chore: add follow-up tasks from Round 73

TASK-203: Empty blob PK encoding divergence (LOW)
TASK-204: Fix PK UPDATE test schema mismatch (LOW)
TASK-205: Fix inventory app test (LOW)

Author: subtleGradient

.tasks/triage/TASK-203-empty-blob-pk-encoding-divergence.md

@@ -0,0 +1,56 @@
+# TASK-203 — Empty Blob PK Encoding Divergence
+
+## Goal
+Fix the PK blob encoding for empty blob primary keys to match Rust/C oracle.
+
+## Status
+- State: triage
+- Priority: LOW (edge case, not blocking sync)
+- Discovered: 2025-12-20 (TASK-133)
+
+## Problem
+
+When a table has an empty blob (`X''`) as its primary key value, the encoded PK in `crsql_changes` differs:
+
+```
+Zig:    0105
+Rust/C: 0104
+```
+
+## Reproduction
+
+```bash
+cd /Users/tom/Developer/effect-native/cr-sqlite
+bash zig/harness/test-pk-blob-parity.sh
+```
+
+Look for test WF-028.
+
+## Impact
+
+This only affects tables with:
+1. BLOB primary keys
+2. That contain empty blob values
+
+This is a rare edge case and doesn't affect sync correctness (the blob value itself syncs correctly, just the encoding differs).
+
+## Files to Investigate
+
+- `zig/src/pack_columns.zig` — PK blob encoding logic
+- `core/rs/core/src/pack_columns.rs` — Rust reference
+
+## Acceptance Criteria
+
+1. [ ] Empty blob PK encoded as `0104` (matching Rust/C)
+2. [ ] `bash zig/harness/test-pk-blob-parity.sh` passes 9/9
+
+## Parent Docs / Cross-links
+
+- Test: `zig/harness/test-pk-blob-parity.sh` (WF-028)
+- Related: `.tasks/done/TASK-133-pk-blob-format-edge-case-parity.md`
+
+## Progress Log
+- 2025-12-25: Created from Round 73 test results.
+
+## Completion Notes
+(Empty until done.)

.tasks/triage/TASK-204-fix-pk-update-test-schema.md

@@ -0,0 +1,56 @@
+# TASK-204 — Fix PK UPDATE Test Schema Mismatch
+
+## Goal
+Fix `test-pk-update.sh` Test 1d to use correct pks table schema.
+
+## Status
+- State: triage
+- Priority: LOW (test bug, not implementation bug)
+- Discovered: 2025-12-25 (Round 73)
+
+## Problem
+
+Test 1d in `zig/harness/test-pk-update.sh` uses outdated column names:
+
+```sql
+-- Test currently uses:
+SELECT COUNT(*) FROM foo__crsql_clock c JOIN foo__crsql_pks p ON c.pk = p.pk WHERE p.pks = X'010901';
+
+-- But the actual schema is:
+CREATE TABLE "foo__crsql_pks" (__crsql_key INTEGER PRIMARY KEY, "id")
+```
+
+The columns `pk` and `pks` don't exist. The actual columns are:
+- `__crsql_key` (the key)
+- `id` (the actual PK column value)
+
+## Fix
+
+Update the test to use:
+```sql
+SELECT COUNT(*) FROM foo__crsql_clock c WHERE c.key = (SELECT __crsql_key FROM foo__crsql_pks WHERE id = 1);
+```
+
+Or use the clock table's `key` column directly:
+```sql
+SELECT COUNT(*) FROM foo__crsql_clock WHERE key = <expected_key>;
+```
+
+## Files to Modify
+
+- `zig/harness/test-pk-update.sh` — Fix Test 1d SQL queries
+
+## Acceptance Criteria
+
+1. [ ] Test 1d passes
+2. [ ] All 16 PK UPDATE tests pass
+
+## Parent Docs / Cross-links
+
+- Related: `.tasks/done/TASK-110-zig-pk-update-compound-text-pk.md`
+
+## Progress Log
+- 2025-12-25: Created from Round 73 findings.
+
+## Completion Notes
+(Empty until done.)

.tasks/triage/TASK-205-fix-inventory-app-test.md

@@ -0,0 +1,57 @@
+# TASK-205 — Fix Inventory App Simulation Test
+
+## Goal
+Fix `test-app-inventory.sh` which fails for BOTH Zig AND Rust/C implementations.
+
+## Status
+- State: triage
+- Priority: LOW (test bug, not implementation bug)
+- Discovered: 2025-12-25 (Round 73)
+
+## Problem
+
+The inventory app test fails with both implementations showing sites don't converge:
+
+```
+Step 3: Verify convergence
+  FAIL: Sites did not converge
+  A: PROD-A|site-a|150
+PROD-B|site-a|75
+  B: PROD-A|site-b|200
+PROD-C|site-b|50
+  C: PROD-B|site-c|100
+PROD-C|site-c|125
+```
+
+Since both implementations fail identically, this is a test design issue, not an implementation bug.
+
+## Root Cause (Suspected)
+
+The test creates inventory at different sites but:
+1. Each site creates different products (PROD-A at site-a, PROD-B at site-a, etc.)
+2. The sync only transfers data, not schema
+3. Sites end up with their local data only
+
+The test should either:
+1. Have all sites share the same products first
+2. Or verify that each site has ALL products after sync (not just their local ones)
+
+## Files to Modify
+
+- `zig/harness/test-app-inventory.sh` — Fix test logic
+
+## Acceptance Criteria
+
+1. [ ] Test passes for both Zig and Rust/C
+2. [ ] Or document as intentional behavior and mark test as xfail
+
+## Parent Docs / Cross-links
+
+- Test script: `zig/harness/test-app-inventory.sh`
+- Related: `.tasks/done/TASK-194-real-world-app-simulation.md`
+
+## Progress Log
+- 2025-12-25: Created from Round 73 findings.
+
+## Completion Notes
+(Empty until done.)