From the catalog the team ships against.

Verified in the shared environment

1. Foundation · Product foundation

   Evidence-backed audit and activation foundation

   Run evidence-backed audits, generate reviewable fixes and commerce artifacts, and export decision-ready reports from one authenticated workspace.

   Outcomes

   • Four-pillar diagnostics and deterministic crawl evidence
   • Fix Engine, content optimization, and commerce artifact generation
   • Versioned reporting across JSON, XLSX, PPTX, and PDF

2. Foundation · Product foundation

   Tenant, provider, and sector configuration

   Manage authorized brands, questions, rival pools, schedules, and encrypted provider credentials within tenant boundaries.

   Outcomes

   • Tenant-scoped authentication, roles, quotas, and audit history
   • Owner-managed BYOK settings with encrypted storage and rotation
   • Editable banking and fashion rival defaults in Brand Management

3. Foundation · Measurement

   Versioned visibility and mention quality

   Inspect brand visibility, recommendation role, rank, sentiment, citations, and evidence with versioned metric lineage.

   Outcomes

   • Evidence-preserving mention classification
   • Explicitly unclassified outcomes when evidence is invalid or unavailable
   • Immutable snapshots that retain historical metric definitions

4. Phase 0 · Governance

   One truthful roadmap and backlog

   See a concise, current delivery view generated from the same catalog the product team uses to govern implementation.

   Outcomes

   • Clear separation between live, active, next, and later work
   • No customer exposure of security, cost, or operational evidence
   • Roadmap status that cannot silently drift from the repository

5. Phase 1 · Operating loop

   Durable operating loop and API convergence

   Carry treatment, approval, schedule, and measurement state across restarts, devices, and authenticated sessions.

   Outcomes

   • Durable treatment lifecycle from generation through measurement
   • A pipeline state with evidence and a clear next action
   • Governed schedule enable, disable, and replacement

6. Phase 1 · Cost governance

   Estimate and confirm provider spend

   Review the pinned models, expected calls, estimate, and hard cap before any provider-backed benchmark starts.

   Outcomes

   • Spend-free estimates before execution
   • Explicit owner confirmation tied to the exact run shape
   • Automatic rejection when pricing, model pins, or scope changes

7. Phase 2 · Benchmark

   Banking benchmark v2

   Benchmark Turkish retail, commercial, and corporate banking with an immutable, human-approved question set and editable active rival pools.

   Outcomes

   • Balanced coverage across all three banking scopes
   • Recommendation, research, and explicit product-comparison questions
   • Traceable source approval and cohort checksum

8. Phase 2 · Measurement

   Product comparison quality

   See which products or providers were compared, which factors drove the answer, and whether cited evidence supports the outcome.

   Outcomes

   • Compared entities, factors, outcome, citations, and confidence
   • Evidence excerpts and definition lineage
   • Explicitly unclassified results when evidence is invalid

Merged or being built

1. Phase 2 · Benchmark

   Brand tag and prompt coherence

   Keep Brand Management prompts aligned with each brand's subcategory tags, including participation vs conventional banking and clear-cut fashion gender lines.

   Outcomes

   • Conventional banks no longer carry participation-presupposing saved prompts
   • New banking workspaces default to the provenance-corrected v3 cohort
   • CI blocks catalog drift that reintroduces tag/prompt mismatches

2. Phase 2 · Reporting

   Benchmark exploration filters

   Explore visibility and comparison results by scope, topic, intent, provider, and active rival pool.

   Outcomes

   • Consistent filters across dashboard and report views
   • Target scoring limited to the editable active rival pool

3. Phase 3 · Product foundation

   Analysis workspace and Growth Cockpit home

   Open Growth Cockpit as home, start System/Custom/Data Feeding analyses from the sidebar, save and reopen them with a horizontal visibility pipeline, and filter cockpit results by time, category, and analysis type.

   Outcomes

   • Growth Cockpit is the default home with working time-frame and category filters
   • New Analysis expands into System Prompts, Custom Prompts, and Data Feeding
   • Saved Analyses lists status, dates, and load/rename/archive/delete actions
   • Open analyses show Measure → Diagnose → Improve → Approve and publish → Re-measure with save and autosave

4. Phase 3 · Product foundation

   Unified Custom and Automated audit report

   Review one middle-way Diagnose report for Custom Audit and Automated Audit, then work issues and guides in Improve.

   Outcomes

   • Shared pillar scores, evidence explorer, and narrative tabs across audit types
   • Platform-styled Diagnose report instead of divergent Custom Audit chrome
   • Actionable issues, action plan, and shopping guides live in the Improve step

5. Phase 3 · Product foundation

   Reachable content, commerce, and governance surfaces

   Open the content, commerce-feed, and governance tools from the Improve and Approve steps of a saved analysis instead of having no route that reaches them.

   Outcomes

   • Content Optimizer and Commerce Protocols open in Improve, working from the run the step already loaded
   • Governance and Monitoring opens in Approve and publish, beside the decision it governs
   • Tools that need product pages say so rather than failing when a run stored none

6. Phase 3 · Measurement

   Rivals observed in comparison answers

   See which brands the models compared you against that you are not yet tracking, with the evidence behind each one.

   Outcomes

   • Suggested rivals appear beside the rival picker, ranked by how many prompts and providers named them
   • Only names the sector catalog accepts are offered, so adding one cannot fail on save
   • Coverage is stated as the number of comparison answers behind the suggestion rather than implied

7. Phase 3 · Measurement

   Trend alerts that survive a significance test

   See which week-over-week movements are larger than this cohort's sampling noise, and which are too small a sample to judge.

   Outcomes

   • A high-severity alert means the drop survived a two-proportion test between the two cycles
   • A large movement that cannot be tested is still shown, labelled as untested rather than hidden
   • Each alert states the responses behind it, so the reader can weigh it

8. Phase 3 · Measurement

   What a run can resolve, before the spend is confirmed

   See the smallest week-over-week change this run could confirm before you approve its cost, so a benchmark is not bought to answer a question it cannot answer.

   Outcomes

   • The cost confirmation states the movement this many responses can resolve
   • A run too small to confirm any actionable movement says so plainly
   • The responses needed for a finer resolution are stated alongside

9. Phase 3 · Measurement

   A scheduled cycle says what changed

   A scheduled measurement that finds a significant drop names it in the email subject and webhook payload, instead of only announcing that a report finished rendering.

   Outcomes

   • The email subject names the metric that moved beyond sampling noise
   • Every alert travels with the responses behind it, so severity is never asserted bare
   • Automations shows whether a schedule reaches a person at all

10. Phase 3 · Measurement

    Sources that cite rivals where the brand is absent

    See which cited sources informed answers that named a rival and not you, so the citation table names places to go and get listed rather than only who was cited.

    Outcomes

    • A cited source is flagged when it validated rivals in answers that never mentioned the brand
    • The flag states which rivals and how many questions, so it can be checked rather than trusted
    • A source cited where the brand was also named is never flagged

11. Phase 3 · Measurement

    Record the surface a measurement came from

    Every measured answer now records where it came from and whether the model actually retrieved from the web, so a future surface cannot be mistaken for an API call.

    Outcomes

    • A measurement states its surface rather than having it inferred from a provider slug
    • Retrieval is recorded from what the call did, not guessed from the model name
    • Existing runs keep working: the columns default to an ungrounded API measurement

12. Phase 3 · Measurement

    Stop offering surfaces the product cannot measure

    The product no longer offers or advertises AI surfaces it has no way to measure, so a configured surface is one that was actually asked.

    Outcomes

    • Default surfaces name only engines the product can call
    • The published description lists what is measured, not what is aspired to
    • A surface silently answered by a different engine fails a test rather than shipping

13. Phase 3 · Engineering

    Engineering and release hardening

    Increase release confidence with smaller domain modules, stable API contracts, stronger dependency controls, and regression gates.

    Outcomes

    • Faster, safer product changes without public URL churn
    • API compatibility and migration checks before release
    • Accessibility, responsive, focus, theming, and performance P1 closure

Gated on an open acceptance step

1. Phase 4 · Beta release

   Cost-capped banking beta proof

   Validate a repeatable Garanti BBVA operating journey with transparent completion, variance, evidence, export, and cost boundaries.

   Outcomes

   • Two separately reported weekly full-cohort cycles
   • Balanced validation-subset variance without overstated trial counts
   • Exact cross-format reconciliation and release evidence

2. Expansion · Sector expansion

   Authoritative fashion benchmark

   Use a deterministic 198-question fashion contract across 99 tagged brands and their editable tag-ranked rival defaults.

   Outcomes

   • Human-approved fashion question cohort
   • Tag-ranked default rivals with tenant editing
   • Mavi configuration and rival-selection acceptance journey

Planned

1. Phase 3 · Reproducibility

   Durable replay and workspace history

   Replay authenticated crawl evidence and revisit saved audit or discovery history after instance replacement or device changes.

   Outcomes

   • Tenant-scoped crawl manifests, hashes, snapshots, and diff history
   • Durable bulk audit and discovery history
   • Retention controls and integrity verification

2. Expansion · Sector expansion

   Additional sector rival packs

   Extend configurable rival defaults to technology, ecommerce, and lodging after each catalog passes tag-completeness checks.

   Outcomes

   • Sector-specific, editable competitor defaults
   • Coverage claims tied to an actual validated catalog

3. Expansion · Production

   Separately verified production environment

   Promote only after a distinct production environment passes independent security, restore, release, and operating checks.

   Outcomes

   • Independent production data, queues, storage, secrets, and identity
   • Backup and restore rehearsal, alerts, incident ownership, and release tagging

4. Expansion · Product expansion

   Revocable sharing and agency portfolio controls

   Share reports safely with clients and manage multiple authorized brand workspaces from an agency portfolio.

   Outcomes

   • Revocable client-safe report links
   • Agency portfolio access with explicit tenant boundaries

5. Expansion · Measurement

   Measured business observation integrations

   Relate visibility work to approved analytics, search, market, CRM, and CSV observations while keeping impact claims appropriately directional.

   Outcomes

   • GA4, Search Console, Semrush, CRM, and approved CSV observations
   • Transparent source lineage and directional impact boundaries