Annotation fixes

End state: the exact current active 21-campaign cohort and its pre-write location criteria, campaign-goal rows, conversion action identity, budgets, statuses and bidding state are captured for customer 4891907898.

Pass when: the account identity is read back; all 21 resource IDs are listed once; the intended phone action is verified as the existing enabled `tel_szam_katt`/phone-interaction action; and a before snapshot can be compared field-by-field with the after snapshot.

Must not: include paused legacy campaigns, infer identity from a browser tab, or start a write before the exact account and object IDs are verified.

End state: every active campaign has a positive `LOCATION` or `PROXIMITY` target that is traceable to verified project/service-area evidence. City campaigns use their verified city target; motorway campaigns use only a provider-supported, evidence-backed geographic scope. Generic keeps its existing verified Budakeszi 40 km target unless the same evidence proves a narrower correction.

Pass when: the post-write provider readback shows the expected geographic criterion for all 21 active campaigns, no unintended location rows exist, presence behavior remains explicit, campaign IDs and statuses are unchanged, and the mapping table is saved with criterion IDs and names.

Must not: invent radii, broaden service areas from campaign names, use unverified business locations, touch paused/legacy campaigns, or change budgets, bids, keywords, ads or schedules.

End state: for each active campaign, the biddable campaign-specific goal is the existing web `PHONE_CALL_LEAD / WEBSITE` phone-interaction goal backed by the verified `tel_szam_katt` action. Its event coverage is verified against the site’s `gfn-track-phone` phone-reveal buttons and floating `widget-phone` link. Other campaign-specific goal categories are not biddable for this target cohort unless the provider model requires an equivalent exact representation.

Pass when: all 21 campaign-goal rows read back with the web phone-interaction goal biddable; non-phone goal categories are read back as non-biddable for these campaigns; the existing action remains enabled and primary/included; and no duplicate conversion action is created.

Must not: create synthetic conversions, place a call, change account-level action status, publish GTM, edit website code, delete conversion actions, or claim that the interaction proves a connected call, lead, job or revenue.

End state: a post-write diff proves that only the requested location criteria and campaign-specific goal rows changed.

Pass when: campaign status, budgets, bidding, ads, keywords, network, devices, languages, URLs, account-level conversion actions, GTM and website hashes match the before-state, except for the explicitly authorized fields.

Must not: treat a successful API response as proof of a narrow diff, accept hidden account-wide changes, or leave any partial write unexplained.

End state: a concise explanation states which campaigns were fixed, what exact goal now drives them, which settings were intentionally untouched, and any campaign that remains blocked by missing geographic evidence.

Pass when: the report contains exact before/after IDs, timestamps, screenshots or scripts, provider receipts, a strict success matrix, and an honest statement of remaining UNKNOWNs. The project state and changelog record the result without claiming more than the readback proves.

Must not: call a partial mutation complete, promise a learning-phase exit date, or publish private Ads evidence.