Selected period
Trials and paid conversion
Trials startedConverted to paid
Business
Business overview
Free-trial activity, paid conversion, current engagement, and the people who may need attention.
Last 30 days
Right now
Live activity
Newest accounts
Recent free trials
Five-day journey
Trial status
Overview
What is driving account creation?
Start here for traffic quality, conversion, and the trend across the selected period.
Measured traffic
All visits and account creation
All measured visitsAccounts createdLikely automated
Signup journey
Where visitors continue—or stop
Live now
Who is on AphasiaHope.ai right now?
A simple, privacy-aware view of active pages, places, and arrival channels.
Last 90 seconds
Live marketing-site visitors
Visible, consented visits on AphasiaHope.ai, refreshed automatically.
0people active nowLikely or confirmed people seen during the last 90 seconds
Pages open
Approximate locations
How they arrived
Foreground-visible time
Active session timers
“Live” is a presence estimate, not surveillance: hidden tabs stop heartbeats, inactive visits age out, and approximate edge location may be missing or imprecise.
Geography
Where is interest coming from?
Country and region counts include identifier-free visits. City-scale maps use only permissioned journeys and never use GPS, precise addresses, postal codes, or stored raw IPs.
Audience map
Approximate visitor locations
Qualified visitors during the selected period.
One marker per approximate cityUse +/− to zoom · drag to move
Locations come from Netlify’s network estimate and are rounded to one decimal before storage—roughly city scale, never a person’s exact position.
Audience geography
Countries
Local patterns
Cities and regions
Acquisition
Which channels deserve more attention?
Compare first-party traffic and switch attribution models without losing the underlying journey.
Acquisition mix
Channel performance
Attribution lens
How channels share credit
Source intelligence
Top traffic, signup, paid, referral, and automated sources
Traffic and referral leaders include identifier-free daily counts. Signup leaders use detailed last non-direct attribution, so a direct return does not erase the source that brought the person in.
Journeys & quality
What happened before signup?
Trace consented multi-visit paths while keeping bots and uncertain traffic visible.
Multi-visit paths
Journeys that became signups
Traffic quality
People, crawlers, and uncertainty
Human and bot labels combine crawler signatures with consented browser behavior. “Unresolved” traffic remains visible instead of being silently discarded.
Content
What are visitors exploring and choosing?
See which landing pages earn attention and which calls to action move people forward.
Content performance
Landing pages
Intent signals
Calls to action
Email delivery
Are account and support emails reaching people?
Postmark delivery, bounce, complaint, and suppression events for the selected time window. Delivery means the receiving mail server accepted the message; it does not prove inbox placement.
Transactional email health
Postmark monitoring
Private server-side monitoring. Message bodies are never stored, and recipient addresses are masked in this view.
Action queue
Recent delivery issues
Hard bounces and complaints remain suppressed until they are safely resolved in Postmark.
Connections
What is collecting data?
Check the first-party collector now; Google and advertising imports can be connected in the next setup step.
Data connections
Measurement and advertising sources
Customer care · Service desk
Every case, every handoff, one clear next step
See what arrived, which role owns the next action, the recorded assessment and approval, and how the team is performing.
Refreshes every 45 seconds
Example preview · Fictional customers and decisions. Nothing here is saved or sent.
Live workflow settings
Automation Controls
Pause a workflow immediately. New customer messages and existing tickets stay visible for staff.
Live workload
Where cases stand
Assigned work
Agent desk
Last 30 days
Service performance
Decision trail
Recent activity
Customer messages
Inbox
Select a conversation
Open a Support or Feedback message to review it and reply.
Customer care · Protocol 2026-09-29.1
One consistent path from message to confirmation
Use this guide with the live checklist on each customer case. A checked step reflects recorded evidence or a completed action, not an AI claim.
1 · Intake
Receive and acknowledge
Save the message under a permanent AH ticket number. Send the acknowledgment around 20 seconds later. Support promises a progress update within 24 hours; feedback receives a separate acknowledgment.
2 · Understand
Classify and check facts
Identify the issue and any urgent or sensitive content. Read only the relevant membership, onboarding, practice-header, and recent-error signals. Missing records remain unknown. Ask one focused question when needed.
3 · Decide
Follow the playbook
GPT-5.4 Mini prepares the service assessment and draft. A separate GPT-5.4 Mini manager pass reviews the reply. Only an approved, fixed routine help answer can send automatically; other cases wait for staff.
4 · Technical
Investigate and verify
GPT-6 Sol records a read-only technical assessment and GPT-6 Astra reviews its evidence. A developer must still reproduce the issue, make and verify any code change, and record a deployed fix reference before support asks the member to retry.
5 · Follow through
Update and confirm
Send a support progress update within 24 hours even if work continues. After a verified fix, ask the member to check. Keep the ticket open until the member confirms resolution. Silence never closes it.
Playbooks
Choose the next safe action
Routine question
Match an approved fact and fixed reply. Auto-send only after the manager approves that exact text.
Technical problem
Record symptom and missing reproduction details; Sol investigates, Astra reviews, developer verifies, then customer checks.
Billing or membership
Check the limited membership record and route to staff. No automatic refund, charge, or entitlement decision.
Account access
Offer approved reset guidance where it fits; route changes and unresolved access to staff. Never request secrets.
Feedback or feature idea
Record the theme for product review; do not promise a release date.
Health or safety concern
Do not diagnose or infer recovery. Route to a person for a careful response.
Unclear or mixed request
Summarize what is known, ask one focused question, and preserve the ticket while it is clarified.
Model failure or disagreement
Stop automatic delivery, show the case in the care queue, and assign a person.
Operating procedure
Case checklist and decision gates
- Receive: save the original private message and permanent ticket number.
- Acknowledge: send the receipt; for support, set a 24-hour progress update target.
- Screen: identify category, urgency, sensitive content, and mixed requests. Urgent cases alert staff.
- Check facts: use bounded account and app signals; mark missing data as unknown.
- Choose a playbook: record observed evidence, gaps, next owner, and one question if needed.
- Review the exact reply: the response manager checks every proposed customer message. Only a fixed, approved routine answer may send automatically.
- Update: send a support progress update within 24 hours even when work is ongoing.
- Verify technical work: require reproduction, an evidence note, and a deployed release reference before saying a fix is live.
- Confirm: ask the member to retry and keep the ticket open until they confirm resolution. Silence never closes it.
Model and approval boundary
GPT-5.4 Mini handles customer triage and a separate response review. GPT-6 Sol investigates technical reports; GPT-6 Astra reviews that investigation. The copied Codex brief requests Sol for repository work and Astra for code review. The in-app technical assessment itself cannot edit code, approve a release, or mark a fix verified. Staff owns those steps.
Failure rule
If a model is unavailable, uncertain, or disagrees with its reviewer, automatic delivery stops and a person takes the case. No customer text is copied into Slack alerts. An opened receipt means the conversation was opened, not that every word was understood.
Source document and version history: docs/CUSTOMER_CARE_PROTOCOL.md.
Foundation system
Algorithm health
100out of 100
All monitored rules are healthy
The deterministic auditor has not found a production failure.
Needs attention
Active findings
Across the application
Module health
Selected period
Time by module
Current presence
Online users
Engagement explorer
Practice by user
Required product tour
Step-by-step completion
Follow-up opportunities
Where people stop
Individual journeys
Product-tour progress
By provider
Provider spend
By module
Module spend
Per user
User cost explorer
Most expensive routes
Provider / operation
Live event trail
Recent API cost events
Repeated problems
Issue groups
The same underlying error is combined into one group. Select a group to inspect its evidence and prepare a developer handoff.
What happened most recently
Latest evidence
Each event shows when it occurred, where the user was, and the technical clue captured at that moment.
Where it happens
Modules and routes
Developer package
Selected issue handoff
No issue selected
Open Crash Reports and select a grouped issue to copy a Codex-ready reproduction prompt.
Not loaded
Masked interaction history
Error-linked sessions
Mux delivery resilience
Video playback and retry health
All product-specific rules
Detector coverage
Grouped by experience fingerprint
Findings that need attention
What happened most recently
Experience event trail
Known rules
Safeguard coverage
Proof of adaptation
Movement checks
Traceable versions
Current production stack
Live production · synthetic QA access
Watch every screen size
Automated browser checks visit the deployed app in desktop, laptop, Android phone, and iPhone views. Open a run to see each step, its result, and the linked evidence.
No customer account accessGuarded test sessionsHourly cloud runs
Latest run—Waiting for history
Latest browser matrix
Device coverage
The Record control is checked on desktop, laptop, and Android emulation. iPhone checks cover layout and tap access; physical microphone behavior requires a real device pass.
Saved checks
Run history
Choose a run to inspect failures and completed steps.
Selected run
Step evidence
Only synthetic account results and sanitized diagnostics are stored here.
Conversation testing
Find edge cases before users do
Generate controlled variations across meaning, speech form, timing, recognition, and dialogue state. Every result is reproducible from its seed.
Deterministic casesNo real user speechChecking storage
Latest pass rate—Run a test to update
Step 2 · Review
Preview test coverage
Inspect representative cases before running the complete suite.
Your preview will appear here
Choose the variables, then select Generate preview. Nothing is saved until you run the suite.
Controlled comparison
Compare quality before changing production
Run identical synthetic cases through a baseline and a candidate profile. Review safety, response changes, and estimated cost before deciding what should move forward.
Same casesProduction unchangedChecking storage
Estimated savings—Quality gate included
Step 2 · Review
Comparison history
Open a recent run or start a new comparison.
No comparisons yet
Configure a baseline and candidate, then run your first safe comparison.
Investigation workflow
Turn a failed check into a reproducible task
Each handoff includes the test seed, evidence, failed safeguards, expected behavior, and a ready-to-copy investigation request.
No production identityEvidence attachedHuman verified
Open items0Across the developer queue
Account-level adaptive care
Personalization Studio
Schedule a private diagnostic window, review what actually happened, test a candidate configuration, and activate or roll it back for one account and module.
Account-isolatedEvery change is versioned for this account and module only. It does not modify the shared global runtime.
Step 1
Schedule diagnostic capture
Step 2
Open investigation case
Step 3
Create a versioned account configuration
Configurations are immutable. Activation is explicit, and rollback creates another traceable version.
Reviewed evidence
Label what the person meant
Use the recording while you review. These labels become development or holdout examples for this account only.
Model and algorithm test
Compare on reviewed examples
Runs the baseline and candidate independently, scores meaning alignment, and checks confident mistakes.
Capture review
Sessions and artifacts
Change control
Version history
Audit trail
Recent account actions
Controlled improvement
Algorithm releases
Promotion gate
CandidateRegression suiteShadow comparisonCanaryProduction
Changes never promote themselves. A human administrator approves every production release after deterministic safeguards pass.