Skip to main content

Examples and Evidence Index

This index explains what each TeaQL example can actually prove. A code block, model description, generated workspace, executed test, and retained CI artifact are different evidence classes and must not be presented as interchangeable.

Inventory last reviewed: 2026-10-28. Each execution row keeps its own date.

For learning, use the seven Order tutorials and Order API usage guide. For documentation work, use the coverage matrix; it distinguishes missing writing from missing behavioral evidence. The inventory below supplies the existing sources and qualification boundaries rather than prescribing a new test campaign.

Evidence Classes​

ClassMeaning
Reproducible fixtureInputs and repeatable commands are committed; execution still needs a recorded run.
ExecutedA named version combination was run successfully on a recorded date.
Retained CI evidenceLogs, manifests, outputs, and a run/artifact identifier are preserved.
Contract-checked illustrationStatic code shape passes documentation rules but was not compiled against a generated model.
Historical evaluationEvidence for an older version; useful for history, not current compatibility.

“Runnable” should identify the exact command, working directory, dependency versions, provider, expected result, and retained evidence. Without those fields, use a narrower status.

Dated Example Inventory​

ExampleLocationCurrent statusWhat it provesWhat it does not prove
Order workflowOrder model and scenario matrixNative SQLite verified 2026-10-28: core 119/119 with 742 checkpoints; isolation, stock, lifecycle and sustained-load scopes each 7/7Seventeen core cases plus tenant isolation, stock reservation and native lifecycle faults. The separate finite soak runs four retained workers per runtime for at least 120 seconds: 13,951 operations, 77 snapshots and 49 lock probes, with transaction, connection and request-task release checks. All scopes have source-bound receipts.HTTP, browser/TFP, other providers, peak capacity and absence of all heap leaks. Java, Go, TypeScript, Rust, Python and .NET retain lifecycle/soak workspace pending-release markers; Swift uses published 0.3.1. Earlier Java published scopes, School and generator qualifications remain retained.
Shared School model and seven applicationsExecutable School baselineLocally executed on 2026-10-27: 7/7 SQLite PASS; six published runtimes, Go WORKSPACE PASS — RELEASE PENDINGOne model, explicit schema, seed idempotence, audited create, bounded query, separate-process readback and unchanged generated libraries; Go source commit and actual module resolution recorded.A passing public Go consumer, other providers, browser, TFP, concurrency or migrations.
School advanced scenarios in seven languagesShared advanced suiteLocally executed on 2026-10-27: 7/7 builds, 238/238 assertions, 7/7 full suites; two published runtimes; Go, TypeScript, Rust, Python and .NET WORKSPACE PASS — RELEASE PENDINGSame five schools and 34 scenarios per runtime: multi-entity rollback, synchronized same/different-School writes, continuous ID cursor plans, retained-ID jumps/empty/overflow/isolation, paging boundaries, empty results, nested/negative relation predicates, filtered children and aggregates, blank intent rejection, Facets, Top-N, audited updates, stale writes and deletion visibility across analytics and relations. Eight baseline checkpoints, four isolated mutation database comparisons and per-read SQLite comparisons per runtime. Go uses relation-isolation and ID-set observation patches; Rust, Python and .NET use recorded pagination patches; TypeScript's actual packed artifact and Node resolution are recorded; the earlier 76/77 published-runtime failure remains available.Passing published Go/TypeScript/Rust/Python/.NET pagination consumers, other providers, sustained concurrency stress or deployed Assist fixes.
Earlier Golden Path modelmodel.xml and fixture POMRetained historical input; replaced by School in the current workflowReproducible input for earlier Java/Rust probes.Current seven-runtime acceptance.
Golden Path workflowgolden-path.ymlSeven-runtime workflow defined; no remote run recorded for this changeBasic and advanced School suites for all seven runtimes, native build, SQLite assertions, restart checks and failure artifact retention.Retained CI proof until a remote run/artifact ID is added.
Rust Quick StartQuick Start — RustExecuted locally on 2026-07-13cargo-teaql 2.0.8, both dynamic targets, TeaQL/SQLite 4.1.1, tests, startup, query, audited mutation, and regeneration.MSRV, non-SQLite providers, load/concurrency limits, or release support lifetime.
Java Golden PathVerification Evidence ArchiveJuly 2026 evaluation/generation/regeneration executed; build blocked in that campaignMaven plugin 1.1.1 target discovery and Java generation behavior, including defects observed in that run.Current Java status or a runnable Java application, query, or mutation.
Model Explorer source catalogsrc/data/models.js and ModelExplorerContract-checked illustration; src/pages/models/** is currently excluded from the site buildFive domain guides, each with six Java/Rust scenario pairs, follow documented intent/audit terminal shapes.A currently published route, source KSML, successful generation, compilation, provider execution, or exact emitted methods for those models.
CookbookCookbook evidence matrixMixed per recipeEach recipe declares its source/runtime review or executed boundary and missing tests.Uniform execution status across every recipe or stack.
Evaluation Report #001PDF reportHistorical evaluation dated 2026-06-09A school-management Rust exercise reported for TeaQL 3.2.2, SQLite/SQLx 0.8, and cargo-teaql 0.2.0.Current TeaQL 4.1.1, current provider architecture, or current client 2.0.8 compatibility.

The PDF is retained as historical engineering evidence. Its commands, crate names, generated shapes, and conclusions must not be copied into current setup guidance without regeneration and verification.

Model Explorer Source Boundary​

The Model Explorer source currently contains five catalog entries:

  • Ecommerce Order Service;
  • Clinic Appointment Service;
  • Hotel Management Service;
  • Helpdesk Ticketing Service;
  • Supply Chain Service.

Each entry contains six scenarios—simple query, graph loading, advanced filter, aggregation, DDD extension, and runtime customization—with both Java and Rust variants. That is 30 scenario pairs and 60 language-specific snippets.

The documentation contract checker validates query intent ordering and mutation audit terminals in those strings. The model data file does not contain the KSML sources named by its sourceDir labels, and those scenarios have not been generated or compiled as current workspaces. Treat every generated facade, entity, relation, filter, and update method as illustrative until the matching model source and generated evidence are committed.

The Docusaurus pages configuration currently excludes models/**, so these components do not constitute a published model catalog. Removing that exclusion is a separate publication decision and should happen only after the evidence and navigation requirements below are satisfied.

Reproduce the Maintained Path​

The Rust Quick Start and Java failure records in this inventory describe the July 2026 campaign. For a new run, start from the current language journey, inspect the live target catalog, and use the commands in the Golden Path workflow. Do not repair generated output by hand. Use Cross-language Conformance Status for current committed capability evidence and the archived Java Compatibility Evidence only when investigating those historical failure signatures.

Evidence Still Missing​

The example collection is not ready for a first-class top-level Examples section until it includes:

  1. a retained current Java generated build, query, audited mutation, and logs linked from this site;
  2. a retained run/artifact identifier for the committed Golden Path workflow;
  3. at least one maintained generated-output sample with its source model and generator command;
  4. compiled/generated evidence for Model Explorer entries or an explicit decision to keep them as design-only guides;
  5. current non-SQLite provider examples with provider/database/toolchain versions and expected output;
  6. small AI task examples with allowed edit surfaces and generated-output review checklists.

Adding or Promoting an Example​

An example entry must record:

  • source model or fixture path;
  • client and generator versions;
  • dynamic target and exact command;
  • generated dependency versions;
  • build/test/run command and working directory;
  • provider/database/toolchain versions;
  • expected output;
  • generated versus handwritten file boundaries;
  • evidence status, date, owner, update trigger, and retained artifact link.

Promote an illustration to executed only after its exact model and methods are generated and run. Promote an executed example to retained CI evidence only after logs/artifacts have a durable run identifier.

Seven-runtime Order workflow​

The Order workflow retains seventeen executable core native scenarios per language: 119/119, 7/7 SQLite databases, with 742 independent database checkpoints in the P2 retained-ID and summary receipt. Those core cases verify actual retained-ID store events, random jumps, empty sets, overflow lower bounds versus exact totals, predicate isolation, real TTL expiry and explicit invalidation; provider summaries, status facets, currency isolation and bounded item previews are checked against the same filtered SQLite rows. Every read preserves all persisted fields and timestamps.

That suite reran all fifteen previous cases. Concurrent writes still repeat three times: 49 rounds and 98 workers. Subchecks and rounds do not inflate case counts. The 56/56 receipt, 77/77 receipt, 91/91 receipt and 105/105 receipt and their sources remain unchanged, alongside the separate School 238/238 baseline.

The retained core stage used only local evidence: the unchanged model and its retained evaluation, existing generated libraries and locally rendered model-aware Assist. It makes no fresh online evaluation claim. Go, TypeScript, Rust, Python and .NET remain pending runtime release; Java and Swift pass on published artifacts. Its .NET compound condition-key repair retains its own failed attempt, patch and tests; earlier runtime failures and generator deployment qualifications remain visible in the receipt chain. TypeScript's store and Swift's explicit-invalidation adapter are application-owned.

Buyer and tenant isolation now has a separate 7/7 receipt on the locally evaluated model revision 1.1.0, with 46 probes per language and 350 new SQLite checkpoints. The application facade constrains rows, counts, sums, items, retained IDs and audited writes by trusted tenant and buyer. Forged identity, forbidden fields, invalid bounds and foreign IDs fail their respective gates. Synthetic server sessions identify the authentication integration seam; arbitrary raw Q access is outside the facade's authorization boundary. Generator mutation intent/session, committed-version and TypeScript Assist repairs remain pending deployment.

All P1 and the first five P2 scenarios now have executable evidence across their recorded model revisions. Stock extensions, fault injection, HTTP, browser/TFP and ecosystem integrations remain planned. Continuous export during concurrent writes is outside these fixed-dataset receipts.

Order inventory reservation​

The Order stock workflow uses a separately evaluated model revision 1.2.0. Its seven-language SQLite receipt records 7/7 stock cells, 47 reservation requests and 56 database checkpoints per runtime (392 total). It includes 42 synchronized two-process oversell and duplicate races, a late receipt INSERT failure after Order/inventory UPDATEs, full rollback and successful same-key retry.

The model, fixture, catalog, read-only oracle, local discovery manifest and execution notes bind the result. All seven native sources are linked from the workflow. This stage used only local evidence. Go, TypeScript, Rust, Python and .NET retain pending-release markers, and generator fixes remain pending deployment. School, original Order 119/119 and isolation 7/7 receipts remain unchanged. The tenant version gate coordinates this facade's writers; the fixture covers one complete single-SKU line, not distributed inventory, multi-SKU fulfillment or sustained throughput.

Order request lifecycle and provider faults​

The Order lifecycle workflow reuses the exact locally evaluated Order 1.2.0 model and generated libraries. Its seven-language receipt records 7/7 lifecycle cells, 315 independent SQLite checkpoints and 147 lock probes. Each runtime runs 11 independent processes covering native cancellation before admission and after writes, a late provider statement error, a real COMMIT BUSY failure under a separately held reader lock, and the successful control. Full snapshots and same-provider recovery verify rollback and observed resource release.

Java lifecycle requires a workspace JDBC fix pending release. Its published 1.555-RELEASE failure retains exact inputs and dirty pooled-transaction evidence. The runtime patch, three regression tests, discovery manifest and execution notes qualify the pass. All five prior workspace runtime markers remain, Swift passes on its recorded published runtime, and earlier Java published scopes remain unchanged. School and every earlier Order receipt stay source-bound and immutable. This stage uses only local evidence. Cooperative cancellation, SQLite rollback-journal commit faults and bounded repetitions do not certify arbitrary SQL interruption, HTTP disconnect handling, sustained stress or ambiguous post-commit outcomes.

Order sustained concurrency and resource release​

The Order sustained-load workflow and source-bound receipt record 7/7 finite SQLite soak checks: four retained native processes per runtime, at least 120 seconds of mixed load, 13,951 operations, 77 independent snapshots and 49 writer probes. Held-transaction calibration precedes warmed native resource gauges, drain, cooldown, same-provider recovery and explicit shutdown. Final SQLite files, per-request ledgers and compressed byte-exact command logs are retained with source hashes. Smoke runs cannot receive formal PASS status.

Use the catalog, resource observation notes and local discovery manifest. The unchanged evaluated Order model and cached generated libraries were reused; no model or evidence was transmitted for online evaluation, Assist or generation. Java, Go, TypeScript, Rust, Python and .NET retain their lifecycle workspace RELEASE PENDING qualifications; Swift retains published 0.3.1. School and all previous Order receipts remain unchanged. This bounded workload does not certify maximum throughput, shared-context thread safety or absence of all heap leaks.