Skip to content

Verification Evidence

Status: public evidence summary, 2026-11-09.

This page indexes the verification claims PRAXO-CORE can support today. It is a public summary, not a production certificate, penetration test, or claim that every implementation and deployment inherits the same result. Raw run records and full environment logs remain in restricted internal storage; this page records the claim level, result, and remaining limits.

For the EC-01 through EC-14 mapping across Core and its product consumers, see the Execution Correctness Evidence Index.

See the test evidence standard for the required environment, configuration, procedure, data, result, and findings record.

Claim Levels

Level Meaning
Contract/unit Deterministic repository tests, fakes, vectors, or API snapshots
Local integration Real local binaries and databases on loopback, with disposable state
Independent/production Reproduced by an independent party or observed in a deployed environment

Current public evidence does not establish the third level unless a row says so. A source tag, complete matrix label, or passing vector suite is not production evidence.

Current Evidence

Area Verified Level Limits
Go Core Tests, vet, build, API snapshot, race, fuzz and module-security CI; current main and release workflow checks; v2.8.0 is the current release (2026-11-14); the current-go-consumer CI job installs the recommended version from the registry and runs deterministic allow/deny checks; v2.6.0 resolves from the public module proxy in a clean external module; a separately recorded clean external-module install and minimal-agent smoke cover v2.5.2 Contract/unit + CI + local integration JSON Schema and plugin manifest fuzz runs have hit a 10s deadline and passed on rerun; slow-input risk remains under investigation; a non-CI clean install smoke for v2.8.0 was not rerun; the recorded install covers one platform and a minimal agent
Go long-running recipe File-backed store with idempotent submit, response loss, child-process restart, separate verifier, fencing with lease expiry/renewal/revocation, deterministic pre/post-rename fault injection, fail-closed corrupt state, a standalone service binary with health/readiness, basic metrics, example Prometheus scrape/retention/alert configs (config-validated; no live collector), environment-based configuration, file-based token/secret rotation (issuance external), and graceful shutdown, configurable file/SQLite/Postgres lease backends with reload-modify-save (SQLite verified; Postgres gated by PRAXO_POSTGRES_DSN and unverified here), process-level crash before and after the rename boundary, a gated 100-submission load/fault test with a fresh-checkout reproduction script, a separate verifier binary with its own token/secret, and a no-lease concurrent-writer boundary test Contract/unit + local integration L4 application recipe; no production deployment, collector/retention/alerting, OS-level power loss, production lease coordination, trusted clock, or exactly-once
Cross-language conformance Python, TypeScript, Java, Rust, Fortran, and COBOL compatibility lab runs deterministic vector, API snapshot/signature, and parity checks; Linux CI and local runs are recorded Contract/unit + CI Six languages only; COBOL was verified locally with GnuCOBOL and on Linux CI; no provider, HITL, checkpoint, sandbox, or production evidence; one platform and toolchain set per run; non-author reproduction pending
Python published package PyPI praxo-core==0.4.1 is published from source tag python/v0.4.1. The release workflow built and smoke-tested the wheel and sdist before publication; the package includes Harmovela and MCP adapters. Historical local checks cover 0.3.2 and 0.2.4 Registry publication + release-workflow artifact smoke + post-publication consumer check One platform/toolchain; same-machine clean venv, not an independent or non-author consumer; no provider or production evidence
Python published-package upgrade scripts/test_python_upgrade.py installs PyPI 0.2.4, runs the unchanged consumer fixture and governed compatibility checks, upgrades to 0.3.0, and reruns the full starter/governed suite CI + registry upgrade One version pair; does not prove arbitrary upgrade paths, independent consumers, or production behavior
TypeScript published package npm @axisvela/praxo-core@0.3.0 is published from source tag typescript/v0.3.0 and is the latest version. The release workflow built and consumed the package tarball; 0.3.0 README includes absolute registry/coding-agent links; the post-publication consumer script installed 0.3.0 from npm in a clean project and passed the starter checks locally; the governed negative suite was not rerun for 0.3.0. The historical smoke covers 0.1.5 Registry publication + release-workflow tarball smoke + post-publication consumer check One platform/toolchain; not an independent consumer; no provider or production evidence
TypeScript published-package upgrade scripts/test_typescript_upgrade.py installs npm 0.1.5, runs the unchanged consumer and governed compatibility checks, upgrades to 0.2.0, and reruns the full starter/governed suite CI + registry upgrade One version pair; does not prove arbitrary upgrade paths, independent consumers, or production behavior
Rust published crate crates.io praxo-core==0.4.0 is published from source tag rust/v0.4.0; the recorded clean temporary-project install smoke covers 0.1.4 Registry publication + historical local install smoke The registry smoke covers 0.1.4, not 0.4.0; one platform; not an independent consumer; no provider or production evidence
Dart published package praxo_core==0.1.1 installed from pub.dev into a clean project; the program imported package:praxo_core/praxo_core.dart and returned hello from a minimal agent Local install smoke One platform; same-machine project, not an independent or non-author consumer; minimal agent only, no provider or production evidence
Ruby published gem RubyGems praxo-core==0.4.1 is published from source tag ruby/v0.4.1; the gem source is axisvela/praxo-core-ruby. 0.4.0, 0.3.0, and 0.2.0 remain available. The recorded clean temporary-directory smoke covers 0.1.0 Registry publication + historical local install smoke Registry 0.4.1 smoke was not rerun; one platform; not an independent consumer; no provider or production evidence
Lua published rock LuaRocks praxo-core==0.4.1-1 is published through .github/workflows/lua-release.yml, which builds the source rock, installs it in a clean tree, and runs the installed smoke. The published 0.4.1-1 rock was also installed into a clean temporary tree and the consumer starter passed outside the repository; 0.4.0-1, 0.3.0-1, and 0.2.0-1 remain available Registry publication + CI install smoke + local consumer install One platform; not an independent consumer; no provider or production evidence
PHP published package Packagist publishes axisvela/praxo-core==0.7.0 from split tag v0.7.0; v0.4.1, v0.4.0, v0.3.0, and v0.2.0 remain available Registry publication No Composer install smoke is recorded for 0.7.0; the recorded starter install covers 0.4.1; one platform; not an independent consumer; no provider or production evidence
C# published package Praxo.Core==0.4.0 was built, tested, packed, and pushed to NuGet by GitHub Actions Trusted Publishing from source tag csharp/v0.4.0 CI + registry No consumer install smoke; one package version; no provider or production evidence
MATLAB File Exchange package PRAXO-CORE 0.3.0 is published at File Exchange entry 185166 with the generated toolbox image and MIT license Registry publication MATLAB runtime verification and consumer install smoke are not recorded; Octave verification does not establish MATLAB behavior
PowerShell Core PowerShell Gallery praxo-core-ps==0.2.0 is published from the versioned module. CI builds the C# Core, runs the C# conformance and behavior suites, wrapper smoke tests, and the PowerShell API snapshot Registry + contract/unit + CI Shared C# runtime, not a separate agent loop; behavior evidence is inherited from the C# Core; no consumer install smoke or provider evidence
Elixir published package Hex praxo_core==0.5.1 is published from source tag elixir/v0.5.1 with HexDocs 0.5.1; 0.5.0 failed before publication; 0.4.1, 0.4.0, 0.3.0, and 0.2.0 remain available. 0.4.1 resolved in a clean Mix project and the consumer starter passed locally; 0.5.1 install smoke was not rerun Registry publication + historical local consumer install One platform; not an independent consumer; no provider or production evidence
Erlang published package Hex praxo_core_erlang==0.4.0 is published from source tag erlang/v0.4.0; 0.3.1 and 0.2.0 remain available. The historical 0.1.1 row was fetched by a clean rebar3 project Registry publication + historical local install smoke Consumer install smoke was not rerun for 0.4.0; one platform; no independent consumer; no provider or production evidence
D published package praxo-core==0.1.0 fetched from DUB/code.dlang.org with dub fetch praxo-core Registry metadata + fetch No compile or independent consumer smoke; one package version; no provider or production evidence
Java published package Maven Central com.axisvela:praxo-core:0.5.0 is published with sources, javadoc, and GPG signatures. Central validated the deployment before publication; the post-publication consumer script resolved 0.4.1 from Maven Central and passed starter/governed checks locally; 0.5.0 consumer smoke was not rerun Registry publication + Central validation + historical post-publication consumer check One deployment/platform/toolchain; not an independent consumer; no provider or production evidence
Java published-package upgrade scripts/test_java_upgrade.py resolves Maven Central 0.3.0, runs the unchanged consumer and governed compatibility checks, upgrades to 0.4.0, and reruns the full starter/governed suite CI + registry upgrade One version pair; does not prove arbitrary upgrade paths, independent consumers, or production behavior
Extended consumer starters (Ruby/PHP/Elixir/Lua) Deterministic allow/deny starters in examples/consumer-ruby, consumer-php, consumer-elixir, and consumer-lua pass in source-mode locally. After configuring the local proxy CA, RubyGems praxo-core==0.4.1, Hex praxo_core==0.4.1, and LuaRocks praxo-core==0.4.1-1 installs were run outside the repository and the starters passed. Composer installed published axisvela/praxo-core==0.4.1 and the starter passed without a bridge; the package declares autoload.files and ships api/0.4.1.txt Contract/unit + local registry installs Extended-language starters; no governed negative matrix; one platform; no provider, identity, isolation, or production evidence
Cross-language agent behavior All sixteen wired languages pass all twenty authorization, lifecycle, cancellation, validation, call-limit, parallel-tool, run-isolation, PII, checkpoint, capacity, SSE-truncation, stream-failure, and HITL-decision scenarios in the source. Agent-capable languages use scenario-level or self-test coverage; extended languages use native-suite coverage. Parallel tools for COBOL, Lua, PHP, and Fortran are delegated to caller-provided batch callbacks. Other implementations are optional. The clean-tree CI report at b6218e5 records 320 passed, 0 unsupported, and no failures Scenario tests + CI All sixteen wired languages reach 20/20 in the source; some extended rows are suite-level and a failure is not isolated to a scenario; the earlier local run at 4ffd83b is historical, not current evidence; not a throughput/stress test, not full feature parity, provider, sandbox, or production evidence
Governed workspace Policy-required configuration, final-argument authorization, single-use approvals, post-approval edits, queued revocation, fail-closed dependency errors Contract/unit Example-level contract; not a Core profile, authenticated identity service, or durable approval store
Context trust Memory, skill, tool-result and summary source separation; digest-bound approved skills; text restore; late System injection rejection Contract/unit Opt-in example path; not a general prompt-injection filter, source authenticator, or durable lineage proof
Containment Windows path/process/WASM controls; Linux path and symlink controls in a verbose CI job; local VFS operations use os.Root after policy checks Contract/unit + CI Windows symlink cases skipped without privilege; policy decisions remain check-then-use; Linux cgroup v2 and independent review remain open; the pinned wazero API does not expose per-instruction fuel metering, so CPU bounding is deadline/cancellation-based rather than fuel-based
Workspace adapters Shared conformance and fake-provider tests for E2B, Daytona, OpenSandbox, CubeSandbox Contract/unit All live provider tests skipped without credentials; no contract promotion
OpenTelemetry Runnable in-memory example emits run.started, tool.call, run.completed; allowlist excludes tool arguments/results Example + unit No collector, OTLP backend, sampling, retention, or production telemetry verification
A2A interoperability Google ADK consumes published Core v2.5.1 A2A with no local Core replacement; success, denial, failure, timeout, authentication, TLS, cross-origin, and concurrent consumer fixtures Local integration Loopback deterministic fixture, not a deployed peer or production load test
Agent IAM contracts Direct and gateway-shaped Agent IAM examples against contract fakes; credential time-window, key, signature, and subject checks Contract/unit NOMIVELA/EIDOVELA/AEGIVELA responses are fakes; the demo has no proof replay cache or body-bound gateway assertion
Coding-agent skill discovery Claude Code 2.1.196 installed agent-skills/praxo-integration as a temporary skills-directory plugin; plugin list reported it loaded and plugin details listed the skill component (2026-11-08, bdd36d7; issue #15) Local structural check One tool and platform; that CLI did not run a model task or published-package integration; not adoption evidence
Coding-agent skill task opencode 1.18.18 discovered agent-skills/praxo-integration from a project-local .opencode/skills/praxo-integration/SKILL.md, then completed a deterministic Python task using published PyPI praxo-core==0.3.2 with explicit tool scope and allow/deny tests (3/3 passed; model deepseek/deepseek-flash; Python 3.14.6; 2026-11-08, 7004a08) Local structural check + local task execution One tool, one task, one platform; first-pass success without repair; no governed negative matrix, live model, authentic identity, durable approvals, effect commit, exactly-once, or isolation evidence; not adoption evidence
AKROVELA embedded PRAXO-CORE AKROVELA embeds published PRAXO-CORE v2.6.0 through a product-owned adapter: exact-operation tools, mandatory policy, bounded budgets, redacted telemetry, and a PostgreSQL Run/reconciliation journal; a deterministic slice with a test model and local PostgreSQL is recorded as locally verified in the AKROVELA repository (b5e5da9, 2026-11-08) Local integration (product repository) No production model provider, Agent HTTP endpoint, observations, independent verifier, ROS 2 execution, hardware effect, live KINETOVELA integration, or E1-E8 pilot run; product-owned evidence, not Core acceptance
Integration recovery Real local Symbivela, Ontovela, Rheovela, Orchadyn, Praxovela, LIMENORA, and integration-domain services; 8 steps, 5 products, 4 gates, 13 evidence artifacts, 11/11 report checks; preserve/verify denials and silent-rerun rejection Local integration Disposable local databases and fixed IDs; no production, network partition, concurrent load, independent reproduction, or deployment verification
Product effect ledgers RHEOVELA effect/durable/crash tests and PRAXOVELA tenant-scoped effect-ledger tests, including in-memory/SQLite lifecycle and reconciliation Contract/unit (product repositories) In-process SQLite/temp files; not a live PostgreSQL cluster or a cross-service transaction

CI links above point to recorded runs. They are evidence for the revision named by the run, not proof that the current main passes. Registry links name the current published version; the consumer-smoke rows above name the version that was actually installed.

Not Claimed

  • Production readiness, deployed SLOs, or capacity under load.
  • Independent security review, penetration testing, or compliance certification.
  • Generic exactly-once execution or a cross-service transaction.
  • Live verification of external workspace providers.
  • Live Postgres verification: a local attempt on 2026-11-09 with PostgreSQL 18.1 on Windows did not complete (server exited with 0xC0000142); the gated Postgres tests require PRAXO_POSTGRES_DSN and a healthy server.
  • Windows symlink containment when the test host cannot create symlinks.
  • Test results for a different commit, tag, binary, or dependency resolution.
  • Package-registry availability inferred from a source tag.

Reporting A Gap

If a claim cannot be reproduced, open an issue with the commit, platform, commands, inputs, expected result, observed result, and whether the mismatch is a test defect or a behavior defect. Do not replace a failed record with a later successful retry; keep both.