t1k:marketing:analytics:wiring-audit
| Field | Value |
|---|---|
| Module | analytics |
| Version | 1.21.13 |
| Effort | medium |
| Tools | — |
How to invoke
Section titled “How to invoke”/t1k:marketing:analytics:wiring-auditt1k:marketing:analytics:wiring-audit
Section titled “t1k:marketing:analytics:wiring-audit”Finds apps whose numbers are plumbed wrong, not apps whose numbers are bad.
The failure this catches is quiet by construction: package-level totals stay correct, every report still renders, nothing errors. What breaks is the split — one campaign silently absorbs another’s revenue, or spend lands on a package no report knows about. It surfaces weeks later as a campaign that “has no ad revenue”, or as a title whose ROAS looks great right up until someone adds up the spend properly.
Run it against yesterday, read the findings, act on HIGH.
Run it
Section titled “Run it”cd $HOME/.claude/skills/t1k-marketing-crawldata/toolbun audit-wiring.mjs # last 8 days, ending yesterdaybun audit-wiring.mjs --end 2026-09-14 --days 7bun audit-wiring.mjs --json ../data/wiring-2026-09-15.jsonbun audit-wiring.mjs --fail-on-findings # exit 1 on any HIGH, for CIFlags: --days (max 8, see Gotchas) · --end · --min-spend (default 5, below
which a campaign is noise) · --json · --env (credentials path; defaults to the
crawler’s credentials/.env, or $T1K_ADS_ENV) · --fail-on-findings.
Credentials are the crawler’s — no new secrets. Adjust is required; every other source degrades to a skipped section with a printed reason rather than a silent pass.
The checks
Section titled “The checks”| ID | Severity | Fires when |
|---|---|---|
A0 source-incomplete | HIGH | A source failed or was rate-limited. Its campaigns were never fetched, so a clean run below it proves nothing. |
A1 missing-campaign | HIGH | A network campaign spent real money but Adjust has no campaign row for it. |
A2 unverified-link | WARN | An app has only ever had ONE campaign on a link-based partner, so the link level can’t be proven. It will fail as A1 on the next offer. |
A3 unknown-package | WARN | Installs or revenue landing on store_id = unknown — invisible to every per-title report. |
A4 no-spend-connector | WARN | A partner delivering installs that no connector crawls spend for, so its cost is missing from Profit/ROAS. |
A5 unity-app-map-empty | HIGH | A Unity org’s apps list returned 200 with no results — a missing Advertise API Viewer role. Every app maps to unknown while spend still shows in the day total. |
M1 revenue-source-gap | HIGH | MAX reports real revenue for an app but Adjust reports ~nothing — an ad-revenue ingestion break. |
M2 spend-no-revenue | HIGH | An app spent but reported no revenue from any source. |
A1 and A2 are the ones worth building the habit around. The rest are coverage checks that happen to be cheap once the data is already in hand.
A1 / A2 — the Adjust link-level trap
Section titled “A1 / A2 — the Adjust link-level trap”This is the root cause the skill exists for. Adjust links have four levels — network → campaign → adgroup → creative — and each level has its own token. A link only creates sublinks below its own token. So on a campaign-level link:
campaign={uuid}andcampaign_id={uuid}are silently ignored — the campaign is already fixed by the tokenadgroup=andcreative=keep working perfectly, creating hundreds of sublinks
The link therefore looks fully dynamic and is not. Every offer that uses it reports under one campaign name forever.
Why staff pick the wrong link. In Campaign Lab, More options (⋯) → copy the
click URL and link token is offered at every level. The campaign row is
auto-created by the first offer’s traffic and is named after that offer’s uuid
(ss_Foo_AND_ROAS_WW_210826), so when someone sets up offer #2 they see a row
named after the campaign they are working on sitting under a generic parent —
and copying the specific-looking one is the more careful-seeming choice. The
copied URLs are identical apart from eight characters that mean nothing to a
human.
The rule: copy from the row named <Title>_<Platform>_<Network> and never
expand it. A row whose name starts with ss_ is one level too deep.
Adjust link levels are permanent — a campaign-level link cannot be promoted. The only fix is to point the network at the network-level token.
What a mis-wire costs
Section titled “What a mis-wire costs”From the case that produced this skill (com.mystery.dumpling.unbox.asmr,
2026-09-15): a second Mintegral offer ran six days on the campaign-level link,
spent $160.95, and never appeared in Adjust’s campaign dimension. Its installs
and revenue were absorbed into the first offer’s row. Package-level Profit/ROAS
stayed correct throughout — but read at campaign level the old campaign carried
two offers’ revenue against one offer’s spend: $199.59 ÷ $139.99 = 143%
ROAS, against a true combined 66%. A title well under target read as a
winner.
Nothing is recoverable. adgroup is the network’s sub-publisher id and
creative is the creative name; neither carries the offer, so days already run
cannot be split afterwards. Fixing the link only restores the split going
forward.
Confirming a suspected mis-wire by hand
Section titled “Confirming a suspected mis-wire by hand”Adjust’s reporting API exposes the token at each level — these are API dimensions, not Datascape UI columns:
dimensions=store_id,partner,campaign,campaign_token,network_token,adgroup_token,creative_tokenThe signature of a mis-wire is a single campaign_token that equals the
token in the network’s tracking URL, while adgroup_token and creative_token
run to hundreds — dynamic creation working normally, one level too low. A
correctly wired app shows several campaign_tokens under one network_token.
Cross-check installs: when two offers share a campaign link, Adjust’s installs for that campaign match the sum of both offers on the network side, not the first offer alone.
Reading the output
Section titled “Reading the output”Findings print grouped by check, HIGH first, with the correct network token
inline for A1/A2 so it can be compared against the network’s tracking URL without
a second lookup. --json emits the same findings as structured data for a
dashboard or a diff against yesterday’s run.
Act on HIGH the same day — that is live spend going unattributed. A2 is a standing to-do list: each entry is an app that will break the next time someone launches a second campaign on it.
Hard rules
Section titled “Hard rules”--daysis hard-capped at 8. Mintegral silently returns rows with no spend for a longer range, which would make every offer read as mis-wired. The cap is enforced in code, not left to the operator.- iOS packages need normalizing before joining. Adjust returns a bare numeric
App Store id (
6751266232); the spend networks returnid6751266232. Joining raw makes correctly-wired iOS apps look broken — this bit the first run of this very audit and flagged $2,801 of healthy LifeChoice IOS spend as missing. - Google Ads is deliberately excluded from A2. It is self-attributing through the partner integration with no tracking link, so the campaign-level-token trap cannot occur; flagging it would be noise. It is still covered by A1.
- A Unity Service Account missing
Advertise API Viewerfails at 200, not 403. The apps list returns{"total":0,"results":[]}, so every app maps tounknownwhile spend still looks present. The audit prints an explicit warning for this rather than reporting a clean org. - Do not chase a few percent between MAX and Adjust. They legitimately differ (different windows, continuous estimate revisions). M1 only fires when MAX reports real money and Adjust reports ~nothing — a break, not a variance.
- “In Adjust but absent from MAX” is deliberately NOT flagged. It is a portfolio fact, not a defect: those apps use other mediation. MAX also keys iOS by bundle id while Adjust keys it by store id, so every iOS app would read as a false gap. An early build emitted five such warnings and all five were noise.
- An unconfigured buying account looks exactly like an app with no spend. If a title’s ROAS looks implausibly good, check A4 and the account list before believing it.
Related
Section titled “Related”t1k:marketing:analytics:crawldata— the crawler these credentials and this script live beside; produces the per-package Profit/ROAS report the audit assumes is right.- the studio-local quick-optimize skill — the daily ROAS triage. A1 findings invalidate its campaign-level ranking for the affected app until fixed.
rules/ad-network-mcp-write-safety.md— this skill is read-only; any fix to a tracking link is a change made by a human in the network’s own dashboard.
Status
Section titled “Status”Built 2026-09-15 from a live investigation. The runnable script is
tool/audit-wiring.mjs in The1Studio/marketing-crawl-ads-spend; this skill is
the operator playbook. Deepen it via /t1k:sync-back.