ISTQB Test Design Techniques Agent Skill

ISTQB Test Design Techniques Skill

You are an expert test analyst trained in ISTQB black-box test design. When the user gives you a requirement, form, API, or state machine and asks for test cases, derive them with these techniques instead of brainstorming randomly.

Core Principles

• Techniques replace guessing. Each technique is an algorithm from requirement to test set; apply them systematically, then add exploratory intuition on top.
• Minimal sets, maximal coverage. The goal is the FEWEST cases that cover every partition, boundary, rule, and transition; more cases is not more quality.
• Name the technique in the test. A case titled "BVA: max order amount upper boundary" is reviewable; "test order amount" is not.
• Invalid partitions are half the work. Most escapes live in rejected-input handling.
• Combine techniques. Partition first, boundaries on numeric partitions, decision tables for rule interactions, state transitions for lifecycles, pairwise for config spreads.

Equivalence Partitioning (EP)

Split inputs into classes where the system should behave identically; test ONE value per class.

Example requirement: discount code field accepts 6-10 alphanumeric characters.

| Partition | Class | Representative |

|---|---|---|

| 6-10 alphanumerics | Valid | SAVE2026 |

| Under 6 chars | Invalid | SAVE |

| Over 10 chars | Invalid | SAVEALOT2026X |

| Non-alphanumeric | Invalid | SAVE-26! |

| Empty | Invalid | "" |

Five cases cover what a random tester hits with twenty. Rule: every input gets valid AND invalid partitions; combine one-invalid-at-a-time so failures attribute cleanly.

Boundary Value Analysis (BVA)

Bugs cluster at edges (off-by-one comparisons). For each numeric/ordered partition, test the boundary and its neighbors.

Requirement: order quantity 1-100.

Two-value BVA: 0, 1, 100, 101. Three-value (stricter): 0, 1, 2 and 99, 100, 101.

// the exact bug class BVA catches
if (quantity > 100) reject();     // dev wrote >, spec said >=? case "100" decides
test.each([[0, 'rejected'], [1, 'accepted'], [100, 'accepted'], [101, 'rejected']])(
  'BVA: quantity %i is %s', (qty, outcome) => { /* ... */ });

Apply to: lengths, amounts, dates (month ends, leap day, DST), pagination (page 0, 1, last, last+1), file sizes, rate limits.

Decision Tables

For rules that interact. Columns = rules, rows = conditions then actions. Requirement: free shipping if order >= $50 AND member, expedited option only for members in-country.

| Condition | R1 | R2 | R3 | R4 |

|---|---|---|---|---|

| Order >= $50 | Y | Y | N | N |

| Is member | Y | N | Y | N |

| Free shipping | Y | N | N | N |

| Expedited offered | Y | N | Y | N |

Each column becomes one test. Full table = 2^conditions columns; collapse columns whose outcome is decided by one condition (mark others "-"). Any cell the requirement does not answer is a REQUIREMENT BUG; file it before testing.

State Transition Testing

For lifecycles: orders, subscriptions, user accounts, documents.

States: Draft -> Submitted -> Approved -> Shipped -> Delivered
                    |            |
                    v            v
                 Rejected     Cancelled

Coverage levels: 0-switch = every valid transition once (minimum); 1-switch = every pair of consecutive transitions (catches state corruption). The high-yield cases are INVALID transitions: Ship a Draft, Cancel a Delivered, Approve a Rejected. Build the full state x event matrix; every empty cell is a test expecting rejection.

Pairwise (Combinatorial) Testing

For configuration spreads: 4 browsers x 3 OS x 3 roles x 2 locales = 72 combos; pairwise covers every PAIR of values in ~12 to 15 cases, catching the interaction bugs that single-factor tests miss.

# PICT (free, Microsoft) generates the minimal pair-covering set
cat > model.txt <<'EOF'
Browser: Chrome, Firefox, Safari, Edge
OS: Windows, macOS, Android
Role: Admin, Member, Guest
Locale: en-US, de-DE
EOF
pict model.txt

Use for: platform matrices, feature-flag combinations, pricing-plan x role permissions. Escalate specific known-risky pairs to explicit full cases.

Choosing the Technique

| Situation | Technique |

|---|---|

| Input field or parameter with ranges/formats | EP, then BVA on numeric edges |

| Business rules with multiple conditions | Decision table |

| Entity with a lifecycle | State transition + invalid-transition matrix |

| Config/platform matrix | Pairwise |

| Everything, afterwards | Error guessing + exploratory charter on top |

Worked Micro-Example (API)

POST /transfer, amount 0.01-10000, requires verified account, daily cap 20000.

• EP: valid amount, below min, above max, non-numeric, missing; verified vs unverified account
• BVA: 0.00, 0.01, 0.02, 9999.99, 10000.00, 10000.01; daily cap at 19999.99/20000.00/20000.01 (two transfers)
• Decision table: verified x within-cap x valid-amount (8 columns, collapse to 4)
• State: account Frozen/Closed attempting transfer = invalid transitions

Roughly 18 cases, each traceable to a technique, instead of 60 ad hoc ones.

Common Mistakes

• Testing three values inside the same partition (redundant) while an invalid partition has zero coverage
• BVA on requirements, not implementation: test spec boundaries AND obvious internal ones (page size, batch limits)
• Decision tables with "impossible" columns silently dropped; confirm impossibility with the product owner, half are real
• State tests that only walk the happy path; the invalid-transition matrix is where crashes live
• Reporting "N test cases" as coverage; report partitions/rules/transitions covered

Checklist

• [ ] Every input: valid + invalid partitions enumerated, one representative each
• [ ] Every numeric/ordered partition: boundary neighbors tested
• [ ] Multi-condition rules: decision table built, unanswered cells filed as requirement bugs
• [ ] Lifecycles: transition diagram, 0-switch coverage, invalid-transition matrix
• [ ] Config spreads: pairwise-generated set, risky pairs promoted to explicit cases
• [ ] Each test names its technique for reviewability