Qualify 1.2.1 test coverage and schema failure handling

.github/workflows/linux.yml

@@ -4,6 +4,12 @@ on:
     branches: [main]
   pull_request:
   workflow_dispatch:
+    inputs:
+      scope:
+        description: Full qualification or coverage-only exporter refresh
+        type: choice
+        options: [full, coverage]
+        default: full
 permissions:
   contents: read
 concurrency:
@@ -17,7 +23,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        lane: [linux, linux-safety]
+        lane: ${{ fromJSON(inputs.scope == 'coverage' && '["linux-coverage"]' || '["linux", "linux-safety", "linux-coverage"]') }}
         architecture: [x86_64, arm64]
         include:
           - architecture: x86_64

CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## Unreleased
 
+- Preserve ordinary invalidity when best-effort schema diagnostics omit an issue
+  that exceeds the byte ceiling. Required work and annotation failures remain
+  terminal. Add 31 behavior and precise failure tests across all five modules,
+  plus reproducible five-module Linux coverage exports on ARM64 and x86_64.
+
+- Extend Codable, exact value, streaming, Patch, schema/regex and public workflow
+  assertions to 318 Swift Testing functions. Preserve precise errors, complete
+  five-module coverage and finite cumulative accounting; add seeded replays and
+  a bounded native mutation pilot. Report a wholly missing native schema bundle
+  as structured metaSchema failure while retaining all resource integrity pins.
+
 - Add local full-language wasm32 WASI support, preserving all four products and
   the empty root dependency graph. Twelve cumulative UInt64 quotas can exceed
   Int.max without changing profiles; allocation capacities retain Int guards.

Documentation/Architecture/API.md

@@ -101,6 +101,13 @@ metered Base64; Date defaults to its Codable representation. Explicit ISO-8601 u
 UTC fractional seconds, finite years 0000...9999 and at most 128 input UTF-8 bytes.
 `JSONCodingOptions.userInfo` stores checked Sendable values.
 
+The qualified 128-bit probe distinguishes exact value projection from typed
+Codable. `JSONNumber.integer(as: Int128.self)`/`UInt128.self` support exact bounded
+projection. Native typed Int128/UInt128 Codable currently uses the SDK container
+defaults and throws unsupported-type EncodingError/DecodingError, including for
+small values. Use exact JSONNumber/JSON values or JSONCodablePayload for lossless
+transport; this is no claim that typed 128-bit Codable is implemented.
+
 Foreign Codable bridges inherit the foreign coder's source fidelity. External
 numeric decoding tries Int64, UInt64 and Decimal, with no Double fallback. External
 encoding requires an exact Int64, UInt64 or Decimal representation. Neither bridge

Documentation/CI/README.md

@@ -44,6 +44,25 @@ streaming regression is documented in ../Performance/README.md.
 Run locally with `python3 Scripts/ci.py --lane native`, `--lane safety` or `--lane docs`.
 The first hosted result must be inspected before marking CI qualified.
 
+## Coverage and test discovery
+
+`python3 Scripts/ci.py --lane coverage` runs the full instrumented native Swift
+Testing suite and exports all five test executables against its merged profile.
+SwiftPM's default export can contain only a subset of executable modules on the
+supported Xcode backend. The lane rejects missing/duplicate source mappings,
+stale unmapped classifications, impossible counters and changed discovery. An
+actual LLVM export omitting the schema executable must be rejected. Branch/MC/DC
+counters are reported as unavailable when the compiler emits none; generated
+Unicode data and executable lookup regions remain separate.
+
+[TestDiscovery.json](../Qualification/TestDiscovery.json) is the exact root,
+parameterized, QA and shared-state discovery contract. Native, safety and WASI
+read that same contract. Native and coverage also run ten Python harness behavior
+tests; these check qualification tooling and do not count as Swift library tests.
+The existing required Native check names/settings and all 49 cap probes remain.
+The coverage lane is available locally; this increment does not activate a new
+hosted check or claim a Linux/WASI coverage run.
+
 ## Linux development lanes
 
 The additive Linux workflow selects the pinned official Swift 6.4.0 Ubuntu 24.04
@@ -63,7 +82,7 @@ distribution, architecture or libc claims.
 
 `python3 Scripts/ci.py --lane wasm` requires the matching Swift.org 6.4.0 compiler,
 SDK and WasmKit plus Node. It records cross-build and runtime outcomes separately.
-It preserves all 192/173 root discoveries, independent corpus policies and 49 cap
+It preserves the exact root discoveries in TestDiscovery.json, independent corpus policies and 49 cap
 outcomes; both hosts also execute the frozen public consumer. The
 [WASI contract](../Architecture/WebAssembly.md) defines host/safety scopes.
 This local lane neither activates hosted CI nor substitutes for Apple/Linux gates.

Documentation/Performance/README.md

@@ -29,3 +29,9 @@ Logical retained bytes differ from process RSS. Darwin ru_maxrss is cumulative;
 allocator end observations are not isolated workload allocations. These results
 make no cross-machine throughput or OS-enforced memory guarantee. Projection and
 JSON text-sequence framing remain deferred.
+
+The proposed 1.2.1 diagnostic repair has a separate
+[matched comparison and disposition](../Qualification/Release1.2.1/ReviewReadiness.md).
+The earlier +11.9833% median remains unresolved; a subsequently authorized
+loaded-host run measured −1.6871%, with competing activity recorded. It establishes
+no intrinsic improvement and does not replace qualifying quiet-host evidence.

Documentation/Qualification/README.md

@@ -89,3 +89,28 @@ The WASI increment adds three boundary/accounting functions: 192 root functions
 invocations and all 49 public metric outcomes. Native 64-bit overflow/rendezvous
 cases remain; wasm32 uses UInt64 ledgers and cooperative checkpoint cancellation.
 Run the [WASI lane](../Architecture/WebAssembly.md) for its explicit per-host scope.
+
+The first [1.2.1 test wave](Release1.2.1/Wave1.md) adds 23 public Codable behavior
+functions: exactly 215 root functions (JSON 127, schema 51, regex 12, streaming 14,
+Patch 11), retaining all 173 parameterized invocations and all 49 cap outcomes.
+The focused QA lane now expects 34/3/3/2 functions in JSON/schema/streaming/Patch.
+Historical counts above stay tied to their original receipts. The exact current
+[discovery contract](TestDiscovery.json) is shared by native/safety/WASI harnesses.
+
+The [second 1.2.1 wave](Release1.2.1/Wave2.md) adds 29 behavior functions:
+244 root functions (JSON 144, schema 51, regex 12, streaming 26, Patch 11), with
+173 parameterized invocations retained. QA expects 51/3/15/2 functions. Safety's
+targeted TSan subset adds the 12 controlled streaming functions to the existing
+six cases, with exact 4/1/1/12 module counts. Historical receipts remain unchanged.
+
+The 1.2.1 [portability follow-up](Release1.2.1/Portability.md) has 245/173 roots,
+72 QA and 19 shared-state functions. ARM64/x86_64 Linux full lanes and scoped
+WASI debug/release QA passed. Atomic native profiling reports 97.46% authored
+lines / 93.04% regions and rejects saturated LLVM counters; earlier percentages
+remain historical measurements, not directly comparable corrected coverage.
+
+The [public workflow/mutation acceptance wave](Release1.2.1/AcceptanceWave.md)
+adds a sixth test-only target while preserving five implementation modules and four
+products. Native, both Linux architectures, scoped WASI and clean source extraction
+are qualified with exact discoveries and materialized evidence. Mutation survivors
+led to real final-chunk cancellation and failed-branch annotation assertions.