ThorFuchs · Oct 12, 2026 · Oct 12, 2026 · Oct 12, 2026
@@ -12,6 +12,9 @@
   qualification remains a release gate. No tag/publication follows this change.
 - Correct native ISO-8601 calendar fields and strict Base64 unused padding bits.
   See API migration guidance; no limits/profile weights or dependencies raised.
+- Release failed Base64 scratch from logical retention while preserving consumed
+  work and terminal operational errors. Add 19 acceptance/conformance tests,
+  public mobile recovery checks and a focused debug/release/ASAN QA lane.
 
 
 ## 1.0.0 — 2026-10-11

@@ -35,7 +35,7 @@ order by default. Exact duplicates fail checked construction and parsing.
 | `JSONObject.build(limits:_:)` | Exclusive noncopyable member accumulator; checks before append |
 | `JSON.buildArray(limits:_:)` | Exclusive array accumulator; nested arrays remain one element |
 | `append(contentsOf:)` on array accumulator | Explicit flattening with charged work |
-| Object subscripts by String/JSONKey | Unmetered linear exact-key lookup; nil means absence |
+| Object subscripts by String/JSONKey | Unmetered exact-key lookup using immutable metadata; nil means absence |
 | `JSON.requiredMember(_:limits:)` | Bounded lookup; throws on wrong kind or missing member |
 | `JSON.presence(of:limits:)` | Bounded lookup separating absent, null and non-null value |
 | `JSONObject.presence(of:)` | Unmetered convenience with the same three states |

@@ -53,3 +53,14 @@ nine release samples after three warmups: median54.406542 ms and process peak RS
 allocation or hard-RSS claims, and no matched mobile optimization comparison.
 Physical iPad measurements remain blocked. Logical bytes, allocator observations,
 process RSS and caller computation remain distinct.
+
+The additional QA wave changes only Base64 failure cleanup in the runtime. Affected
+public Data/payload paths have three matched, alternating release-native process
+pairs (27 samples per source/workload, 100 invocations/sample). Valid Data medians
+range from −1.28% to +4.98%; payload decoding is +1.16%, corruption recovery −0.09%.
+Input hashes/work counters match; no new 10% timing trigger is crossed. The corrected
+recovery ledger reports final live zero instead of 300 and peak 16 instead of 300.
+This is truthful accounting of freed scratch, without an RSS/allocator improvement
+claim. The unchanged index and twelve-workload dispositions above are retained.
+The expanded physical QA harness has different work and cannot be compared with
+the earlier phone median as a matched speed result. See [QA](QA.md).
@@ -1,20 +1,68 @@
 {
   "schema_version": 1,
   "target_release": "1.1.0",
-  "milestone": "implementation, runtime qualification and reviewable candidate",
-  "tested_core_sha": "a48eb337b70eb075ce37c99dc3d150032f720d00",
-  "preserved_history": ["1.0.0", "W00-W12", "G01-G12", "RC01-RC08", "F01-F06", "J01-J04", "R0 turn 001"],
-  "root_function_counts": {"JSON":86,"JSONStreaming":11,"JSONSchema":38,"JSONRegexRuntime":12,"JSONPatch":9},
-  "root_function_total":156,
-  "parameterized_invocations":173,
-  "cap_probes":49,
-  "readme_examples":10,
-  "authored_public_docc_symbols":582,
-  "payload_fixtures":8,
-  "r5_matrix": {"macOS":"executed","iPhone_simulator":"executed","iPad_simulator":"executed","physical_iPhone":"executed","physical_iPad":"blocked; unavailable"},
-  "conditional_features": {"SIMD":"deferred","parallel_batches":"deferred","E11":"deferred","E12":"deferred"},
-  "evidence_policy":"Actual final native/safety/docs/mobile receipts and materialized Turn002 archive identify commands, source hashes, counts and limitations. Documentation-only successors map unchanged core/test inputs explicitly.",
-  "publication_performed":false,
-  "remaining_release_blockers":["Physical iPad smoke, disk persistence and release measurements for the agreed R5 matrix", "Complete maintainer review and protected PR integration before final release readiness"],
-  "performance_review":"IntegratedPerformance.md; unchanged 10% trigger, retain initial regression measurements and final accepted dispositions"
+  "milestone": "additional acceptance and conformance wave",
+  "tested_core_sha": "5d655f2beab569ac804890493bfafcf6045caabe",
+  "preserved_history": [
+    "1.0.0",
+    "W00-W12",
+    "G01-G12",
+    "RC01-RC08",
+    "F01-F06",
+    "J01-J04",
+    "R0 turn 001"
+  ],
+  "root_function_counts": {
+    "JSON": 97,
+    "JSONStreaming": 14,
+    "JSONSchema": 41,
+    "JSONRegexRuntime": 12,
+    "JSONPatch": 11
+  },
+  "root_function_total": 175,
+  "parameterized_invocations": 173,
+  "cap_probes": 49,
+  "readme_examples": 10,
+  "authored_public_docc_symbols": 582,
+  "payload_fixtures": 8,
+  "r5_matrix": {
+    "macOS": "executed",
+    "iPhone_simulator": "executed",
+    "iPad_simulator": "executed",
+    "physical_iPhone": "executed",
+    "physical_iPad": "blocked; unavailable"
+  },
+  "conditional_features": {
+    "SIMD": "deferred",
+    "parallel_batches": "deferred",
+    "E11": "deferred",
+    "E12": "deferred"
+  },
+  "evidence_policy": "Turn003 materializes actual native/safety/QA/docs/SDK/simulator/iPhone receipts, raw measurements and reproducible source archive. Turn002 and original history remain immutable. Source bindings identify unchanged runtime bytes and expanded test cases.",
+  "publication_performed": false,
+  "remaining_release_blockers": [
+    "Physical iPad smoke, disk persistence and release measurements for the agreed R5 matrix",
+    "Complete maintainer review and protected PR integration before final release readiness"
+  ],
+  "performance_review": "IntegratedPerformance.md; unchanged 10% trigger, retain initial regression measurements and final accepted dispositions",
+  "prior_milestones": [
+    {
+      "turn": 2,
+      "tested_core_sha": "a48eb337b70eb075ce37c99dc3d150032f720d00",
+      "root_function_total": 156,
+      "source_manifest_sha256": "2b69a683002eaa4e112bd005b3eb06e63dc8e06bfcd2b5c8485f076870f26797",
+      "return_sha256": "151ce56037763e1ca3d6d0439521020f34ca368ef77b50c31b8ef4e27fb282b6"
+    }
+  ],
+  "qa_wave": {
+    "added_functions": 19,
+    "baseline_swift_test_files_retained": 48,
+    "scope": "QA.md",
+    "native_full": "passed on tested_core_sha",
+    "safety_full": "passed on tested_core_sha; TSAN5",
+    "mobile": "both SDKs/simulators/physical iPhone passed; iPad physical blocked",
+    "supplemental_boundary_review": "passed at e6eb46da346bee9892119fdc414abd7783e66b2c: 19 exact cases each in debug/release/ASAN",
+    "docs": "passed at e6eb46da346bee9892119fdc414abd7783e66b2c: four catalogues, 582 authored public symbols, 10 README examples; refresh final documentation successor",
+    "accounting_fix": "Failed Base64 scratch released; work/terminal health retained; caps/weights/version unchanged"
+  }
 }
@@ -0,0 +1,69 @@
+# Additional 1.1 acceptance and conformance wave
+
+The Turn003 wave adds 19 Swift Testing behavior functions to the 156-function
+Turn002 candidate. Exact discovery is JSON97, Streaming14, Schema41, Regex12 and
+Patch11: 175 functions, with all 173 prior parameterized invocations and all 48
+earlier Swift test files retained. No independent policy exclusion is reclassified.
+
+| Subject | Additional evidence |
+| --- | --- |
+| Exact transport | Fixed seed `0x534A1100A11CE001`, 192 documents across six root kinds; compact token bytes independently check native projection, Version 1 opening and native/Foundation JSON/property-list transport |
+| Numeric projections | 772 integral spellings, independently checked signed/unsigned 8-bit ranges, including equivalent fractional/exponent spellings |
+| Keys/index publication | Checked construction, accumulator, parser, edits and reusable scopes; actual member boundaries 7/8/9, 31/32/33, 63/64/65 and 127/128/129; ordered scalar-key oracle |
+| Base64 | 259 canonical byte fixtures; 1,152 mutated unused-padding-bit cases; repeated caught corruption, prefix ownership, quota failure and in-loop cancellation |
+| Scoped coding | Permanent ceilings after ordinary failure, prior unselected-text admission, caller-error scratch release and mapped Unicode UTF-8 byte coordinates |
+| Payload budgets | Six roots, creation/opening work and live-retention inclusive/minus-one matrix; identical preserved bytes |
+| Shared state | 16 workers ×20 openings/read scopes over shared immutable inputs; checked Sendable and five-case TSAN selection |
+| Schema | Explicitly rooted definitions and negative cases; wide persisted instances/evaluated scalar locations, plan reuse after invalidity, cumulative regex search boundary |
+| Patch | Ordered scalar-key edits and real 32→31→32 stages; cumulative quota refusal publishes no partial result |
+| Streaming | Every byte split of a wide scalar-key/lexeme document and malformed duplicate; stable second-key location, terminal pulls, cancellation after three accepted sink bytes |
+
+The first executed wave reproduced a logical-retention accounting defect: shared
+Base64 decoding left partial scratch charged after ordinary corruption. A caller
+catching that error could accumulate freed-byte reservations and later get a false
+resource refusal. Decode now releases only its invocation's failed reservation.
+Consumed work and terminal resource/cancellation state remain charged/latched;
+successful Data ownership retains its prior accounting. This corrects Version 1
+accounting without changing weights, caps, grammar or the public API. It is not an
+allocator-memory leak or an RSS improvement claim. Original failure receipts remain.
+
+A new projection test initially calibrated its limit using native rounded decoding
+then added a larger exactness-proof operation. Its correction measures the same
+complete operation on both sides. Compiler harness corrections use an explicit
+JSONKey initializer and an actor-owned Sendable sink; no unsafe annotation or
+compiler bypass is introduced. A separate review expanded generated field widths
+to include actual member boundaries after adding the special keys.
+
+Use `Scripts/ci.py --lane native`, `--lane safety`, `--lane docs` and the supplemental
+`--lane qa` on the supported host. Full native execution retains JSONTestSuite
+318 discovered/287 evaluated/31 policy and Schema2337/2300/37, 82 authored vectors,
+nine budget vectors, 4,396 malformed replays, 31,824 regex cases/442 patterns and
+18,853 Unicode cases/94,265 comparisons. All49 public cap outcomes remain required.
+The rooted public consumer retains20 cases and distinct operational errors.
+
+The public mobile harness additionally checks100 caught malformed Data projections
+under32 logical bytes and wide indexed payload/Schema/Patch behavior. Both SDKs,
+isolated iPhone/iPad simulators and the physical iPhone execute the updated harness;
+disk save/reopen/update uses three independently exited processes and eight frozen
+fixtures. Physical iPad qualification remains blocked. Simulator execution does
+not establish physical-device performance.
+
+The affected Base64 paths have three serial alternating release-native pairs,
+each with3 warmups,9 samples and100 public invocations/sample, finite balanced caps,
+matched input hashes and identical work. Across27 samples/side, valid decode
+medians range from−1.28% to+4.98%; native payload decode is+1.16% and corruption
+recovery−0.09%. No new10% timing trigger is crossed. Recovery's corrected logical
+peak is16 rather than300, with final live0 rather than300; work remains506. Keep
+raw variance/ranges and process-wide RSS separate from these logical counters.
+The earlier wide-index and integrated streaming dispositions remain unchanged.
+
+Physical iPhone14ProMax/iOS27.2 updated public QA measures9 release samples after3
+warmups: median50.457167ms, process-wide peak25,133,056 bytes, thermal state0 at
+both observations. This harness adds work relative to Turn002; the two medians
+are not a matched speed comparison. SDK/UIKit/process costs are included.
+
+Actual receipts, failed-source reproduction, raw samples, source bindings, clean
+archive rebuilding and compiled catalogues are materialized in the local Turn003
+return. Hosted Native job artifacts supply remote source/command receipts. This
+wave does not merge the protected PR stack, qualify an MCP server, publish1.1,
+select SIMD/parallel batches, or close the unavailable physical iPad gate.
@@ -33,3 +33,29 @@ actual commands/source and nonzero counts. Source archives exclude private input
 Git/build trees, signing profiles and device identifiers. Physical iPad remains
 blocked; conditional SIMD/parallel batches stay deferred, with no unqualified
 performance claim. No merge/tag/publication is implied by this review.
+
+## Additional Turn003 review
+
+The same implementing agent performed a separate source, fixture, receipt and
+test-body review against parent `9e61f158`. This remains distinct from independent
+human review or GitHub approval. Native/full safety ran at `5d655f2`; the reviewed
+boundary additions ran through the exact 19-case debug/release/ASAN QA lane at
+`e6eb46d`. All runtime library bytes are unchanged between those two stages.
+
+- Medium: shared Base64 failure retained freed scratch in the logical ledger.
+  Preserve the 400 failed recovery assertions; restore only the failed decoder's
+  live reservation. Successful ownership, work and terminal health remain intact.
+  Full native/ASAN, quota/cancellation-prefix tests and a public 32-byte consumer
+  verify the correction. Three matched benchmark pairs retain identical work.
+- Qualification: special keys added to generated widths obscured actual final
+  index boundaries. Expanded the corpus to actual 7/8/9, 31/32/33, 63/64/65 and
+  127/128/129 members; Patch now explicitly visits 32→31→32. All added cases pass
+  debug/release/ASAN without changing any prior test file or policy exclusion.
+- Documentation: the API table still described direct object lookup as linear.
+  It now describes immutable exact-key metadata. README distinguishes released
+  1.0's historical stream regression from the candidate's separate +15.76% pair.
+
+The final source binding, clean reproduction, actual catalogue outputs and hosted
+CI receipts accompany the local Turn003 return. Both SDKs, both simulator families
+and the updated physical iPhone public/disk harness pass. Physical iPad remains
+blocked; protected integration and publication require separate maintainer action.
@@ -12,14 +12,23 @@ and seven additional nonparameterized cases. The first production lookup increme
 the independent runner adds nine budget/report vectors. There are three separate
 JSON/JSONSchema consumer tests. All discoveries must be nonzero.
 
-The current integrated 1.1 candidate checks exactly 156 root functions:
+The Turn002 integrated 1.1 candidate checked exactly 156 root functions:
 JSON 86, JSONStreaming 11, JSONSchema 38, JSONRegexRuntime 12 and JSONPatch 9.
 All 173 parameterized invocations remain. The earlier counts above describe their
 historical stages, not this source. Three additional public/mobile Swift Testing
 functions exercise eight pinned payload fixtures and three independently exited
 SwiftData disk phases. The seven-function R0 subset, all 49 cap probes, rooted
 20-case consumer and independent policies retain their exact checks.
 
+The additional QA wave adds 19 behavior functions, for exactly 175:
+JSON 97, JSONStreaming 14, JSONSchema 41, JSONRegexRuntime 12 and JSONPatch 11.
+All 48 earlier Swift test files and 173 parameterized invocations are retained.
+The focused `python3 Scripts/ci.py --lane qa` checks exact counts 11/3/3/2 in
+debug, release and Address Sanitizer. It supplements the full native/safety lanes
+and does not replace independent qualification or any of the 49 cap probes.
+Thread Sanitizer checks five shared-state cases, including concurrent payload
+opening and separate read scopes over one immutable root. See [QA](../Architecture/Release1.1/QA.md).
+
 | Independent input | Observed policy |
 | --- | --- |
 | JSONTestSuite, pinned in FixtureRegistry.json | 318 discovered: 101 accepted, 186 rejected, 31 strict/finite-policy rejections |

@@ -365,12 +365,16 @@ documentation. Receipts record actual commands, toolchains, source hashes and ca
 counts. Policy exclusions are reported separately from evaluated successes. See
 [qualification](Documentation/Qualification/README.md) and [CI](Documentation/CI/README.md).
 
-Performance is measured separately on native release builds. The current reviewed
+Performance is measured separately on native release builds. The released 1.0
 streaming comparison is approximately **11% slower**, exceeding the unchanged 10%
 review trigger, with its accepted disposition and measurement limits documented in
 [performance notes](Documentation/Performance/README.md). Logical retention remains
 separate from process RSS. Scalar scanning and plan-local immutable regex sharing
 are the qualified choices. Projection and JSON text-sequence framing remain deferred.
+The 1.1 candidate's separate matched stream comparison is **15.76% slower**; its
+[integrated review](Documentation/Architecture/Release1.1/IntegratedPerformance.md)
+retains the accepted disposition. The [additional QA wave](Documentation/Architecture/Release1.1/QA.md)
+records conformance coverage and the Base64 accounting correction.
 
 ## Contributing, support and licensing
 

@@ -6,7 +6,7 @@
 from preflight import collect
 
 ROOT = pathlib.Path(__file__).resolve().parents[1]
-SHARED_TESTS = 'independentCodecFacadesAndExactValuesCanBeSharedAcrossTasks|concurrentSchemaEvaluationsHaveExclusiveDynamicRegexAndAnnotationState|sharedRegexProgramHasExclusiveConcurrentSearchState|immutableIndexedObjectCanBeSharedAcrossTasks'
+SHARED_TESTS = 'independentCodecFacadesAndExactValuesCanBeSharedAcrossTasks|concurrentSchemaEvaluationsHaveExclusiveDynamicRegexAndAnnotationState|sharedRegexProgramHasExclusiveConcurrentSearchState|immutableIndexedObjectCanBeSharedAcrossTasks|qaSharedPayloadAndRootHaveIndependentConcurrentReadScopes'
 R0_TESTS = 'readerProbe|indexProbes|duplicateRecognitionPrecedesValueAndClosingSyntax'
 
 def snapshot():
@@ -15,6 +15,12 @@ def snapshot():
 def commands(lane, output):
     # The qualification executable resolves output paths from its repository root.
     output = output.relative_to(ROOT)
+    if lane == 'qa':
+        return [
+            ['swift', 'test', '--filter', 'qa'],
+            ['swift', 'test', '-c', 'release', '--filter', 'qa'],
+            ['swift', 'test', '-c', 'release', '--sanitize', 'address', '--scratch-path', '.build/ci-address', '--filter', 'qa'],
+        ]
     if lane == 'native':
         return [
             ['python3', '-B', 'Scripts/public_check.py'],
@@ -71,7 +77,7 @@ def check_result(command, code, log):
     counts = [int(n) for n in re.findall(r'Test run with (\d+) tests? in[^\n]*passed', log)]
     passed = code == 0
     if command[:2] == ['swift', 'test']:
-        expected = [7] if R0_TESTS in command else [3] if 'Tools/MCPReadiness' in command or 'Tools/MobileQualification' in command else [1] if 'Tools/Qualification' in command else [1, 1, 2] if 'thread' in command else [9, 11, 12, 38, 86]
+        expected = [2, 3, 3, 11] if 'qa' in command else [7] if R0_TESTS in command else [3] if 'Tools/MCPReadiness' in command or 'Tools/MobileQualification' in command else [1] if 'Tools/Qualification' in command else [1, 1, 3] if 'thread' in command else [11, 12, 14, 41, 97]
         passed = passed and sorted(counts) == expected
         if 'Tools/Qualification' in command:
             probes = re.findall(r'^CAP_PROBE (\w+) (\S+)$', log, re.MULTILINE)
@@ -89,7 +95,7 @@ def check_result(command, code, log):
 
 def main():
     parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument('--lane', choices=['native', 'safety', 'docs', 'r0'], required=True)
+    parser.add_argument('--lane', choices=['native', 'safety', 'docs', 'r0', 'qa'], required=True)
     parser.add_argument('--output')
     args = parser.parse_args()
     output = ROOT / (args.output or '.build/ci/' + args.lane)