GMAIL DELIVERY TEST: HANDOFF FOR CODEX RUNNING ON THE CORTEX SERVER Prepared: 2026-10-21 START HERE You are running on the Cortex server. This directory contains a fresh objective, access instructions, and experiment boundaries. It contains no previous test results and no passwords, tokens, application encryption keys, or connection secrets. REMOTE CORTEX FILES - This handoff: /var/www/cortex/environment/repo/gmail-delivery-test/CORTEX-GMAIL-TEST-HANDOFF.txt - Operator prompt: /var/www/cortex/environment/repo/gmail-delivery-test/CORTEX-GMAIL-TEST-PROMPT.txt - Gmail API server technical reference: /var/www/cortex/environment/repo/gmail-delivery-test/API-SERVER-GMAIL-REFERENCE.txt - Cortex project: /var/www/cortex - Existing Cortex API documentation: /var/www/cortex/environment/repo/GMAIL-API-HANDOFF.md OBJECTIVE Use controlled tests between the owner's own mailboxes to investigate whether genuine replies, Gmail-initiated conversations, or explicit recipient feedback improve later Inbox delivery. Determine which information in an incoming reply is useful, and report measured Gmail message labels rather than assuming delivery. Research relevant official and firsthand unofficial sources where necessary. TEST IDENTITIES - Gmail mailbox: zombiescloud454501@gmail.com - Domain test mailbox: admin@good-ideas.info - Sending IP to verify in the actual sender: 151.80.142.110 - These are the only allowed recipients. No worker mailboxes or bulk campaigns. TWO SERVERS: DO NOT CONFUSE THEIR PATHS OR CREDENTIALS 1. CORTEX: you are here. Inspect /var/www/cortex and the existing sender's actual single-message test mechanism, settings, rendering, and attempt records. 2. GMAIL API SERVER: separate host, application domain runwaydo.com, Laravel project /var/www/html. Its encrypted Gmail connection and Google OAuth credentials must stay there. The technical reference describes its existing PHP services. IMPORTANT ACCESS FACTS - A connection from the Gmail API server to Cortex was verified for placing these documents. That does not establish a reverse connection from Cortex to the API server, nor does it transfer the Gmail authorization to this machine. - No reverse SSH alias, API-server SSH credentials, or test-control HTTP endpoint is supplied or implied by these documents. Inspect existing authorized access configuration before claiming that the two-server test is ready to run. - Read API-SERVER-GMAIL-REFERENCE.txt. Every /var/www/html path and its PHP preflight belong ON THE GMAIL API SERVER, not on this Cortex host. - If an authorized private channel already exists, use it to execute the Gmail operations on that server. Return only sanitized verification and observations. - Do not copy Google tokens, refresh credentials, OAuth project JSON, Laravel APP_KEY, database credentials, or Cortex integration secrets onto another host or into a report, prompt, shell output, browser script, or log. - If access back to the API server is missing, inspect and prepare the Cortex single-message sender/harness, exact message plans, and the API-side commands first. Then report precisely which API-server access mechanism is missing and ask the owner to configure it privately. Do not send a blind test campaign. IMPLEMENTATION AND TEST SCOPE - Inspect current source and applicable AGENTS.md instructions before changes. - Preserve production campaign and worker workflows. Use an isolated test harness and private evidence directory; do not expose a public credential/test endpoint. - The Repo API is not automatically an arbitrary-header single-message interface. Read its documented contract and inspect the real sender rather than inventing endpoints, campaign IDs, SMTP credentials, or server access. - Normal campaign placeholder replacement must still work when using campaign rendering. Do not inject a Repo tracking div into these Gmail API test messages. - Maximum per run: six domain-to-Gmail test messages and two Gmail-to-domain test messages. Dispatch one attempt at a time, using the exact listed recipients. - Persist a run plan and unique physical Message-IDs before dispatch. A send timeout after possible submission means outcome unknown; never resend it automatically. - Start with baseline and genuine-reply cases, then a separate Gmail-initiated conversation case. Use actual current parent IDs and a meaningful message body. - A Not-spam intervention is limited to an identified test message, only after its initial labels are saved. Record the intervention separately from arrival results. - Do not add Contacts, Never-spam filters, forwarding, routing changes, or broad allowlists in the default experiment. Do not delete conversations or unrelated mail. - Honor Gmail budgets, cooldowns, bounded polling, and documented retry handling. - Discover arrivals using exact RFC Message-ID plus identity checks. Check INBOX, SPAM, TRASH, and UNREAD on each incoming message. Conversation grouping and SMTP acceptance alone do not establish Inbox delivery. - Keep this experiment independent from worker completion and payment records. - Document a specific blocker if a required capability is genuinely unavailable. Do not replace missing access with fabricated results or a claim of success. REPORT For each new case: actual sender/IP and recipient, parent linkage, physical send attempt and RFC Message-ID, Gmail arrival ID, authentication results, conversation match, initial labels, intervention if any, final outcome, and limitations. Distinguish evidence from hypotheses. No universal Inbox guarantee. No previous experiment history, secrets, or unrelated mailbox content in the report. Placing these files does not start any test. The owner invokes the prompt to begin the bounded test run, with Gmail operations remaining on the API server.