docs: add hypothesis invalidation tasks (TASK-127, TASK-128)

Author: subtleGradient

.tasks/backlog/TASK-127-experimental-parity-invalidation.md

@@ -0,0 +1,27 @@
+# TASK-127: Experimentally invalidate "full parity" hypothesis
+
+## Goal
+The current hypothesis is that the Zig implementation has achieved full oracle parity with the Rust/C implementation (18/18 tests pass).
+We need to **invalidate** this hypothesis experimentally by finding at least one divergence that is not yet covered by our test suite.
+
+## Scope
+- Create a new test harness `zig/harness/test-fuzz-parity.sh` (or similar).
+- Implement a simple stochastic/fuzzing approach:
+  - Generate random schemas (tables with random column types, PKs).
+  - Generate random operations (INSERT, UPDATE, DELETE, transactions).
+  - Run identical SQL against both Zig (loadable ext) and Rust/C (oracle via sqlite-cr wrapper).
+  - Compare `crsql_changes`, `crsql_db_version`, `crsql_site_id`, and table contents.
+- Run the fuzzer until a divergence is found.
+
+## Files to Modify
+- `zig/harness/test-fuzz-parity.sh` (new)
+- `zig/harness/test-parity.sh` (optional, to include the new test)
+
+## Acceptance Criteria
+- [ ] A new test script `zig/harness/test-fuzz-parity.sh` exists.
+- [ ] The script runs against both Zig and Rust/C oracle.
+- [ ] The script identifies at least one divergence (a "counter-example" to the full parity hypothesis).
+- [ ] The divergence is documented in the completion notes.
+
+## Parent Docs
+- `research/zig-cr/92-gap-backlog.md`

.tasks/backlog/TASK-128-expand-parity-suite.md

@@ -0,0 +1,23 @@
+# TASK-128: Expand parity suite with invalidation findings
+
+## Goal
+Once TASK-127 has identified divergences (invalidating the "full parity" hypothesis), we must expand our regression test suite to cover these edge cases permanently.
+
+## Scope
+- Analyze the divergence(s) found in TASK-127.
+- Create deterministic reproduction cases in the appropriate `zig/harness/test-*.sh` script (or create a new one if needed).
+- Fix the divergence in the Zig implementation (if it's a bug) or document it as a known limitation.
+- Verify the fix with the new test case.
+
+## Files to Modify
+- `zig/harness/test-*.sh` (existing suites)
+- `zig/src/*.zig` (implementation fixes)
+
+## Acceptance Criteria
+- [ ] Deterministic regression tests exist for all divergences found in TASK-127.
+- [ ] Zig implementation passes these new tests.
+- [ ] `make -C zig test-parity` includes these new tests.
+
+## Parent Docs
+- `research/zig-cr/92-gap-backlog.md`
+- `.tasks/backlog/TASK-127-experimental-parity-invalidation.md`

research/zig-cr/92-gap-backlog.md

@@ -11,6 +11,10 @@
 
 ## Now (next parallel assignments)
 
+### Hypothesis Invalidation (Next Priority)
+- [ ] **TASK-127** — Experimentally invalidate "full parity" hypothesis via fuzzing `.tasks/backlog/TASK-127-experimental-parity-invalidation.md`
+- [ ] **TASK-128** — Expand parity suite with findings from TASK-127 `.tasks/backlog/TASK-128-expand-parity-suite.md`
+
 All oracle parity tests pass. Zig implementation is now wire-compatible with Rust/C oracle.
 
 ### Open Gaps (Parity Divergences)