Review

Review the current branch's changes against master (or another base if specified). Produce a structured review covering correctness, architecture, tests, and repo-specific concerns. Adapt depth to the PR type — migration PRs get extra migration-checklist items; feature/fix PRs skip migration sections.

Step 1 — Orient

git fetch origin master                    # local master may be stale
git log --oneline origin/master..HEAD      # commits in this PR
git diff --stat origin/master...HEAD       # files changed
git diff origin/master...HEAD              # full diff (read selectively)

Use the three-dot range for git diff: it diffs against the merge base, so commits landed on master after the branch was cut don't show up as spurious reversions in the PR. (git log keeps the two-dot form — for log it already means "reachable from HEAD, not from master".)

Identify the PR type:

• ObjC→Swift migration — files under PrebidMobile/Objc/ deleted and matching files under PrebidMobile/Swift/... added
• Feature / bug fix — changes to existing Swift or ObjC logic
• Tooling / CI — script or config changes
• Docs only — .md files only

Step 2 — Run the standard checks

# Build all XCFrameworks (catches CocoaPods + SPM + Carthage regressions)
./scripts/buildPrebidMobile.sh

# PR test suite (must pass; re-run once if only PBMBidRequesterTest.testBanner_300x250 fails)
./scripts/testPrebidMobile.sh --latest --quick

# SwiftLint (warnings are ok; errors block merge)
swiftlint --config .swiftlint.yml

If the full suite is warranted (final PR in a phase, or touching networking):

./scripts/testPrebidMobile.sh --latest

Step 3 — Code review

For any PR

Correctness

• Logic matches the intent described in the commit message / PR description
• No obvious off-by-one, nil-crash, or force-unwrap in production paths
• Error paths are handled; no silent swallowing of errors
• Thread safety: any shared mutable state accessed from multiple queues?

Swift quality

• No single-letter variable names (d, n, e, etc.)
• No unnecessary @objc on internal-only Swift code
• public access only where needed; prefer internal
• Prefer let over var where value never changes

iOS SDK specifics

• No changes to public API surface without explicit intent (check PrebidMobile/Swift/ public types)
• fetchDemand call sites and ad unit lifecycle unaffected
• Min deployment target stays iOS 13.0 — no API calls that require higher
• No new ObjC dependency added to the Swift layer without justification
• Never log consent strings, IDFA, or EIDs (ORTBUser.ext) — this SDK handles all three

Ad rendering — the SDK draws into someone else's window

• SDK-drawn controls (close, skip, mute, Learn More) keep a ≥44×44pt tap target. Note AdViewButtonDecorator.getButtonSize() computes 0.1 × screenWidth, which is under 44pt on most of the device fleet — don't propagate that pattern.
• No UIScreen.main.bounds for sizing or positioning. The host app's window may be a fraction of the screen under Split View, Slide Over, or Stage Manager; size from the containing view's bounds. Likewise avoid UIApplication.statusBarOrientation (deprecated since iOS 13) — read the trait environment instead.
• Overlay chrome composited over arbitrary advertiser content meets contrast floors: 4.5:1 for text, 3:1 for UI components. A translucent scrim over unknown video does not guarantee either.
• Text the SDK draws either scales with Dynamic Type (UIFontMetrics.scaledFont(for:)) or is deliberately capped for a fixed-height overlay — not silently frozen at a point size.
• When ad chrome changes, spot-check with VoiceOver on, text at Accessibility XXXL, and Reduce Transparency enabled.

Tests

• New tests run in the PR plan. PrebidMobilePRTests.xctestplan selects by exclusion (skippedTests), so new classes are picked up automatically — only verify the class isn't listed there.
• Existing tests not weakened. In particular, assertForOverFulfill = false, XCTAssert(true), and assertions relaxed to !error.localizedDescription.isEmpty hide the failure instead of fixing it; prefer isolating the test (e.g. tagging fixtures per test instance so a shared singleton's callbacks can be filtered).
• No mocked database / network when a real one is feasible

For ObjC→Swift migration PRs (additional checks)

File structure

• Swift twins land under PrebidMobile/Swift/PrebidMobileRendering/ORTB/Request/ (or mirrored path)
• Filename: ORTBFoo.swift (no PBM prefix)
• Class declaration: @objc(PBMORTBFoo) public class ORTBFoo: NSObject, PBMJsonCodable

ObjC bridge completeness (Gaps 6 & 7)

• Class is @objc public class — not just public class
• All @objc-visible properties have @objc public var
• Init: @objc(initWithJsonDictionary:) public required init(jsonDictionary:) — non-optional, calls super.init() first
• Encode: @objc(toJsonDictionary) public var jsonDictionary: [String: Any]

JSON parity

• Every ObjC property has a corresponding Swift property with the exact same JSON key string
• Empty arrays are included as-is (not suppressed) — pbmCopyWithoutEmptyVals only strips nil/NSNull (Gap 9)
• Child ORTB objects use json[.key] = childObj — empty-dict suppression is automatic (Gap 2)
• lat/lon on ORTBGeo must use NSDecimalNumber(decimal:) on encode to preserve decimal precision

NSMutableDictionary properties (Gap discovered S1.4)

• Any property typed NSMutableDictionary * in ObjC must be decoded as: NSMutableDictionary(dictionary: extDict) — NOT as? NSMutableDictionary (silently returns nil)

NSCopying (Gap discovered S1.4)

• If the ObjC type conformed to <NSCopying>, the Swift twin must also conform
• Implement via JSON round-trip: Self(jsonDictionary: jsonDictionary)

JSONObject.dict private(set) (Gap 10)

• Cannot write json.dict["ext"] = val from outside the struct
• Use var result = json.dict; result["ext"] = val; return result for untyped sub-dict injection

ObjC file cleanup

• .m file deleted ✓
• PrivateHeaders/*.h deleted ✓
• PBMORTBAbstract+Protected.h import removed from the deleted .m (not from Phase 3/4 consumers)
• PBMORTB.h umbrella updated (deleted type removed from imports)
• Remaining ObjC consumers updated: #import "PBMORTBFoo.h" → #import "SwiftImport.h"
• Xcode project: .m/.h refs removed, .swift ref added (check project.pbxproj diff)

Test file updates

• Swift test files use new ORTBFoo names (not PBMORTBFoo) — no 'has been renamed' compiler errors
• Used perl -pi -e 's/PBMORTBFoo\b/ORTBFoo/g' (not sed, which has unreliable \b on macOS)
• codeAndDecode<T: PBMORTBAbstract> overload removed once PBMORTBAbstract.m is deleted
• Decode coverage includes a partial JSON payload, not only a fully-populated fixture — round-tripping a full fixture cannot catch a resurrected class default (Gap S2.5-C)

Docs

• docs/migration/playbook.md updated if a new gap was discovered
• The PR's docs/migration/pr-phase-*.md updated with step summary and ticked test-plan boxes, and no superseded draft docs left alongside it

Step 4 — Known flaky test

PBMBidRequesterTest.testBanner_300x250 fails intermittently under full-suite simulator load but passes in isolation. Do NOT flag this as a regression unless it also fails when run alone:

xcodebuild ... -only-testing PrebidMobileTests/PBMBidRequesterTest/testBanner_300x250 \
  test-without-building

Step 5 — Produce the review

Structure the output as:

## Summary
One paragraph: what the PR does, and the overall verdict (approve / request changes / comment).

## Blockers
Numbered list of must-fix items (empty if none).

## Suggestions
Numbered list of non-blocking improvements.

## Migration checklist (migration PRs only)
[ ] ObjC bridge annotations present
[ ] JSON key strings match ObjC originals
[ ] ObjC files deleted, consumers patched, project.pbxproj clean
[ ] Test files renamed, new classes registered in PRTests plan
[ ] Playbook updated if new gap discovered
[ ] Build + quick tests green

## Nits
One-liners: style, naming, comment quality. Low priority.

Keep the review concise. Lead with blockers. Skip sections that have nothing to report.