/reporting_protocol.md)\n- [Security model](docs/security_model.md)\n- [Backend author guide](docs/backend_author_guide.md)\n- [Plugin guide](docs/plugin_guide.md)\n- [Release checklist](docs/release_checklist.md)\n\n## Development Checks\n\n```powershell\nuv run pytest\nuv run ruff check .\nuv run mypy src\n```\n\nFull-profile development:\n\n```powershell\nuv sync --extra dev --extra full\nuv run pytest tests/full\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkadubon%2Fcertified-workflow-conversion","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fkadubon%2Fcertified-workflow-conversion","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkadubon%2Fcertified-workflow-conversion/lists"}