Kertújítók · Google Ads only

Verification and Constraints

READY FOR REVIEW

Set up the agreed campaigns properly. Prove the settings, spending control and actual serving.

Current item: review this contract25 September 2026Executor not launched

Goal and finish line

End goal: Set up and verify all six agreed Google Search services so the right work is advertised within the 200,000 Ft monthly limit.

Done means: The three priority services receive the main emphasis, turf and standalone irrigation retain smaller budgets, and both maintenance offers are distinguished. Ads, keywords, areas, destinations and bidding fit the meeting. Real impressions prove serving. Unproved measurement, payment or review remains explicitly pending or blocked.

6agreed services

200,000 Ftmaximum per month

0 / 9final criteria verified

Rules that stay fixed

Google only. All six services remain in scope. No Meta or proposal/calendar/CRM build.
Maximum 200,000 Ft per calendar month across the whole Ads account, including spend already incurred. Péter handles payment.
Maintenance: 10 km around the documented Szigetszentmiklós targeting center, presence targeting. Keep its source and exact coordinates clear, without inventing a home address. Preserve justified areas for other services.
Reuse current campaigns, KPC analysis, exclusions and working landing pages. Make only campaign-critical landing/tracking corrections. Cloudflare only, no GAJA/GHL.

The agreed campaign mix

Service	Emphasis	What we want to attract
Talajfúrás	Priority	Actual auger-drilling work, not generic earthmoving or well drilling
Betontörés	Priority	Concrete breaking and supported demolition work
Térkövezés	Priority	Paving installation
Gyepszőnyegezés	Smaller budget	Keep autumn turf enquiries possible
Öntözőrendszerek	Smaller budget	Standalone irrigation enquiries, without complete-garden projects
Kertfenntartás	Included	Separate short local mowing/leaf jobs from larger machine-supported maintenance

Current starting point: Fresh API read, 25 September 2026 at 08:08 Budapest: five campaigns enabled but marked Limited, maintenance paused. September-to-date returned zero impressions. These are starting facts, not a diagnosis or proof of finished setup.

Source: September 9 meeting, Matt’s later Google-only decisions, saved KPC analysis and a fresh account read. Existing maintenance landing page ↗

Three deliverables. Clear proof.

Expand to inspect the tests

Align campaigns with the meeting

All six services covered by appropriate keywords, ads, locations and working destinations.

0 / 3

How we verify it is done

Observable tests

When the final campaign/ad-group tables are compared with the meeting, then a six-row checklist maps every service to matching keywords, truthful Hungarian ads, assets and a working destination, with short and larger maintenance offers distinguished.Proof: Before/after Ads readback and actual Ads/landing screenshots. Reuse the justified structure, without forcing six campaigns. Map each maintenance offer separately, or record the evidence and reason for any unsupported offer. Irrigation excludes complete-garden intent.

D1-1 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When location settings are read for all six services, then maintenance shows the agreed 10 km radius and presence mode, and the six-row location matrix records each service’s center/areas, exclusions and network settings.Proof: Current readback and screenshot, with the maintenance center’s source and coordinates. Label a selected town center honestly. The radius is a targeting setting, not a physical-location guarantee.

D1-2 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When current keywords and search terms are checked against the meeting offer and KPC findings, then unsupported demand is excluded and valid commercial demand remains covered in the final keyword/negative-keyword inventory.Proof: Dated search-term coverage, source-backed retained/changed decisions, before/after keyword export and actual Ads screenshots. State whether terms are complete or sampled. Preserve useful exclusions and commercial price/installation demand.

D1-3 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

Protect

Do not reduce the six-service brief to only the three priority services, or apply the maintenance radius to every service.
Preserve useful history and exclusions. No unverified capabilities, alpine tree felling, invented prices or guarantees. Do not blend the son’s short jobs with larger maintenance promises.

Control spending and conversion goals

A reconciled monthly budget and campaign goals that reflect verified enquiry signals.

0 / 3

How we verify it is done

Observable tests

When all enabled campaigns, distinct/shared budgets and month-to-date cost are read before and after each enable or budget edit, then the worksheet proves total monthly exposure stays within 200,000 Ft, with the agreed service weighting.Proof: Dated account-wide readback, budget screenshot and conservative remaining-month arithmetic. Include incurred/reported cost, reporting lag, highest change-day budgets and remaining days. Use billed cost where available without treating credits as extra headroom. A daily-total × 30.4 estimate alone is insufficient.

D2-1 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When the spend-control calculation is tested at the remaining-budget limit, then it requires reduction or pause before the worst-case exposure exceeds the allowance, and records the responsible owner, next check and safe next-month settings.Proof: Calculation and near-limit test, actual budget/status readback, pause threshold and residual exposure through the next check. Prefer conservative provider spending bounds over a new automation build. Daily budgets are not a hard account-wide cap. Keep affected spend paused if the bound cannot be proven.

D2-2 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When a labelled controlled submission is accepted through the live form, then its saved record and matching event are linked in the evidence packet, and each campaign’s goal and bidding strategy are read back against that verified signal.Proof: Test ID across form/storage/event, saved Ads goals and actual diagnostics. Exclude test data from real-lead reporting and do not invent paid attribution. Until verified, explicitly use justified Maximize Clicks/Manual CPC with any cap recorded, or pause. The fallback does not pass measurement. Record Ads processing delays separately.

D2-3 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

Protect

Never raise the monthly ceiling or change Péter’s payment card/funding. Include other enabled campaigns and prior spend.
No fabricated business enquiries, paid ad clicks or attribution. The Ads executor owns narrow campaign-critical Cloudflare/tracking fixes after checking current ownership. Broader dependencies get a named owner and remain blocked. Preserve existing consent and form protections.

Prove launch and record remaining blockers

One six-service result table showing configured, eligible, serving and unresolved states separately.

0 / 3

How we verify it is done

Observable tests

When post-change Google Ads diagnostics are opened, then the six-service table records working URLs, exact policy/status reasons, schedules and billing state, with an evidenced cause or an explicitly unknown cause for each restriction.Proof: Fresh readback and actual diagnostics screenshots. Record edit/appeal time and next review check, normally after one business day and again after two full business days if still pending. Appeal submitted, generic Limited status and Eligible alone never prove serving.

D3-1 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When performance is read at the predeclared observation deadline, then each service is marked SERVING only with real post-change impressions, BLOCKED with an evidenced restriction, or UNKNOWN when serving remains unproved, with counts and times in the result table.Proof: Before activation, fix Europe/Budapest start/end and read times, normally 24 and 48 hours after launch, covering scheduled serving time. Do not reset the window to select better results. Save metrics and actual performance screenshots. Zero impressions never pass. Pending policy/billing gets its own finite checkpoint, owner and next action.

D3-2 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

When the final handoff is opened, then it shows the six-service states, exact changes, budget control, measurement limits, safe rollback and any Péter-owned payment action, with links to verified evidence and dated next actions.Proof: One concise handoff and updated board. No client message is sent.

D3-3 · Evidence pending. Attach the provider/file receipt. For GUI tests: actual post-test screenshot + blind description + comparison. Status: UNVERIFIED.

Protect

Keep configured, approved, serving and verified leads distinct. Never guarantee enquiries or sales.
Preserve safe rollback without restoring excessive budgets. The observation deadline ends this check, not the requirement: any unproved serving or measurement remains unchecked and the whole goal is not complete.

Evidence rules and unresolved factsWhat can stop a pass

Every GUI test needs a screenshot of the actual changed interface after the stated test. An independent gpt-6-luna high reviewer receives the screenshot without the expected result and describes only what is visible. Attach screenshot, blind description and parent comparison to that criterion. Missing or mismatched proof leaves it unchecked. Provider exports stay private; publish only redacted summaries/screenshots.

Open, to verify: The cause of Limited/zero serving, current billing/policy state, lead-goal reliability and final allocation require fresh executor readbacks. Historical product-release notes are not current blockers without checking. Scope, Péter’s payment ownership and the 10 km choice are settled.

Assumption: existing campaigns and pages can be corrected without a broader rebuild. The executor checks this and reports any specific campaign dependency.

Zero checked criteria means this new execution has not run. It does not erase earlier work. No new report pack or product build is included.

Handoff and execution authorityReview first, then Launch

Meeting briefPrivate meeting transcript, 2026-09-09

Keep the promised services.

Setup analysisPrivate KPC + fresh account snapshot

Reuse facts and useful changes.

Verified campaigns

Show what changed and works.

Result + next action

Make remaining blockers explicit.

PREPARE only: no executor or account changes now. After Matt reviews this board and later sends Launch, one gpt-5.6-sol medium executor may implement and activate this Google-only scope within the cap. Before launch, read the latest prompt, board comments and inline rewrites, record the check including zero comments, and reconcile feedback. Material changes need review of the revised board. No client sends, changed commercial commitments, or more than USD15 shared vendor spend. New review/deployed surfaces must use bare publicly reachable URLs with no added security gate. Keep credentials and sensitive source data off them. Existing Google account access stays private.

PARENT · PREPARE25 September

Ads-only contract prepared for review. Account read was read-only. No campaign changes or executor launch.

Executor messages will appear here after Launch.

Private executor inputs: executor-source-map.md and the reviewed contract. On completion: deliverables.md, main-prompts.md, session.txt and a clean gist. Update this board and its proof counters on the same URL.