recon_qa · v1.0.0 · 2026-11-02 · sha256 cd17bcc0f6527e9a
recon_qa v1.0.0A
Immutable. This exact content never changes and is served at /api/v1/blob/cd17bcc0f6527e9a.
--- name: recon_qa description: > Quality assurance for reconciliation workbooks and outputs. Validates formula accuracy, cross-sheet consistency, balance tie-outs, category completeness, and common spreadsheet errors. Companion skill to recon_analysis — run after export to verify output integrity. Triggers: QA recon, audit recon workbook, check recon output, validate reconciliation, sanity check recon, review recon export, reconciliation quality check, verify recon results. data_domain: "reconciliation output validation, workbook audit, formula accuracy, balance tie-out, cross-sheet consistency, data integrity check" created_date: 2026-05-20 version: 1.0.0 license: Snowflake Skills License. See LICENSE.txt for complete terms type: snowflake language: en status: Published --- # Reconcile QA ## Intent Post-reconciliation quality assurance skill that validates the integrity of reconciliation outputs (Excel workbooks, CSV exports, or in-memory result sets). Designed as a companion to `recon_analysis` — run after a reconciliation export to verify the output is accurate, internally consistent, and ready for sign-off. For controllers, audit teams, and accounting managers who need confidence that recon outputs are correct before distribution. This skill is for output validation only. Not for: running reconciliations (use `recon_analysis`), modifying data, creating journals, or building dashboards. ## Behavior ### Routing Logic Detect scope and route to the appropriate validation depth: | User Intent | Scope | Description | |---|---|---| | "QA this recon", "check my recon output" | **Standard** | Run formula + consistency checks on the recon workbook | | "Full audit", "deep QA", "sign-off ready?" | **Deep** | Standard + balance tie-outs + completeness + reasonableness | | "Quick check", "sanity check" | **Quick** | Summary-level consistency only (match rate math, row counts) | | Points to a specific sheet or range | **Targeted** | Validate only the specified scope | ### Standard Validation Checks (All Scopes) #### 1. Structural Integrity | Check | What to look for | Severity | |---|---|---| | Sheet presence | Expected sheets exist (Summary, Discrepancies, Matched, Unmatched, Discrepancy Analysis, Metadata) | CRITICAL | | Column headers | Required columns present in each sheet | CRITICAL | | Row count consistency | Sum of Matched + Discrepancy + Unmatched rows = Total in Summary | CRITICAL | | Empty sheets | Any data sheet with zero rows (unexpected) | WARNING | | Data types | Numeric columns contain numbers, date columns contain dates | WARNING | #### 2. Mathematical Accuracy | Check | What to look for | Severity | |---|---|---| | Match rate calculation | (Matched + Within Threshold) / Total × 100 = reported match rate | CRITICAL | | Category counts | Sum of per-category counts = total discrepancy count in summary | CRITICAL | | Absolute diff correctness | For each discrepancy row: ABS(Source1 - Source2) = reported diff | CRITICAL | | Percent diff correctness | Diff / Source2 × 100 = reported percent (handle div-by-zero) | WARNING | | Threshold classification | Items marked "Within Threshold" actually fall within configured thresholds | CRITICAL | #### 3. Cross-Sheet Consistency | Check | What to look for | Severity | |---|---|---| | Key uniqueness | Each join key appears at most once per sheet (no accidental duplication) | CRITICAL | | Discrepancy ↔ Analysis | Every row in Discrepancies sheet has a corresponding row in Discrepancy Analysis | WARNING | | Source data coverage | Keys in Matched + Discrepancies + Unmatched = union of keys across source sheets | CRITICAL | | Category labels | Categories in detail sheets match exactly the categories in Summary breakdown | WARNING | #### 4. Metadata Validation | Check | What to look for | Severity | |---|---|---| | Timestamp present | Generated timestamp exists and is within last 24 hours | INFO | | Parameter recording | Date range, thresholds, profile name all recorded | INFO | | Source identification | Both sources clearly identified | WARNING | | Row count audit | Metadata row counts match actual sheet row counts | CRITICAL | ### Deep Validation Checks (Deep Scope Only) #### 5. Balance Tie-Outs | Check | What to look for | Severity | |---|---|---| | Source1 total | SUM of Source1 values across all matched rows = expected source total (if available) | CRITICAL | | Source2 total | SUM of Source2 values across all matched rows = expected source total | CRITICAL | | Net discrepancy | Total discrepancy amount = Source1 total - Source2 total (for numeric fields) | CRITICAL | | Orphan value impact | Sum of Source1 Only amounts + Sum of Source2 Only amounts accounted for | WARNING | #### 6. Category Completeness | Check | What to look for | Severity | |---|---|---| | All discrepancies categorized | Zero rows with NULL or empty Category | CRITICAL | | All categories have reasons | Zero rows with NULL or empty Comment/Reason | WARNING | | Owner assignment | If triage fields configured: all discrepancies have an owner | WARNING | | Action assignment | If triage fields configured: all discrepancies have an action | WARNING | #### 7. Reasonableness | Check | What to look for | Severity | |---|---|---| | Match rate vs history | If prior run metadata available: match rate within ±5% of historical average | INFO | | Discrepancy concentration | Single entity/key accounting for >50% of total discrepancy value | WARNING | | Threshold abuse | >30% of "matches" are actually "within threshold" (suggests thresholds too loose) | WARNING | | Sign consistency | Discrepancies predominantly one-directional (Source1 > Source2 or vice versa) | INFO | | Stale data | Source data dates significantly older than recon run date | WARNING | ### Output Present findings as a validation report: ``` RECONCILIATION QA REPORT ========================= Workbook: <filename> Scope: <Standard | Deep | Quick | Targeted> Checks run: N Checks passed: N FINDINGS: | # | Sheet | Cell/Range | Severity | Category | Issue | Impact | |---|---|---|---|---|---|---| SUMMARY: Critical: N | Warning: N | Info: N Verdict: <CLEAN | ISSUES FOUND | BLOCKED> ``` **Severity definitions:** - **CRITICAL** — Output is incorrect or misleading; must fix before distribution - **WARNING** — Potential issue; review before sign-off - **INFO** — Style or best-practice observation; no action required **Verdict logic:** - **CLEAN** = 0 Critical, 0 Warning - **ISSUES FOUND** = 0 Critical, ≥1 Warning - **BLOCKED** = ≥1 Critical — do NOT distribute until resolved ### Guardrails - This skill validates — it does NOT fix. Report findings; let the user decide on corrections. - NEVER modify the workbook or recon results. If fixes are needed, suggest re-running `recon_analysis` with corrected parameters. - If the workbook structure is unrecognizable (not a recon output), inform the user and suggest pointing to the correct file. ### Confidence Handling | Confidence | Behavior | |---|---| | HIGH (workbook structure matches expected recon output format) | Run all applicable checks for the selected scope | | MEDIUM (workbook has some expected sheets but non-standard structure) | Run checks that apply; skip structural checks with a note; report what was validated | | LOW (file is not a recognizable reconciliation output) | Report: "This doesn't appear to be a reconciliation workbook. Expected sheets: [list]. Found: [list]." Suggest correct file. | ### Error Handling | Scenario | Action | |---|---| | File not found | Report path error; ask for correct file location | | File is not Excel/CSV | Report format error; list supported formats | | Password-protected workbook | Report: cannot open; ask user to provide unprotected version | | Partial workbook (some sheets missing) | Run available checks; note missing sheets as CRITICAL finding | | Very large workbook (>100K rows) | Sample-check rather than full-scan; note sampling in report | ## What to Provide To run QA, point this skill at one of the following: - **A reconciliation workbook** (Excel `.xlsx`) — produced by `recon_analysis` or any equivalent reconciliation tool. Expected sheets: Summary, Discrepancies, Matched, Unmatched, Discrepancy Analysis, Metadata. The skill will validate all sheets that are present. - **A CSV export** — single-sheet export from a reconciliation run. The skill will run applicable checks for the available structure. - **An in-memory result set** — if you just ran a reconciliation in this session, refer to it directly and the skill will validate the results in context. Optionally, provide **prior run history** (a table or export of past reconciliation runs for the same profile) to enable reasonableness checks — e.g., flagging when the current match rate is significantly lower than historical average.