RUM-18135: Add a cold-start benchmarking harness and methodology guide

[Valpertui]

What does this PR do?

Adds a cold-start benchmarking harness, a customer-facing methodology guide, and an agent skill: the tooling and the written contract for answering "how much does the SDK cost at app startup?" on a given app and device.

Nine commits, purely additive (12 new files, no SDK code touched). Each adds one reviewable unit in its final form; no commit re-touches an earlier one, so the history reads foundation -> prerequisite -> measure -> analyse -> attribute -> prove -> document:

#	commit	contents
1	Add shared device controls for the cold-start harness	lib.sh, .gitignore
2	Add the SDK-liveness oracle	verify_sdk_active.sh
3	Add the A/B cold-start runner	coldstart_bench.sh
4	Add the paired per-block statistics	ab_stats.py, fp_simulation.py
5	Add optional Perfetto trace attribution	capture_trace.sh, verify_trace.py
6	Add regression tests for the harness contracts	tests/test_harness_regressions.py
7	Add the operator reference for the harness	tools/coldstart-benchmark/README.md
8	Document how to measure the SDK's cold-start impact	docs/benchmarking_sdk_cold_start.md, linked from the root README
9	Add a coldstart-benchmark skill for coding agents	.claude/skills/coldstart-benchmark/SKILL.md

All nine are prefixed RUM-18135:. Reviewing commit-by-commit is the intended path; commit 6 is the contract the five preceding commits are held to.

The harness:

• verify_sdk_active.sh: step zero. Installs, md5-attests the install against the local file, launches via the real launcher intent, and proves Datadog.initialize() actually ran. The oracle is the datadog-* thread CoreFeature.initialize() always creates (setupExecutors() then an immediate NTP-sync submit, CoreFeature.kt:265-266) and that R8 cannot rename, because the name is built at runtime from a string template.
• coldstart_bench.sh: the A/B (or A/A). ABBA-counterbalanced blocks, a verified uninstall and md5 attestation per install, pre-granted runtime permissions whose effective outcome is hashed into the CSV and re-checked on every install, a pre-registered warm-up discard, and a per-launch assertion that the launch reported Status=ok / LaunchState=COLD with a numeric TotalTime, that the app owned the foreground for the whole collection window, and that the SDK was as live in the measured process as the arm claims. Aborts rather than recording a sample it cannot vouch for. Every launch that conditions the ramp, the liveness probe and the warm-ups alike, is held to the same gates, since those are what define the cell's post-install state.
• ab_stats.py: statistics. The primary endpoint is a paired test over per-block deltas; Welch and permutation are printed as [diagnostic]. Most of the file is refusals: pooling CSVs from different builds, arms, devices, Android users or protocols; counting the same file twice, including a byte-identical copy; reading an aborted run as a complete one; recovering from a file whose whole blocks are not a consecutive prefix from block 1, which the sequential collector cannot produce; dropping a launch the harness rejected or a metric missing from some launches (both are outcome-dependent, so excluding them can manufacture an improvement). It also refuses to pool runs whose arms ended up with different effective runtime permissions, and suppresses the primary interval when a row omits its validity verdict or a contributing block has no record of which arm ran first. Every override is named and leaves the primary interval suppressed.
• fp_simulation.py: reproduces the false-positive table that justifies that choice.
• capture_trace.sh / verify_trace.py: optional Perfetto attribution, with a liveness gate, an endpoint gate and a whole-window foreground check.
• lib.sh: shared helpers, so the three device-touching workflows cannot drift into different definitions of the same thing: tool resolution, the Android-user snapshot, the permission grant, the animation-scale and radio controls, the verified uninstall, cold-launch validation and the datadog-* liveness probe.
• tests/test_harness_regressions.py: 128 regression tests for the cross-script contracts, driven by a fake adb, so they need no device. They assert the refusals, not the happy path.

Motivation

Customers ask what the SDK costs at startup. The honest answer depends on their app, their device mix and their feature set, so the useful thing to publish is not a number but a method they can run, plus the caveats that decide whether the number it produces means anything.

The naive protocol does not work. Running the same APK in both arms, so the true difference is zero by construction, on a mid-range device (Helio G95, Android 12):

protocol	measured "difference"	95% CI	p	verdict
fixed arm order, permission dialogs unhandled	+12.7 ms	[+3.0, +22.5]	0.011	false positive
counterbalanced order, dialogs unhandled	+5.0 ms	[−4.4, +14.4]	0.30	null, but noisy
counterbalanced + permissions pre-granted	−0.8 ms	[−6.9, +5.2]	0.79	correct

The first row is a statistically significant regression between two identical builds. Anyone running that protocol would have reported it.

Two design decisions are worth flagging for review:

The primary endpoint is paired on blocks, not pooled over launches. Launches inside one arm×block cell share an install, an AOT compilation and a thermal state, so an unpaired test estimates the standard error from within-cell scatter only. Counterbalancing removes the ordering bias; it does nothing about this. By simulation (true effect zero, within-launch sd 11 ms, per-cell shift sd σ_b), false-positive rate of a nominal-95% interval:

design	σ_b = 0	σ_b = 2	σ_b = 4	σ_b = 8
2 blocks × 15 launches, unpaired	4.8%	10.0%	23.4%	45.5%
8 blocks × 4 launches, paired on block deltas	5.1%	5.0%	4.9%	5.2%

A 4 ms between-block shift is ordinary. ./fp_simulation.py reproduces this table using ab_stats.py's own interval code, so it cannot drift away from the tool.

Most controls bias the measured cost downward, and the guide says so. Discarded warm-ups, am force-stop (which does not evict the page cache), TTID-only measurement and pre-granted permissions all shrink the number. Two do not, and the table says which way they go instead: the default speed-profile compile lands at verify on a fresh install, which is pessimistic rather than flattering, and animations off enlarged the total on the app measured (+88.6 ms against +40.2 ms on TTFD) despite suppressing per-frame SDK work. docs/benchmarking_sdk_cold_start.md tabulates the direction of each. The metric is a process-cold, page-cache-warm start to first frame, and is closer to a lower bound than a worst case.

Additional Notes

Findings the guide documents that contradict the intuition, each measured rather than assumed:

• TTID is the wrong endpoint on framework apps. On one React Native app first frame was under a third of startup (~630 ms TTID against ~2075 ms TTFD), so a TTID-only comparison could not have seen SDK cost in the other two thirds.
• Session Replay's startup increment was ~9 ms, inside the noise, despite being the heaviest feature while recording. The cost was in the core SDK.
• The largest block of SDK CPU we measured is not a startup cost. ~149 ms of com.datadog.* JIT (~390 ms with Session Replay), across ten traces. Timestamped: zero of it began before reportFullyDrawn(); the first started 406–500 ms after, in all ten traces. It is post-launch background CPU. A Baseline Profile would not have moved TTID or TTFD on that app.
• Traces tell you what work exists and where, not how much it costs. Tracing itself lengthened the measured window by ~7% in every arm, and five traces per arm resolved it only to ±32 ms.

On docs/sdk_performance.md: the guide explicitly marks those figures as predating this protocol (5 launches per arm, fixed order, no A/A, no confidence intervals) so they are not mistaken for a comparable baseline. No numbers there are changed by this PR.

Safety. All three device-touching scripts uninstall/reinstall the app under test and pre-grant its runtime permissions, including verify_sdk_active.sh, which is documented as step zero and is just as destructive. Every readable device setting they touch is snapshotted and restored from an EXIT trap; INT/TERM exit into that trap, so Ctrl-C stops the run and restores the device, once. Two controls, fixed-performance mode and the background dexopt job, have no Android getter, so those are reversed on a best-effort basis and only when the script successfully issued the command itself.

Package state on Android is per-user, so one numeric Android user is resolved once and passed explicitly to every package, permission and activity command, and stamped into the CSV header. Only permissions found denied are granted, and only those are revoked afterwards, so a grant the app already held is left alone, and pm reset-permissions, which would drop every other app's grants, is never used. The scripts refuse to run when the package is also installed for another user, because host-side adb uninstall has no user selector and would delete that profile's app data.

A preflight refuses to run when PKG does not match the APKs (every block runs adb uninstall $PKG) or when the two arms declare different versionCode/versionName; a missing aapt2 is a hard failure rather than a downgrade to a warning. .gitignore covers results_*.csv, bench_*.log and *.pftrace, since a Perfetto capture is device-wide.

No CHANGELOG entry: tooling and docs only; no shipped SDK code changes.

Operator-visible behavior worth knowing before you run it

Three things will change what an operator sees, beyond the statistics:

• capture_trace.sh never overwrites a .pftrace. The name is checked before any device
  mutation and again immediately before adb pull. This matters because a rejected capture is
  deliberately left on disk for diagnosis, so the file that explains the rejection is also the
  file that now blocks re-running the same name. Move it or pick another name. The old behavior
  destroyed the evidence silently, which is worse, but the iterate loop is two commands now.
• TRACE_ENDPOINT (total_ms default, or ttfd / app_trace_ms) makes the capture prove it
  reached the same endpoint the A/B measured, and fails the capture when it did not. A trace that
  stops short of the window under study cannot explain that window.
• ALLOW_MISSING_LAUNCH_MARKER exists because the whole-window foreground check prefers
  ActivityManager's global launching: <pkg> slice, which is an ActivityMetricsLogger
  implementation detail rather than a documented contract. On a device that never emits it, set
  this to 1: the lifecycle-only checks still run and the verdict becomes held-lifecycle-only
  rather than held, so a capture whose foreign-takeover clause never ran is never reported as
  clean.
• A run aborts rather than proceeding on an unproved control. settings put reports success
  on a device that ignored it, so the three animation scales and both radio settings are read
  back and required to match before anything is measured; an app that declares a permission it
  can never hold (a hard-restricted one such as SMS) aborts unless ALLOW_PARTIAL_PERMISSIONS=1
  accepts the weaker guarantee; and the results CSV and its log are reserved with exclusive
  creates before the first device command, so two runs started against two devices in the same
  second cannot interleave or truncate each other's evidence. Each of these has a named override
  or a clear message naming the file or setting involved.
• AIRPLANE is a radio-state control, not a connectivity one. It sets and verifies Wi-Fi and
  mobile data, in both directions. It deliberately does not claim association, DNS or endpoint
  reachability, since nothing runnable at preflight establishes those for a whole run, so keeping the
  external network condition stable is the operator's job, and the CSV records only what was
  proved.

How to validate

Static analysis (from tools/coldstart-benchmark/):

shellcheck -x -s bash *.sh              # clean
python3 -m py_compile *.py              # clean
python3 -m pyflakes *.py                # clean
python3 -m unittest discover -s tests   # 128 tests, no device needed
./fp_simulation.py                      # ~20s; reproduces the table above

End-to-end, against a device or emulator:

cd tools/coldstart-benchmark
export PKG=<your.app.id>

./verify_sdk_active.sh app-with-datadog.apk "$PKG"          # exit 0 = SDK live

EXPECT_A=1 EXPECT_B=1 LABEL_A=A1 LABEL_B=A2 WARMUP=1 \
  ./coldstart_bench.sh app.apk app.apk 1 4                  # A/A, ~7 min
./ab_stats.py results_<timestamp>.csv --baseline A1 --treatment A2

An A/A run must produce a mean block delta near zero with a CI straddling it, and no significant order effect.

Review checklist (to be filled by reviewers)

• Feature or bugfix MUST have appropriate tests (unit, integration, e2e)
• Make sure you discussed the feature or bugfix with the maintaining team in an Issue
• Make sure each commit and the PR mention the Issue number (cf the CONTRIBUTING doc)

[datadog-official]

✅ All CI checks and tests passed. Datadog automation helped this PR pass.

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

🔄 Datadog retried 2 tests - 2 passed on retry

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 71.76% (+0.02%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: e5d34a5 | Docs | View more details | Give us feedback!}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dcc1de7649

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5e229ec8a7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 597ffb303b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4dc32909e2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 652dbe8e37

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[OliviaShoup]

thanks for the PR! it looks great. the only must-fix issues are British spellings because our style guide prefers American English (sorry lol)

also flagging: em dashes are used very heavily and consistently throughout all three new prose files. every use is correct, but it's dense enough to read as fatiguing or AI-generated to some readers. worth a look if you want to vary it up, but not a blocker

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d8983034d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7775505e82

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f9a1b2e758

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 716d40bfd9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 96d8209f6c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cb50eae5a8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a8d8fec2e3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 081ef091ff

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a5e0a003eb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 37c315ad3b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 800f46cfc9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6b7d454638

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 450dfa73fe

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a964771385

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ccaf9d5eab

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 149f67fcc1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d0099ed7cd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 88a7831fe3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Enumerate processes that use an explicit global name

When a component declares a fully qualified android:process name that is not <applicationId> or <applicationId>:..., this name-based filter omits a process owned by the APK. If Datadog initializes only there, the treatment arm is rejected as inactive; worse, if another conventionally named app process exists, an accidentally SDK-active baseline can pass the absence gate without inspecting the process that owns the SDK thread. Enumerate processes by the installed package UID (failing closed for shared UIDs) rather than assuming every app-owned process uses the package-name prefix; the equivalent name filter in verify_trace.py needs the same treatment.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e5d34a50e3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Normalize decimal counts before Bash arithmetic

When a caller supplies a zero-padded count such as RUNS=010 or BLOCKS=010, this validation accepts it, but Bash arithmetic interprets it as octal while the Python analyzer parses the stamped header as decimal. The collector therefore records only 8 launches/blocks while declaring 10, writes # RUN COMPLETE, and the analyzer later refuses the expensive run as an incomplete matrix; values such as 08 can instead make the arithmetic loop error and collect nothing. Reject leading zeroes or normalize these inputs to base 10 before using or recording them.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep version validation independent of the package override

When aapt2 is unavailable and the operator follows the documented recovery of setting ALLOW_UNVERIFIED_PKG=1 after manually confirming the application ID, this branch skips all of check_apk_pair(), including its separate versionCode/versionName equality check. Baseline and treatment APKs from different app releases can therefore run without ALLOW_VERSION_MISMATCH=1, producing a completed result whose delta is not attributable to the SDK. Require the version-mismatch acknowledgement separately when the override prevents automatic version validation, rather than silently disabling that invariant with the package check.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Arm trace cleanup only after reserving the output

When $NAME.pftrace already exists or another process wins the reservation race, dd_reserve_output_files() calls die, but this EXIT trap has already been installed and cleanup() unconditionally writes every snapshotted animation, timeout, stay-awake, and radio value back through settings put/svc. The advertised pre-mutation collision check can therefore mutate the device despite refusing the capture, and can overwrite a setting changed by another actor after the snapshot. Reserve the trace path before arming mutating cleanup, or track whether each control was actually changed before restoring it.

Useful? React with 👍 / 👎.