V2.0 — Shipped (tag v2.0.0, 2026-08-11)

:::info The only bucket with a checkable 100%-done claim

Everything in this file is delivered in the tagged v2.0.0 release — the project's first-ever GitHub Release, published 2026-08-11. The full execution ledger (492/492 items closed) is .antigravitycli/v20_todo.md; this file is the published, requirement-shaped summary of that ledger, not a duplicate of it — for execution history (dates, commits, individual task closures) follow the links below into v20_todo.md rather than expecting that detail here.

Requirement ids below are promoted, unedited in substance, from .antigravitycli/requirements_v3_next.md's internal REQ-V2-01..REQ-V2-10 numbering (that file's own banner explains the _next filename refers to the next unshipped release, v3.0 — its content, despite the name, is the historical v2.0 record). Each entry keeps its GAP-NN alias permanently, since v20_todo.md and the TASK-<phase>.<group>.<seq> mappings cite by GAP-NN, not by the REQ-V2-* id.

:::

Cluster A — Infrastructure

REQ-V2-01

Global Config Field Activation

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-09
• Traces to: REQ-BUS-12
• Evidence: .antigravitycli/v20_todo.md closeout ledger; not independently re-verified against engine/ude/config.py in this pass — flagged for a follow-up spot-check.
• History: v1.0 had only error_policy active; v2.0 activates all documented GlobalConfig fields.

All fields in ude_global_config.json (doxygen_path, log_level, log_file, cache_root_dir, global_templates_dir, error_policy) are read, validated via a Pydantic GlobalConfig model, and operationally applied at engine startup. Unknown keys are silently ignored (ConfigDict(extra="ignore")); missing fields default without raising ValidationError. translation_service stays reserved — correctly: it is a v4.0+ dependency (see REQ-BUS-06 in v4.0-backlog.md), not an oversight.

REQ-V2-02

Unified Logging Infrastructure

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-12
• Traces to: REQ-BUS-09
• Evidence: .antigravitycli/v20_todo.md closeout ledger; not independently re-verified against engine/ude/ in this pass.
• History: supersedes REQ-FUN-25's v1.0 baseline logging in scope (v1.0 entry preserved for traceability in v1.0-mvp.md).

A single scoped "ude" root logger is configured via logging_setup(cfg: GlobalConfig) at startup, supporting StreamHandler (always) and optional FileHandler (when log_file is set). The incorrect logger label "ude.renderers" in interfaces.py is corrected to "ude.interfaces". Calling logging_setup() twice does not accumulate handlers.

REQ-V2-03

L2 Render Cache Activation

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-07
• Traces to: REQ-BUS-03, REQ-BUS-09
• Evidence: engine/ude/storage.py (BuildCacheManager, existed since v1.0); wiring into all 16 renderer render() calls per .antigravitycli/v20_todo.md closeout — not independently re-verified against every renderer class in this pass.
• History: activates the L2 cache described in v1.0-mvp.md's baseline form.

BuildCacheManager's L2 cache is wired into all 16 concrete renderer render() calls, keyed on the IR hash plus the Jinja2 template hash. The orchestrator passes cache_dir (from GlobalConfig.cache_root_dir) via a cache_manager kwarg forwarded through the __new__ factory in all three renderer families (static_html, hugo_markdown, the ODA family). When cache_root_dir is absent, behavior is identical to v1.0 — no regression.

REQ-V2-04

Doxyfile 3-Tier Key-Level Merge

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-11
• Traces to: REQ-BUS-13
• Evidence: engine/ude/collectors/doxyfile.py; key wiring in engine/ude/collectors/doxygen.py (verified: :642-657 implements the related exclude/exclude_patterns/file_patterns runtime keys, part of the same T3 runtime tier this item's key-level merge governs — see REQ-V3-09 for the filtering-specific requirement this evidence primarily supports).
• History: replaces v1.0-mvp.md's source-concatenation approach.

Doxyfile assembly uses an explicit 3-tier key-level merge (T1 global template → T2 target-specific template → T3 runtime parameters), with T3 always winning and key conflicts logged at DEBUG. Existing ude_doc_config.json files using doxyfile_template continue to work identically.

Cluster B — Library API & CLI Unification

REQ-V2-05

UdeOrchestrator Public Library API

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-05
• Traces to: REQ-BUS-09
• Evidence: engine/ude/orchestrator.py:77 (find_product_json), :99 (deep_merge) — verified: both utilities now live in orchestrator.py, consistent with this requirement's consolidation goal (not independently confirmed that cli.py no longer duplicates them).

UdeOrchestrator exposes stable public methods parse(config, config_dir), render(catalog, config, config_dir, out_dir), and run(doc_config_path). The previously duplicated deep_merge()/find_product_json() utilities are consolidated into orchestrator.py; cli.py is reduced to argument parsing plus a thin orchestrator delegation.

REQ-V2-06

CLI Subcommands

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-01
• Traces to: REQ-BUS-09
• Evidence: engine/ude/cli.py.
• History: landed 2026-07-19, ahead of the rest of the v2.0 cluster (per v1.0-mvp.md, the v1.0-era flat invocation ude --global-config <path> --sdk-config <path> --doc-config <path> remains supported as a backward-compatible alias).

The CLI supports four subcommands (compile, parse, render, audit) via argparse subparsers. The v1.0 flat flag interface remains fully operational. ude parse --output-ir followed by ude render --input-ir produces byte-identical output to ude compile.

Cluster C — Typed Intermediate Representation

REQ-V2-07

7-Model Typed Entity Schema

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-03
• Traces to: REQ-BUS-01
• Evidence: engine/ude/models.py.
• History: replaces v1.0-mvp.md's ClassEntity discriminated union. Prerequisite for REQ-V2-08.

The v1.0 ClassEntity discriminated union is replaced with seven explicit Pydantic models: ClassModel, MethodModel, ParameterModel, EnumModel, VariableModel, ConstantModel, TypeAliasModel — all carrying model_config = ConfigDict(extra="ignore") for forward compatibility. Backward-compatible aliases (ClassEntity = ClassModel, etc.) ship in the same commit. fields: List[str] becomes fields: List[VariableModel]. Old v1.0 .json.gz IR files deserialize without ValidationError.

Cluster D — QA & Testing Completeness

REQ-V2-08

Documentation Coverage Gate

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-10
• Traces to: REQ-BUS-08
• Evidence: engine/ude/coverage.py; .antigravitycli/v20_todo.md FIN-9.13 — real command run, ude.cli audit --mode reject-undocumented --threshold 98, exit 0, 100.00% coverage, 351 entities.

Two enforcement modes are configurable via GlobalConfig: reject-undocumented (non-zero exit if coverage is below coverage_threshold) and allow-undocumented (warnings only, exit 0). ude audit prints a per-entity-type Markdown coverage table. No LLM calls occur in v2.0 — the enrichment modes (auto-document, verify-document) are v4.0+ scope, see v4.0-backlog.md. The coverage gate runs during ude compile and ude audit, not during ude parse/ude render.

:::note Three v1.0-era requirements folded into this delivery, with a narrower scope than originally specified Found during this restructuring — none of the three had a home in any prior draft of this catalog:

• REQ-FUN-12 (Standalone Coverage Reporting Command) is substantially satisfied: ude audit is exactly the independent, standalone coverage subcommand this requirement asked for (under a different name than the ude coverage the original text proposed).
• REQ-FUN-15 (Quality Gate Scope & Completeness Criteria) is only partially satisfied. Verified against engine/ude/coverage.py:180-193 (_record/compute_coverage): an entity counts as "documented" if docstring is not None — a shallow presence check. The full criterion this requirement specifies — every parameter and the return value must independently carry a non-empty description — is not implemented; nor are enums, constants, or type aliases counted yet (noted in the function's own docstring). This gap is real, outstanding V3.0 work — see REQ-V3-13 in v3.0-next.md.
• REQ-FUN-16 (Quality Gate Automatic Exclusions) is not implemented as specified. Verified against engine/ude/coverage.py:56 (_is_excluded_from_coverage): exclusion is purely name-pattern based (private _foo, dunder __foo__, logger, and a small hardcoded false-positive list — see user-docs/docs/exclusion-gates.md's coverage-gate section). There is no semantic exclusion of trivial property getters/setters or of methods inherited unchanged from a base class/stdlib (Equals/GetHashCode/ToString, __str__/__repr__) — a grep for those names in coverage.py found nothing. This is real, outstanding V3.0 work — folded into REQ-V3-13. :::

REQ-V2-09

External Integration Script Confirmation

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-31; covers TEST-INT-01 (run_regression_tests.py), TEST-INT-03 (verify_pages.py), TEST-INT-04 and TEST-INT-06 (check_links.py) — those four ids keep their full specifications in srs/integration_tests_specification.md (retained, not retired)
• Traces to: REQ-BUS-08
• Evidence: tests/integration/regression/{run_regression_tests.py,verify_pages.py,check_links.py}; full resolution history and path-migration corrections in .antigravitycli/requirements_v3_next.md's REQ-V2-09 section.

The three integration scripts originally referenced under a Tests/ path (run_regression_tests.py, verify_pages.py, check_links.py) are located at their current canonical path, tests/integration/regression/. The run_all_integration_tests.{sh,bat} aggregator script itself is orphaned (positional-arg calling convention no longer matches the scripts' real CLIs, and it is referenced by zero CI workflows) — resolved not by resurrecting it but via its actual working successors: per-repo check_links.py inside each deploy.yml, and verify_pages.py --remote-url inside integration_tests.yml's scheduled verify-live-pages job, both confirmed passing against live sites.

REQ-V2-10

Per-Language Integration Test Suites

• Status: PARTIAL — suites are delivered; four per-language rendering residuals are reassigned to V3.0, see below
• Target: v2.0
• Alias: GAP-32
• Traces to: REQ-BUS-08
• Evidence: engine/tests/ per-language integration suites (file-level; individual test files not enumerated in this pass).

Full parse → render integration test suites are implemented for all four supported languages, covering entity types not exercised by the golden-master regression suite, via a shared LanguageIntegrationBase mixin. What this delivered is test-suite existence and coverage of already-shipped behavior — it did not deliver new rendering features. Four residual per-language rendering gaps identified during this work are not closed by this requirement and are reassigned as V3.0 scope, folded into REQ-V3-01/REQ-V3-04 in v3.0-next.md because they are the same underlying work as GAP-28 (structural render parity): C++ category landing pages, C# interface/delegate/event rendering, Java extends/implements relationship rendering, and (kept separate, since Python already has partial support) Python fget/fset property and dunder-method edge cases.

:::note Classification note This entry (and REQ-V2-09) are catalogued as functional deliverables, not non-functional: "a test suite exists and covers X" is a functional deliverable. Only coverage thresholds and pipeline resilience properties belong in nfr.md's Operability category. :::

Pulled forward from the original v3.0 scope

REQ-FUN-50 (v2 form)

sidebar.toml as Mandatory Single Navigation Source

• Status: IMPLEMENTED
• Target: v2.0
• Alias: GAP-06
• Traces to: REQ-BUS-10
• Evidence: engine/ude/renderers/static_html.py:526-535 — verified: HtmlRenderer raises RendererError when sidebar_config has no [[sidebar]] entries, with an inline comment ([IMP-32.10]) explicitly stating "there is deliberately no default navigation fallback here."
• History: see v1.0-mvp.md for the v1.0 per-renderer-class JSON mechanism this replaces.

The per-renderer-class toc_<Class>.json files are removed entirely; navigation is sourced from a single sidebar.toml per project (not per renderer class). The file is mandatory — missing, malformed, or empty (sidebar = [], or zero [[sidebar]] entries) all raise UdeException/RendererError; there is no default navigation at any tier. Resolution follows a 3-way deep-merge cascade (global → SDK → document). The [groups] folder-taxonomy table is the one exception that keeps an engine-tier default, described under v1.0-mvp.md.

REQ-V2-11

Static / Inline / Redirect Page Node Rendering

• Status: IMPLEMENTED — flat nodes only, see the boundary note below
• Target: v2.0
• Alias: GAP-26
• Traces to: REQ-BUS-10, REQ-BUS-11
• Evidence: engine/ude/renderers/static_html.py:559-588 (verified: elif item_type in ("static", "inline") at :559, elif item_type == "redirect" at :571, each hard-coding "children": [] on its nav-tree node — :569, :584, :587); engine/ude/renderers/hugo_markdown.py:515-554 (verified: elif item_type in ("static", "inline", "redirect") at :515, same per-type handling); test coverage engine/tests/test_static_pages.py (not independently re-read this pass).

sidebar.toml root-level static, inline, and redirect node types render correctly in both renderer families. static nodes extract <body> content from an HTML source file or all content (minus front-matter) from a Markdown source file; inline nodes embed content directly in sidebar.toml; redirect nodes emit a meta-refresh/JS redirect (Hugo) or an external_link nav entry (standalone HTML) depending on whether content is present. Both engines support Jinja2 variable interpolation in static/inline content.

:::caution Boundary — this is exactly where GAP-30 begins Every node rendered by this requirement is flat. static_html.py's sidebar loop (:539-588) iterates sidebar_items at the top level only and hard-codes "children": [] on every custom-page node it creates. hugo_markdown.py's only other children-shaped structure (:588-602, _build_child_namespace_map) is the unrelated namespace-nesting map, not sidebar-item children. Neither renderer recurses into a static/inline/redirect node's own children array — confirmed by grepping both files for item.get("children")/item["children"]: no matches. Nested composition of these node types is REQ-V3-02 (GAP-30), not this requirement. Do not cite this requirement's delivery as evidence that GAP-30 is done — see v3.0-next.md. :::

Content-alignment achievement (do not conflate with GAP-28)

REQ-FUN-49 (v2 scope)

Docomatic Semantic Alignment at Scale

• Status: IMPLEMENTED
• Target: v2.0
• Traces to: REQ-BUS-08, REQ-BUS-10
• Evidence: ude_docomatic_coverage.md, ude_docomatic_alignment_rules.md (rules 1–22); engine/tests/test_docomatic_alignment.py:490-494,694-701 — verified: MAX_ALLOWED_DIFFERENCES = {"cpp": 906, "csharp": 241, "java": 392}, and the ODA-renderer-family parametrization gates oda_html for cpp/cs/java plus oda_hugo_markdown for cpp.
• History: extends v1.0-mvp.md's baseline harness to full production scale.

:::info Post-freeze correction (2026-08-21) — evidence path moved, achievement unchanged The cited engine/tests/test_docomatic_alignment.py no longer exists at that path: DEL-A04 (Clean Room extraction) relocated oda.py and its alignment test/baseline to flude-oda-plugin (now flude-oda-plugin/tests/test_docomatic_alignment.py), a separate repository. This requirement's 39-combination/16-SDK achievement itself is unaffected and remains real — only its evidence now lives in a different, ODA-plugin-specific repository, consistent with flude-oda-plugin being the sole legitimate place for comparison against real ODA/Doc-O-Matic output going forward (see REQ-V3-44). engine itself retains an orphaned, now-unused tests/assets/golden_master/html_oda/ fixture directory from before this move — dead weight, not a functional dependency (confirmed: PIPELINE_COMPLEXES in engine/tests/test_golden_master.py has no oda mode referencing it). :::

Docomatic semantic alignment achieved across 39 combinations / 16 SDKs: every residual missing/extra difference is either a shipped fix, a documented rule (ude_docomatic_alignment_rules.md, rules 1–22), or individually verified as real content Docomatic does not itself document.

:::danger Anti-pattern to avoid citing .antigravitycli/planned.md:198 cites this requirement's "≥98% Docomatic match" as evidence that GAP-28 (structural render parity) is done. These are different properties. REQ-FUN-49 measures text/content alignment via an allowances engine that tolerates known structural differences; GAP-28 measures whether a renderer emits the same set of pages at all. Verified this pass: static_html.py's HtmlRenderer._post_render (:232, no-op body at :240) and oda.py's ODAHtmlRenderer._post_render (:290, no-op body at :298) are both no-ops on the base class shared by the C++/C#/Java subclasses; only the Python subclasses override it (static_html.py:928, oda.py:921). The ≥98% figure belongs here, at REQ-FUN-49, and must never be used as evidence for GAP-28 — see v3.0-next.md. :::