Changelog

September 6, 2026 - Visible investigation progress and calmer evidence

This release improves current-plan continuity and interface clarity. It creates no collection, Agent run, proposal, Change Set, approval, provider write, campaign change, launch, bid, budget, delivery, or spend action by itself.

• A fresh collection now records a new evidence check even when Oobi keeps the same current campaign decision, so an unchanged plan no longer looks abandoned or stale.
• Every current plan states why its investigation is still open, how many evidence checks Oobi has completed, and the exact conditions that will close it as an actionable Change Set, required creative, or a keep-unchanged decision.
• TikTok lifecycle values such as Enabled are no longer presented as delivery causes. When a newer response shows no delivery but only repeats lifecycle status, Oobi preserves an earlier provider-verified out-of-budget cause as historical evidence and rechecks it.
• Insights keeps its manager read first and folds sources, measurement, creative, performance, Signal Watch, learning, budget, and keyword safety into one supporting-evidence disclosure instead of many competing drawers.

September 6, 2026 - Stricter plan truth and automatic evidence-gap routing

These changes improve plan truth, evidence continuity, and engineering handoff. They create no Change Set, approval, provider write, campaign change, launch, bid, budget, delivery, or spend action by themselves.

• An extreme mismatch between provider clicks and attributed website sessions now opens a controlled measurement-reconciliation diagnosis instead of falling through to a misleading healthy-monitoring plan, even when shallow website activity exists.
• When a currently non-delivering TikTok response omits its secondary reason, Oobi can retain an earlier provider-verified budget-exhausted cause as historical evidence and explicitly recheck it instead of replacing it with an unsupported unknown-cause claim.
• Agent Evidence v3 now requires a materiality classification for every evidence family Oobi does not collect. Blocking gaps automatically enter the deduplicated Capability Factory; non-blocking boundaries do not create engineering noise.
• Connected-agent instructions now distinguish the OobiMax product, Agent Gateway 3.0 connection, and Agent Evidence v3 context so a protocol version is never presented as the product version.

September 6, 2026 - Business outcomes and richer campaign-agent evidence

No provider credential, user-level record, provider access, attribution claim, causal claim, Change Set, approval, campaign change, launch, bid, budget, delivery, or spend authority is created by this release.

• Workspace owners can separately allow a connected customer agent to report source-labelled aggregate qualified leads, booked calls, purchases, subscriptions, revenue, or refunds for campaigns already inside its Oobi scope.
• Oobi rejects people, contacts, customer/order/click/session identifiers, credentials, raw rows, and sensitive URL queries. Existing active connections can add the permission after a new consent confirmation without changing their token.
• Exact-campaign business evidence can regenerate Oobi's deterministic current plan and reveal a business-system-versus-ad-provider reporting gap. Brand aggregates remain context only, low-confidence submissions wait for confirmation, and return math requires matching campaign, period, source, and currency.
• Authorized customer agents and the managed Oobi Campaign Agent can now use Google/Meta geography plus aggregate Google/Meta/TikTok video-retention milestones when supported collection returns them. Empty supported evidence is not mislabeled as a product gap.
• TikTok geography/device and unconnected commerce, CRM, YouTube Studio, Etsy, and organic-channel systems remain explicit source boundaries until a verified connector or authorized aggregate submission exists.

September 5, 2026 - Google geography and RSA evidence become Oobi-owned work

No campaign, location, ad, creative, approval, provider write, bid, budget, delivery, or spend state changes from this release by itself.

• Current Google campaign plans now promote a material concentration of physical user-location clicks outside the campaign's collected target countries into one Oobi-owned targeting-quality investigation.
• Google campaign collection now records the exact positive and negative location-target modes so Oobi can distinguish configured reach from observed physical delivery without guessing campaign intent.
• Current Search plans now promote enabled responsive search ads rated Poor into a controlled asset-quality investigation and name exact stronger siblings when a like-for-like comparison exists.
• Location and ad-strength signals remain diagnosis, not automatic edits. Oobi checks exact targets, market intent, outcomes, sibling ads, assets, pinning, search terms, and creative history before it can prepare one reviewed change.
• Customer-agent geography and RSA-quality proposals now resolve to these native Oobi investigations instead of opening duplicate engineering gaps.

September 5, 2026 - Deeper customer-agent evidence without provider access

Agent Access remains exact-tenant and exact-campaign scoped. It exposes no password, token, cookie, private key, provider credential, raw provider payload, click identifier, user-level event, Change Set authority, approval, provider write, launch, or spend authority.

• Connected customer agents now start one complete Oobi analysis cycle that lists every current campaign and the deeper evidence Oobi can provide for each one.
• An authorized agent can fetch bounded campaign performance, Google search terms and keywords, structure and targeting, ad and creative evidence/history, Meta placements, Google geography, Google and Meta device evidence, measurement paths, learning outcomes, and cited public context where that evidence applies.
• Every evidence manifest distinguishes available data, an empty or inapplicable result, and evidence Oobi does not yet collect. An agent must fetch available evidence before claiming Oobi has a product gap.
• A suggested value attached to a read-only investigation is treated only as an untrusted hypothesis: Oobi discards it, owns the evidence recheck, and creates no Change Set. Mutating requests and unsupported actions still fail closed.
• The Capability Factory now separates missing read-only evidence collection from missing governed execution support, and gives each category the correct engineering acceptance contract.
• Campaign Plans and exact plan detail state whether Oobi routed an agent idea to Change Review, an Oobi-owned investigation, evidence engineering, execution engineering, customer creative input, or rejection.

September 5, 2026 - Complete customer-agent cycles and reliable capability work

This release changes analysis recording, adjudication routing, and interface truth only. It creates no Agent Access token, collection, Change Set, approval, provider write, campaign change, message, bid, budget, creative, delivery, or spend action by itself.

• A connected customer agent now submits one durable conclusion for every current campaign context it reads. Agreement or no additional action is recorded with an empty action list instead of leaving a silent blank.
• Campaign Plans shows current conclusion coverage and clearly marks any plan the customer agent did not conclude after the latest refresh, while keeping Oobi's own plan complete and authoritative.
• Read-only campaign measurement, qualified-lead outcome, campaign-goal, and optimization-goal diagnoses now open the same dated Oobi-owned investigation across Google, Meta, and TikTok when their exact target, operation, and evidence contract passes; support no longer depends on a coincidental current-plan action.
• Earlier recorded diagnostic gaps now show their current covered state without rewriting their historical audit. Mutating operations or values, wrong targets, and unknown evidence still fail closed or remain genuine engineering gaps.
• The Capability Factory no longer stalls indefinitely: its generated JavaScript is validated, loading has a bounded timeout and retry state, and earlier requests now handled by Oobi are separated from real engineering gaps.

September 5, 2026 - Visible customer-agent proposals and clear engineering handoff

This release improves visibility and ownership only. Customer agents still receive no provider credential, provider API access, Change Set, approval, apply, launch, budget, or spend authority, and copying an engineering brief starts no code or deployment.

• Campaign Plans now shows recent customer-agent proposals, the exact campaign each belongs to, Oobi's verdict, what happens next, and a direct route into the current campaign plan.
• Each exact plan keeps the latest customer-agent proposal visible in the first-read workflow even after a newer managed Ask Oobi Agent run, so one analysis can no longer hide another.
• Capability gaps now tell platform administrators which Oobi product capability is missing, how much validated request evidence supports it, and the next engineering step before showing lifecycle or technical acceptance detail.
• The Capability Factory copies a customer-anonymous, secret-free engineering brief. Customers are not sent to perform unsupported work manually in an ad platform.

September 5, 2026 - Connected-agent operating loop and reliable refresh

This release clarifies the existing governed Agent workflow and repairs refresh coordination. It creates no Agent Access token, schedule change, campaign change, approval, provider write, message, bid, budget, creative, delivery, or spend action by itself.

• Agent Access now gives Codex, Claude, and compatible agents the exact Streamable HTTP setup, a secret-safe starter instruction, and one clear operating loop: Refresh Oobi first, read current Oobi context, then submit proposals back to Oobi.
• Connected agents use Oobi as the source of ad-account and campaign facts. They may add cited public research as a hypothesis, but they receive no provider credentials, provider API access, approval, apply, launch, pause, bid, budget, or spend authority.
• Supported proposals continue through Oobi's review, approval, execution, read-back, and learning controls. Credible unsupported proposals are recorded by Capability Factory as deduplicated, secret-free engineering gaps instead of being discarded.
• Refresh Intelligence now claims each queued account job atomically across web processes, preventing duplicate workers from falsely stopping or overwriting another worker's completed account refresh.
• A TikTok Smart+ limitation that affects only optional detailed ad-review reasons now appears as a scoped evidence note. Fresh campaign and delivery evidence remains usable, while policy-dependent actions stay safely blocked without turning the entire workspace into a false action item.

September 4, 2026 - One Insights workspace and exact-time automation

This release simplifies existing workflows and scheduling. It creates no schedule change, Agent Access key, campaign change, approval, provider write, message, bid, budget, creative, delivery, or spend action by itself.

• Insights is now the one customer-facing place for current exact-campaign priorities, compact approved-change outcomes, pending rechecks, and Weekly Brand Performance Briefs. Earlier Results links open the same page.
• Campaign Builder-created campaigns now appear once in the normal Campaigns inventory with a small origin badge instead of a separate duplicated campaign lane.
• The exact plan now states whether Ask Oobi Agent added material evidence or only confirmed Oobi's existing decision, and it names whether 10 or zero credits were used.
• Settings now exposes Connect Codex or Claude through Agent Access & Security. Customer agents receive only explicitly authorized Oobi intelligence and proposal tools, never ad-platform credentials or execution authority.
• Automatic and hybrid workspaces can choose an exact local daily Refresh Intelligence time and timezone, while Signal Watch keeps its separate interval.

September 3, 2026 - Oobi-owned campaign follow-up

This release changes decision ownership, truth, and presentation only. It creates no campaign change, approval, provider write, message, bid, budget, creative, delivery, or spend action.

• Every current campaign plan now leads with four answers: Decision, Evidence, What Oobi does next, and When Oobi reports back.
• Diagnosis-only plans are explicit Oobi investigations with opened, last-checked, and next-report times instead of customer homework in an ad platform.
• The current Campaign Optimization Plan is authoritative in Insights. Older diagnosis and Signal Watch observations for that exact campaign no longer create duplicate or contradictory priorities.
• Campaign metrics use the plan's real goal-specific result unit and cost basis. Zero results show no cost per result instead of a misleading zero CPA.
• Ask Oobi Agent is clearly optional paid synthesis. Refresh already analyzes every active campaign; a completed Agent run updates only the current plan and cannot approve or apply a change.

September 3, 2026 - Clear Results and focused Insights

This release changes evidence truth and presentation only. It creates no campaign change, approval, provider write, message, bid, budget, creative, delivery, or spend action.

• Results now shows only successfully applied changes, measured outcomes, pending rechecks, and genuine execution issues; campaign diagnosis and old recommendation history no longer compete with outcome proof.
• The account-performance snapshot stays folded, while detailed platform, goal, campaign, freshness, and return evidence remains available through the tenant-scoped CSV export.
• Insights now opens with at most four exact-campaign priorities. A complete Oobi diagnosis wins over a generic Signal Watch observation for the same campaign, preventing duplicate or conflicting cards.
• Source, creative, performance, Signal Watch, learning, and budget detail remains available behind deliberate foldouts instead of filling the default screen.
• Fresh collection state is now evaluated separately from delivery activity, so a current paused or non-delivering campaign is not mislabeled stale only because it has no recent metric date.

September 3, 2026 - Exact Google source boundaries

No campaign, mapping, provider account, approval, budget, bid, creative, delivery, message, or spend state is changed by this correction.

• A direct Google Ads source now shows and maps only the exact account verified for that source, even when the same Google login can open other unrelated accounts.
• An MCC source now discovers only reachable child accounts inside that configured manager hierarchy; unrelated top-level accounts no longer appear beneath it.
• Settings labels direct accounts and MCC sources explicitly, uses readable OAuth wording, and distinguishes connected accounts from MCC child accounts.
• Existing source discovery is corrected safely on read, while campaign collection remains limited to the workspace's explicit brand-to-account mappings.

September 1, 2026 - Clear plan states and a quieter workspace

These changes simplify the existing workflow and strengthen evidence truth. They create no campaign change, approval, provider write, message, bid, budget, delivery, or spend action.

• Every current campaign plan now has one shared state across Home, Campaign Plans, Campaign Editor, and Results: Ready to apply in Oobi, Needs your creative, Diagnosis only, Monitoring, or Keep unchanged.
• Campaign Editor leads with Oobi's decision, what happened, what happens now, what Oobi will check next, and exact current-to-proposed values before supporting evidence.
• Campaign inventory replaces the primary numeric score with Oobi Decision, keeps raw IDs and provider signals folded, and uses a simpler campaign, delivery, action, type, results, spend, and CPA comparison.
• Primary navigation now keeps Home, Campaign Plans, Campaigns, and Results visible while deeper evidence, building, settings, and administration remain available under More.
• Scheduled connector failures preserve the exact safe recovery action and never claim an account was processed when no fresh evidence was collected.

September 1, 2026 - Oobi-owned diagnosis and truthful Campaign Builder evidence

These changes improve the existing plan and builder flows. They add no agent connection, autonomous approval, campaign launch, provider mutation, or spend authority.

• A post-click Oobi diagnosis now stores the exact campaign click/outcome gap, collected destination inventory, optimization event, and separately scoped Measurement context instead of sending the operator to an ad platform to investigate.
• Plan spend uses the mapped account currency, optional deep-AI wording fails closed if it invents another currency, and analysis times render from explicit UTC in the operator's local time.
• Learning says it was applied only when recorded evidence actually influenced a current action; otherwise the plan says history is available or the loop is ready.
• Campaign-specific public context now rejects weak one-token category collisions as well as sports and viewing noise.
• Campaign Builder now recognizes the same tenant-owned brand across spaces, hyphens, and underscores, and its score ledger includes the visible starting value and cap state.

August 30, 2026 - Exact changes before any campaign draft

Preparing a draft still changes nothing in Google or Meta. Change Review, explicit approval, live pre-read, provider apply, fresh read-back, and outcome learning remain separate.

• An executable plan action now shows the exact campaign object plus the literal current and proposed values before the operator prepares anything.
• When one supported action is available, the plan uses one clear Prepare draft for review command instead of a checkbox and a second competing button.
• Evidence, public context, learning detail, and the general Campaign Editor stay available but folded behind the recommendation so the decision arrives first.
• Search copy plans now reject proposals that repeat an intent already covered by another unchanged responsive-search-ad asset. Earlier saved proposals that fail the current rule are blocked before Change Set compilation.
• Campaign-level public research keeps sports, schedules, streaming, and other watch-word collisions outside the exact campaign decision.

August 30, 2026 - A calmer Home and prioritized Campaign Plans

This release changes navigation and prioritization only. It creates no campaign change, Change Set, approval, provider write, budget, bid, delivery, message, or spend action.

• Home no longer promotes an arbitrary campaign plan. It now opens Campaign Plans and summarizes current, executable, diagnosis/review, and keep-unchanged plan authority.
• Completed setup progress, the onboarding checklist, and the Beta Guide shortcut leave the daily Home surface once the workspace is ready; incomplete workspaces keep the exact setup path.
• Campaign Plans now places executable work first and labels diagnosis/review and keep-unchanged plans truthfully instead of calling every card generically ready.
• Benign refresh results use plain intelligence-status language instead of exposing aggregate snapshot counts as a primary success message.

August 30, 2026 - Truthful deep analysis and executable Search copy

Selected RSA actions still become a draft Change Set first. Oobi re-reads the live ad, validates the full copy field, and requires Change Review and explicit approval before any Google Ads write.

• Campaign Plans now derives current work from the latest collected campaign inventory. A paused or inactive campaign moves to history even when an older saved plan snapshot still says it was delivering.
• Campaign inventory now says Open current plan because that link opens the automatically maintained plan; it does not pretend to run a new analysis.
• Run fresh deep AI analysis is the separate explicit Sol action. A completed run shows its completion time, elapsed duration, and model.
• The 10-credit analysis charge is captured only when Sol completes. If the managed synthesis fails or is unavailable, Oobi keeps the deterministic plan, releases the reservation, and changes no campaign.
• Exact Google Search RSA recommendations can now become selectable plan actions when Oobi has the exact synced ad, complete current copy, and a complete platform-valid proposed field.

August 29, 2026 - Current plans learn from fresh signals

Supported non-creative work stays inside Oobi through Change Set review, approval, apply, and outcome checking. Supplying replacement media, completing a provider appeal, or restoring access remain explicit human exceptions. Automatic planning changes no campaign, budget, delivery, message, or spend state.

• Refresh Intelligence now maintains one current Optimization Plan for every active campaign. Unchanged evidence reuses the same-engine plan instead of creating noise, while a plan-engine upgrade refreshes it once so improved logic reaches every active campaign.
• The four-hour Signal Watch performs a bounded campaign mini-collection before it updates Measurement, outcome learning, movement evidence, and current plans.
• Brand-relevant Google News RSS and Google Trends RSS can add cited, timestamped public context for timing, demand, seasonality, topics, and positioning. Public context never becomes campaign attribution or approval authority.
• Oobi's private learning loop uses recorded operator decisions, applied changes, fresh read-back, measured outcomes, and earlier plan history to adjust priority without exporting private decision history.
• Deep AI uses the managed gpt-5.6-sol model only for material automatic cases or an explicit 10-credit analysis request. Stable routine coverage remains deterministic.

August 29, 2026 - Campaign plans now act on material declines

This diagnosis is Assisted, not an automatic creative, bid, target, budget, delivery, or spend change. Current ad inventory is not presented as proof of historical creative causation.

• A material, decision-ready decline in the exact campaign's goal result now creates a prioritized diagnosis instead of appearing only in summary prose or a generic Watch item.
• Demand Gen and Video plans use exact-account daily ad and ad-group evidence when available, and Campaign Editor uses the same source for the same date window.
• The plan localizes the largest collected loss, explains what the evidence cannot prove, and gives one controlled next step. Partial or stale windows do not create a trend verdict.
• After a Demand Gen video refresh, OobiMax advises holding target CPA and budget stable and normally allowing at least 14 days for a new baseline unless delivery or policy breaks.

August 29, 2026 - Exact plan windows and TikTok campaign analysis

No campaign, budget, delivery, message, or spend state changes from opening or reading these plans. Automatic current plans are included in Refresh; an explicit deep Sol synthesis remains a 10-credit analysis action.

• Campaign Optimization Plans now use the same calendar-day evidence window as Campaigns and the exact-campaign view. Sparse provider rows stay visible as stored-day coverage instead of silently extending the date range.
• Saved plans created under the earlier observed-day rule now identify that legacy evidence basis, so operators do not mistake it for a current calendar-window comparison.
• TikTok campaign links now open the exact campaign plan surface. TikTok plans remain analysis only: actions are review-only and no Change Set or TikTok provider write is available.

August 24, 2026 - One complete plan for one exact campaign

Automatic current Campaign Optimization Plans are included in Refresh Intelligence. Reading a saved plan is free; a deliberate deep Sol synthesis costs 10 Oobi Credits and never changes an ad platform.

• Choose one exact Google, Meta, or TikTok campaign and generate one complete Campaign Optimization Plan from current campaign evidence and relevant brand context.
• Each plan separates Do now, Test next, Watch, and Keep unchanged, and explains proof, uncertainty, expected effect, risk, and the next review point.
• A healthy campaign may correctly receive a complete no-change plan. OobiMax does not invent work just to fill a queue.
• Only selected executable actions with exact current and proposed values can enter a draft Change Set. Review, approval, apply, and fresh-data outcome checking remain separate.
• Earlier recommendation records and Slack or Telegram cards remain folded compatibility history during migration; they are not a second current decision queue.

August 10, 2026 - A clearer public product story

The redesign changes public explanation and navigation only. It adds no campaign, budget, Pixel, recommendation-send, purchase, or spend authority.

• The homepage now explains OobiMax through one short operating story: connect evidence, prioritize the next decision, approve the supported move, and recheck the outcome.
• The public homepage now uses a real OobiMax product view instead of a simulated dashboard scene.
• Google Ads, Meta Ads, and TikTok Ads appear as the current paid-media platforms. Measurement remains provider-specific evidence and is never presented as a universal cross-platform conversion or return total.
• Start free now routes directly to workspace creation, while Security, Status, beta intake, pricing, and product documentation use the same current platform truth.

August 9, 2026 - Simpler TikTok setup and Pixel evidence

No account change is required. Continue using the mapped advertiser and run Refresh Intelligence normally; campaign writes remain disabled unless the separate execution gate is explicitly accepted.

• TikTok setup now follows the same simple path as Google and Meta: connect TikTok, then map the selected brand to one exact advertiser account.
• TikTok now uses the same visible three-stage setup as Google and Meta: connect one source, review the saved source, and map the exact advertiser to the brand. Provider access containers remain internal.
• Refresh Intelligence can now read customer-owned TikTok Pixel discovery and aggregate event-health evidence for the mapped advertiser when TikTok makes those approved reads available.
• TikTok Pixel evidence stays brand-scoped and provider-specific. OobiMax stores no user-level events, performs no Pixel writes, and does not blend TikTok events with Meta or Google totals.
• Unavailable Pixel evidence does not invalidate otherwise successful TikTok campaign collection, and an advertiser with no Pixel is not automatically treated as a campaign problem.

August 5, 2026 - Safer Google workspace sign-in

Existing users do not need to register again. Use Google sign-in; enter the Workspace Key only when the same email belongs to more than one OobiMax workspace.

• Google sign-in now checks the verified Google identity against existing OobiMax workspace users before any new beta workspace is created.
• When the email belongs to one workspace, OobiMax reuses that account. If the workspace is still pending, the existing request is shown without creating or notifying a duplicate.
• When the same operator belongs to multiple workspaces, OobiMax asks for the Workspace Key. Inactive users and conflicting Google identities remain blocked instead of being silently relinked.
• A new workspace request is created only from the explicit Create Workspace flow after the required consent.

August 3, 2026 - Decision integrity and workflow reliability

This dated entry describes the retired recommendation workflow. Current campaign decisions start from one exact Campaign Optimization Plan and supported actions move through a Change Set. If Settings -> Measurement shows Access needed for a brand, reconnect that brand's read-only Google Measurement source and run Refresh Intelligence. Review Weekly Brief recipients before enabling or changing scheduled delivery.

• Home now prioritizes current work across brands through the Agency Command Queue. Failed-action recovery, verified policy or delivery issues, active-campaign evidence gaps, and concrete Inbox recommendations appear ahead of inactive or housekeeping work.
• Optimization Readiness now separates the interpreted business goal, the exact platform optimization event, and the bid strategy. Inactive, stale, draft, removed, and no-delivery campaigns remain Not assessed instead of becoming false warnings.
• Accepted, rejected, reviewed, skipped, and dismissed recommendations remain immutable decision history. If materially different evidence appears later, OobiMax creates a new recommendation identity and REC # instead of silently reopening the handled item.
• Google campaign conversion-goal evidence is now shown separately from bidding strategy. Connected GA4, Search Console, Pixel, and CAPI evidence remains brand context unless an exact campaign-level relationship is proven.
• Results and Insights no longer describe zero-result spend as a winner, combine provider-specific result units, or count queue-only decisions as measured learning.
• Inbox, Slack, and Telegram recovery now separates retryable delivery failures from historical messages that can no longer be updated, keeping the current decision state aligned without replaying handled actions.
• Google Measurement connections are isolated by brand, shared agency sources remain supported, and Settings returns cleanly after read-only Measurement consent instead of entering a redirect loop.
• Refresh Intelligence no longer shows a false failure when a successful completion summary mentions archived failed actions. Weekly Brief delivery can recover an eligible missed Monday send, prevents duplicate weekly sends, and distinguishes configuration readiness from verified delivery.

August 2, 2026 - Decision intelligence and evidence boundaries

This dated entry describes the retired recommendation workflow. Current campaign decisions start from one exact Campaign Optimization Plan; approve only supported actions through a reviewed Change Set.

• Results now turns current campaign, recommendation, measurement, policy, and weekly movement evidence into a client-ready decision brief with the next action, owner, expected effect, risk, unknowns, recheck window, and canonical recommendation ID when one exists.
• Google Ads, Meta Ads, GA4, Search Console, Pixel, CAPI, CRM, and offline observations keep their own scope and attribution. OobiMax highlights disagreement but does not blend provider outcomes into a fake conversion, value, CPA, or ROAS total.
• Campaign diagnosis now checks current delivery, goal-event alignment, configuration, post-click funnel evidence, and like-for-like peers before suggesting a bounded creative test.
• Google creative intelligence now includes Responsive Search Ad inventory and ad-strength evidence plus exact YouTube and Demand Gen ad-level delivery where collected. Search inventory does not claim ad-level performance, and AI provenance remains unknown unless the provider exposes it explicitly.
• Google, Meta, and TikTok are the live paid-media intelligence providers. Every future provider must pass the same tenant isolation, freshness, integrity, attribution, recommendation, approval, audit, delivery, and compatibility gates before production registration.

July 28, 2026 - Platform policy and delivery integrity

No action is required unless an integrity item appears in a current Campaign Optimization Plan. Oobi re-reads the exact provider issue and prepares any supported non-creative repair through a Change Set. If Google or Meta requires replacement media or a justified provider appeal, complete that named exception, then run Refresh Intelligence so Oobi can verify it. Refresh maintains current plans but never approves or changes a campaign.

• Refresh Intelligence now checks supported Google and Meta policy and delivery evidence as dedicated integrity work instead of treating it as ordinary creative or performance optimization.
• Meta checks current ad-account restriction state and ad-level eligibility/problem evidence across active and paused inventory, so a blocked paused ad can be found before anyone resumes it.
• A current finding becomes a scoped Action Plan that names the affected account and platform objects. It does not automatically appeal, edit, enable, launch, or change spend.
• Inbox, Slack, and Telegram show the exact collected account, campaign, ad-set, and ad identifiers plus configured/effective status and policy evidence, while the Refresh log reports Meta policy integrity separately from ordinary Meta optimization checks.
• An existing incident stays current if its exact check fails or becomes stale. OobiMax retires it only after that scope refreshes successfully and the platform no longer returns the issue.
• Every future production ad-platform connection must ship the same dedicated integrity and recovery contract before it can join the live platform set.

July 28, 2026 - Broader recommendation review

Existing pending keyword cards are revalidated immediately. Refresh Intelligence now formally retires older unsupported positive and negative keyword cards while preserving legitimate Brand Profile-supported demand, and nothing is sent to Slack or Telegram unless you explicitly send it.

• Refresh Intelligence now explains which supported optimization dimensions were checked and shows the recommendation-family mix that reached the current action-ready queue.
• When you request a Campaign Optimization Plan, Oobi compares the strongest current evidence across the campaign goal, measurement, delivery health, budget and bidding, structure, targeting, creative and assets, search intent, destinations, and conversion-path integrity.
• A concentrated queue now says plainly that it is not a health certificate for every other campaign field. OobiMax keeps evidence and safety thresholds intact instead of inventing recommendations to create artificial variety.
• Legitimate Brand Profile-supported search demand remains eligible when it is the strongest concrete next action.
• Positive-keyword actions recheck retailer, marketplace, named third-party watch-brand, offer-scope, and broad category-only demand against the current Brand Profile before they can enter a Campaign Optimization Plan or an approved Change Set.
• Supported manufacturer and product-family demand such as Seiko remains eligible when the Brand Profile explicitly supports that offer; unrelated competitor demand and unqualified single-word watch queries stay out of the action queue.

July 27, 2026 - Clearer decisions, safer recommendations

No action is required. If Settings -> Measurement shows Access needed, reconnect Google Measurement and run Refresh Intelligence.

• Campaign Builder now recommends one evidence-backed starting strategy from the selected Brand Profile, current account and campaign context, saved plans, and approved creative evidence. It warns about overlapping plans, routes existing campaign changes to Campaign Editor, and keeps supported creation paused until a separate launch or spend approval.
• Results now leads with executed changes, measured outcomes, and open follow-up checks. Insights ranks risks, opportunities, evidence gaps, and watch items, then routes each investigation to the workspace that owns the next step.
• Google Search negative-keyword actions require explicit Brand Profile support. Retailer or marketplace expansion is blocked unless the Brand Profile clearly authorizes that relationship.
• Meta analysis now uses current review-window evidence and corrected conversion, conversion-value, CPA, and ROAS calculations. When no recommendation is appropriate, OobiMax explains whether evidence is missing or the campaign remains inside its safeguards.
• Recommendations tied to removed campaigns or ad groups no longer return as actionable work after Refresh Intelligence.
• Failed approved actions now direct operators to Retry for temporary failures, Reconnect for access problems, or Refresh Intelligence when the saved target is stale or no longer safe.
• Partial Refresh Intelligence runs preserve useful collected data, leave the running state correctly, and identify the connection that needs attention.
• Inbox, Slack, and Telegram reconcile against the same current recommendation state, reducing stale or conflicting external actions.

July 2026 reliability and workspace control

• Removed accounts and sources now leave the active refresh scope cleanly, while remaining valid connections continue collecting.
• Partially completed refreshes now leave the live running state automatically and show the completed follow-up result without requiring a page reload.
• Incomplete source refreshes preserve useful collected data, identify the connection that needs attention, and do not send external recommendations or consume refresh credits.
• Dashboard, Slack, and Telegram now resolve recommendation actions against the same current decision state; stale external actions fail closed.
• Workspace permissions now protect source setup, brand mapping, measurement mapping, and user administration, including last-manager safeguards.

July 2026 decision quality

• The public product story now explains OobiMax through the real operator job: prioritize the next paid-media decision, inspect the exact proposed change, approve supported actions, preserve brand learning across account changes, and keep missing evidence explicit.
• Results now use business-goal-aware outcomes, account-native currency, and explicit missing-value states instead of inferred revenue or conversion claims.
• Insights and learning views now qualify conclusions by evidence coverage, sample size, platform scope, and freshness.
• Recommendation relevance guardrails use each brand's offer, audience, market, campaign state, and exclusions before seasonal or keyword ideas can surface.
• Keyword-overlap recommendations now exclude keywords that are already paused, so the Inbox does not ask operators to repeat a completed action.
• Historical ad accounts can contribute read-only learning while active recommendations and approved execution stay tied to the current mapped account.

June 2026 campaign and measurement

• Campaign Editor provides exact current-versus-proposed review for supported Google and Meta changes, with approval, audit history, apply, retry, and recheck states.
• Approved GA4, Search Console, and Meta event-health connections contribute aggregate read-only evidence without granting campaign or spend authority.
• Campaign Builder creates reviewed paused plans and paused campaigns only; existing campaign edits remain in Campaign Editor.
• Weekly Brief and Signal Watch summarize stored performance, measurement coverage, recommendation state, and verified outcomes.

Control model

Release notes describe customer-visible behavior. Internal implementation and unshipped work are intentionally excluded.

• OobiMax remains approval-first: recommendations do not change campaigns until an authorized user approves a supported action.
• Credits meter product work and never grant campaign, budget, launch, or spend authority.
• Every supported execution keeps a workspace-scoped audit trail and a recoverable failure state.