What shipped.

A copy correction, not a deployment of new capability. The 31 July entry claimed every surface implying native delivery had been corrected. That claim was itself untrue: three were still live. They are named here rather than quietly fixed.

• Billing described a volume limit as if it were operating. The plans page presented a per-plan volume limit as something a merchant could exhaust, that would halt delivery once exhausted, and that would restore delivery on a billing reset or a tier change. All three describe a delivery path that does not run: the release gate refuses before the allowance is ever consulted. No plan has an outbound volume a merchant can consume, nothing halts, and nothing restores. The campaign-safe CSV export is not rate-limited.
• Getting started ended the journey at a delivery. It walked a merchant from install to a first controlled delivery, and described approval as the trigger for one. Approval freezes the audience and authorises the export; it contacts nobody. The journey now ends where it actually ends — at the first campaign-safe export, with the control already withheld.
• This changelog overstated its own completeness. The 31 July entry asserted that every surface implying native delivery had been corrected, and described the recovered-revenue ledger in the present tense of a running delivery path. Both are corrected above. A changelog that certifies its own thoroughness is the one place that claim should never appear unchecked — a reader cannot verify it, and it was wrong.

A documentation and disclosure change, not a deployment. The six legal pages were revised on 31 July 2026; where a change is materially contractual it takes effect on 14 August 2026. Nothing below describes new software running for connected stores.

• Six documents revised. Privacy, Terms, the DPA, the Acceptable Use Policy, the SLA and the service-provider list now carry a last-updated date of 31 July 2026 and state, in the document itself, that material contractual changes take effect on 14 August 2026. Corrections that only describe an existing practice more precisely apply immediately, because they were already true.
• Sending from Retrics is described as unavailable. Direct email to your customers, SMS, WhatsApp, scheduled delivery and the Klaviyo audience handoff are built into the product and switched off in code. No plan, limit, approval, connection or onboarding step turns them on. Approval freezes the audience and authorises the export; it contacts nobody. What is available is the campaign-safe CSV export: the frozen audience with your global control already dropped, which you send from your platform of choice. This entry originally claimed every surface implying native delivery had been corrected — it had not, and three were still live when it was published; they were corrected on 2 August 2026 and are listed in that entry.
• Delivery providers listed as inactive. The service-provider page now separates providers that are active from delivery providers that are not. Twilio, Meta's WhatsApp Business platform and Klaviyo receive no customer data at all in this release, and Resend carries only transactional and authentication email addressed to you — never messages to your customers. Each is listed with its status stated rather than removed, so the disclosure is not made by omission.
• Revenue, stated as revenue. Retrics ingests no product-cost, COGS, shipping or returns data, so it cannot compute a margin figure of any kind. Fourteen labels and claims across the site implied it could; each now says what the product actually reports — expected revenue for a forecast, store-verified revenue once orders land, and a recovery ledger whose figure stays withheld until the holdout comparison can settle it. Paid advertising and loyalty rewards are both graded against store-verified revenue.
• Autopilot, Google and AI drafting restated in place. Unchanged in substance, restated where the documents describe them: Autopilot recommends and never creates, edits, pauses or changes a campaign; a connected Google account is read-only and receives no customer list; Meta is the only advertising platform that receives customer data, as irreversible email hashes; and customer-facing AI drafting is switched off platform-wide, so you write every message.
• In-product notice. Signed-in workspaces now carry a one-line notice of this revision with a link to the terms, on every authenticated screen.

The first public release. Everything below is shipping now for connected Shopify stores.

• Customer Intelligence 360. A single profile for every customer — lifetime value, order history, product affinity, cadence, and lifecycle stage — assembled from imported Shopify orders, customers, and products. Retention math relates internal customer and order IDs; an irreversible email hash supports matching and suppression alongside the contact fields needed for the profile and merchant-approved actions.
• Health & return probability. Each customer carries a health score and a modeled probability of returning, trained on your store's own purchase history. Scores update as new orders sync.
• Cohort & retention analytics. Retention curves by acquisition cohort, so you can see how each month's customers repeat over time and where the drop-off happens.
• RFM segmentation. Recency, Frequency, and Monetary scoring across your whole base, with segments (champions, at-risk, hibernating, and more) you can filter and act on.
• Recovered-revenue ledger. A running record of each campaign and what came back after it, so the impact of retention work is measured rather than assumed. A control is withheld when the audience is frozen, which establishes who was deliberately left alone; the attributed dollar figure is withheld today while the comparison is rebuilt to count only the customers a message actually reached. Retrics does not deliver customer messages in this release — you export the campaign-safe audience and send it from your own platform — so no delivery is recorded against that control.
• Reorder & win-back boards. Prioritized worklists of customers who are due to reorder and customers slipping toward lapse, ranked so you can start with the highest-value moves.
• Second-order targeting. Surfaces first-time buyers at the moment they are most likely to place a second order — the window where our backtest found the flag most precise. That is a richer pool to spend attention on, not a measured gain from messaging it.
• Command palette. Jump to any customer, segment, or view from anywhere with a keyboard-first command palette.
• Saved views. Save any filter, sort, and segment combination as a named view and return to it in one click.
• CSV exports. Export any segment or worklist to CSV for your own analysis, spreadsheets, or downstream tools.

• Cohort engine. Rebuilt retention and cohort computation for accuracy at scale and faster refresh on large order histories.
• Return-probability model. Validated predictive flags against a backtest with a slice of history held back: customers flagged inside the second-order window returned at 3.2× the rate of unflagged customers in the same window, which set the direction for launch.
• Protected customer data. Completed Shopify protected-customer-data handling and wired the mandatory data-erasure webhooks end to end.

• Shopify sync. Read-only ingestion of orders, customers, and products through Shopify's API, with incremental syncing to keep profiles current.
• Customer profiles. The first end-to-end customer profile view, assembled from live store data.