CDGT cheat sheet

SDLC phases map to test gates: requirements review, design review, implementation (unit tests), integration, system/alpha, beta/soft launch, release candidate, live operations.

Verification asks whether the build matches the spec; validation asks whether the game is fun, fair, and shippable - game testing requires both.

Alpha = feature-complete, internal testers; beta = content-complete and feature-locked, external testers; launch candidate = only critical fixes accepted.

Entry criterion is a testable build plus documented expected behavior; exit criterion is a measured defect threshold (e.g., zero open S1/S2, S3 rate below agreed cap).

Shift-left: a defect found in requirements costs roughly 1x to fix; in production 30-100x, because it may require a client patch and certification re-run.

Regression suite scope grows with each feature; every fixed defect should add at least one permanent regression case.

Live-service titles add seasonal content and hotfix cycles, so test plans must cover patch compatibility with saved data from previous versions.

Risk-based prioritization ranks by severity x likelihood; monetization, save data, and account systems rank highest because failures are unrecoverable.

Equivalence partitioning: one representative per valid and invalid class (e.g., currency values 0, 1, max-1, max, max+1, negative).

Boundary value analysis targets min-1, min, min+1, max-1, max, max+1 for every numeric limit such as inventory slots or party size.

Decision tables enumerate condition combinations (e.g., has item x has currency x inventory full) to derive rules and collapse impossible rows.

State transition testing models menus, matchmaking, and save/load flows; cover every state, every valid transition, and every invalid transition attempt.

Pairwise/orthogonal array design reduces full combinatorial coverage (e.g., 5 resolutions x 4 GPUs x 3 OS) to a tractable set covering every pair.

A well-formed test case states ID, precondition, steps, test data, expected result, and postcondition; expected results must be observable.

Positive, negative, and exploratory charters are complementary; scripted cases prove known behavior, charters find unknown behavior.

Traceability matrix links requirement IDs to test case IDs so coverage gaps are visible before execution begins.

Combat: verify damage formula inputs (base, modifiers, resistances), rounding, critical multipliers, cooldown timers, and immunity/overflow edge cases.

Physics: test collision at high velocity (tunneling), slopes, moving platforms, ragdoll recovery, and frame-rate-independent movement.

Progression: XP curves, level caps, respec, and unlock gating; confirm no sequence break allows content skipping.

Inventory and economy: stack limits, item duplication via trade/mail/auction, currency overflow, and rollback after a failed transaction.

Save/load: verify autosave triggers, corrupted-save handling, save version migration, and cross-device cloud sync conflicts.

AI behavior: verify aggro radius, pathfinding around obstacles, state machine transitions, and behavior under unreachable targets.

Quest systems: prerequisite chains, parallel quests, abandonment, and reward duplication on repeated turn-in.

UI/HUD: verify input mapping on keyboard, mouse, and gamepad, focus order, and correct display of localized and numeric values.

Certification (TRC/TCR/Lotcheck) failures block release; a failed submission can delay launch by days to weeks depending on queue.

Controller requirements: system button behavior, disconnect/reconnect handling, and correct button glyph display per platform.

Account and entitlement: sign-out during gameplay must pause or exit gracefully; entitlement loss must revoke content without data loss.

Storage: handle full storage, removable storage removal, and save corruption without crashing; respect platform save-size limits.

Network loss (e.g., suspend/resume, airplane mode) must not hang the title; defined error messaging is required.

Parental controls and privacy: age-gate, purchase restrictions, and telemetry consent must follow platform policy.

Achievements/trophies must unlock reliably and be recoverable after offline play; avoid unobtainable or retroactively awarded states.

Submission package: correct metadata, age ratings, region settings, and store assets; verify the exact build submitted matches the certified build.

Test under simulated latency, jitter, and packet loss (e.g., 100/200/400 ms, 1-5% loss) rather than only on a clean LAN.

Host migration and peer disconnect must preserve match state or end the match cleanly without desync or duplicated rewards.

Client-side prediction and server reconciliation: verify no rubber-banding, teleporting, or hit-registration mismatch under lag.

Matchmaking: verify skill/region/party-size filters, queue timeouts, backfill, and party leader transfer.

Anti-cheat and authority: confirm the server validates movement, currency, and damage; client-only trust is a defect.

Dedicated server capacity: test at peak concurrent users, server spin-up time, and graceful degradation when capacity is exceeded.

Online services: login, session tokens, friends, leaderboards, and store transactions; test service outage and retry behavior.

Cross-play/cross-progression: verify account linking, platform-specific restrictions, and data consistency across platforms.

Frame time budget at 60 FPS is 16.67 ms; at 30 FPS it is 33.3 ms. Measure frame time percentiles (1% low, 0.1% low), not only average FPS.

Memory: watch peak and sustained usage, leaks across level transitions, and platform memory limits that trigger OS termination.

Soak testing runs 8-72 hours to expose leaks, handle exhaustion, and save corruption; log memory and object counts over time.

Load-time and streaming tests cover first boot, level transitions, and texture streaming hitches during traversal.

Compatibility matrix covers min/recommended spec, GPU vendors, OS versions, resolutions/aspect ratios, and input devices.

Thermal throttling on mobile/console: test sustained sessions and confirm frame rate stays within the certified floor.

Crash triage uses call stack, minidump, and repro rate; a 1-in-1000 crash on a common path may outrank a frequent cosmetic bug.

Network and disk I/O performance matter for online titles; measure server tick rate and database latency under load.

Pseudo-localization (accented, expanded text) exposes hard-coded strings, truncation, and layout overflow before real translations arrive.

Text expansion: German and Russian commonly run 30-50% longer than English; verify UI containers, buttons, and subtitles accommodate it.

Locale rules: number, date, currency, and plural formats; right-to-left languages require mirrored layouts and correct glyph shaping.

Font coverage: verify CJK, Cyrillic, Arabic, and symbol glyphs render without tofu boxes; check fallback fonts.

Audio/video localization: verify subtitle sync, dub tracks, and region-specific content edits.

Accessibility: remappable controls, subtitles/captions, colorblind-safe cues (do not rely on color alone), and scalable UI text.

Screen reader and menu narration support varies by platform; verify focus order and that all interactive elements are announced.

Difficulty and motor accessibility: adjustable timing, hold-to-toggle options, and assist modes must be tested for correct behavior.

Severity = impact on the product (crash, data loss, blocker); priority = urgency of fix. They are independent and can disagree.

A reproducible bug report contains build/version, platform, preconditions, exact steps, expected vs actual, frequency, and evidence (log, video, dump).

S1/S2 typically mean crash, data loss, progression blocker, or certification failure; S3/S4 are cosmetic or minor.

Triage ranks by severity, frequency, and exposure; a rare crash on a mandatory path can outrank a common cosmetic issue.

Deduplicate by root cause, not symptom; link related reports and keep one canonical repro.

Regression verification requires re-testing the fix plus adjacent areas; reopen if the original repro still fails.

Defect metrics: defect density per feature, reopen rate, escape rate (bugs found post-release), and mean time to fix.

Never close a bug as cannot-reproduce without documenting the attempted environment and build; mark it for periodic retest.

Automation ROI is highest for stable, repetitive checks: smoke tests, build verification, save/load, and data validation.

Game automation often uses telemetry-driven bots and scripted input; visual validation uses screenshot diffing with tolerance thresholds.

CI pipeline: every commit triggers build, automated smoke suite, and crash check; a red build blocks merges.

Test data management: seeded RNG and deterministic replays make automated results reproducible.

Coverage metrics: requirement coverage, code coverage (unit level), and platform matrix coverage; none alone proves quality.

Defect metrics: open count by severity, arrival rate vs fix rate, and defect removal efficiency (pre-release found / total found).

Flaky tests must be quarantined and fixed; a flaky suite erodes trust and is often worse than no automation.

Log and telemetry analysis (crash reporting, funnel data) supplements manual testing and reveals issues at scale.