The questions a security or procurement review asks, answered in order.

Each outbound flow has its own condition, and turning one on does not turn on another. The full list is in section 04 of the Trust page.

• On the Mac. Flow reads the AI tools' local history and live hook events, recognises what each event was, and produces the count. Flow is a macOS app (macOS 12 or later).
• The collectors are open source. The hooks for Claude Code, Codex CLI, Cursor, Gemini CLI and Windsurf, and the VS Code and Cursor extension, are published under the Apache-2.0 licence with a public event schema, at github.com/ascendaone-com/ai-engineer-tools. A reviewer can read exactly what each event carries.
• Installed per user, without privileges. The hooks install into the engineer's home directory. No administrator rights, no sudo, no global package install, and nothing added to the shell profile or PATH. They run on Node 20 or later.
• Releases are signed. Each release is built only by the public release workflow and signed with keyless Sigstore build provenance. Checksums are published with the release so a build can be verified before it runs.
• Inspect before anything is sent. Installed with --no-pair, the hooks write events to a local file and send nothing. A security team can read that file to see the exact payload before approving a pilot.
• In Ascenda's cloud. Events arrive over HTTPS, are stored against the purposes the engineer granted, and are aggregated into the cohort's Report or Study. Our servers refuse an event sent under a purpose the engineer has not granted.

• On the Mac, the engineer's journal, reflections, timeline and check-in history are held in a local database encrypted with SQLCipher.
• In the cloud, data is held in managed database and object-storage services, encrypted at rest by the platform and in transit over TLS.
• Region. Ascenda's cloud resources are provisioned in an Australian region.
• Retention. Retention periods are configured by data category. An engineer who revokes AI data processing has the live tool telemetry held for them purged.

• The engineer sees their own record in Flow, in full.
• The organisation sees aggregates of ten or more people in a Report or Study. A view of an individual does not exist in the product: not restricted, not on request. The organisation never learns who joined, who did not, or which mode anyone chose.
• Ascenda operates the service. Sign-in is delegated to an external identity provider and the backend stores no passwords. Every request's access token is verified for signature, issuer and expiry before it is trusted, and identity comes only from that token. Administrative endpoints require specific roles, evaluated from the verified token.

• A floor of ten. A figure is shown only when at least ten people contribute to it, and withheld below. Small teams inside a cohort are suppressed rather than shown.
• Pseudonymous contributions. Work telemetry is keyed to an installation, not a name, and repository names arrive as salted hashes.
• No score, ranking or prediction about any person is produced. Ascenda observes measurable patterns of work. How the work felt is reported only as what people told us in their own check-ins, in aggregate.
• Currently opt-in in every jurisdiction. Nobody is counted unless they join. If that changes anywhere, the Trust page will say so first.

• No access to your systems. No repository access, no single sign-on, no CI or ticketing integration, no change to your AI tooling or its licences.
• Network. Outbound HTTPS from each engineer's machine to Ascenda. Nothing inbound.
• On each machine. Flow for macOS, or the collectors alone where Flow cannot be installed. Both are installed by the engineer, without administrator rights.
• Restricted environments. The collectors can run unpaired and write to a local file, which lets a security team review the payload, or forward events to a receiver the organisation runs itself, before any Ascenda pilot begins.
• Your existing controls stay in place. The hooks run inside the AI tools your engineers already use and add no new agent or model.

• Per purpose, by the engineer. Every consent can be withdrawn on its own from Flow's settings. Withdrawing one changes nothing else.
• Leaving a Report or Study stops the engineer being counted from the next report build. Reports already issued are not recalled.
• Revoking AI data processing purges the live tool telemetry already held for that engineer.
• Deletion on request. Any engineer or organisation can ask for data already sent to be deleted: privacy@ascenda.one.
• Offboarding. When an engagement ends, engineers withdraw the purposes they granted for it in the same way, and the organisation's report remains with the organisation.

• Delegated sign-in and verified tokens. An external identity provider handles authentication; the backend stores no passwords and verifies every access token cryptographically. Development-only authentication shortcuts are structurally absent outside development.
• Purpose-bound ingestion. Telemetry is accepted only under a purpose the engineer has granted.
• Abuse protection. Per-client rate limits apply across the API, tighter on identity and AI inference endpoints.
• Boundary controls. Browser access is limited to an explicit allowlist of origins. Interactive API documentation and job dashboards exist only in development. Unexpected errors return no internal detail.
• Secrets and change control. Runtime secrets come from environment configuration and a managed secret store. Schema changes are versioned migrations in version control. Builds and deployments run through automated pipelines, and infrastructure is described as code.
• Verify it yourself. The Trust page can be checked by proxying the app's HTTPS requests. Security researchers are welcome to get in touch.

1. Step 1

   Setup call

   We agree the cohort, the question the Report should answer, and the consent text engineers will see. Procurement, security and privacy reviewers are welcome on the call.

2. Step 2

   Engineers join

   Each engineer installs Flow, or the collectors, on their own machine and chooses whether to join, per purpose. The organisation never learns who did.

3. Step 3

   Thirty days

   The tools' own records are read on each Mac. The Report can draw on history the tools already hold, so the first read can cover weeks already worked.

4. Step 4

   Readout

   A written report of the cohort's AI exposure and work composition, with baseline findings and a proposed measurement plan. Figures are withheld wherever fewer than ten people contribute.

The 90-day Study follows the same path, runs live, and adds a declared change against a frozen baseline. Both are priced per engagement and quoted after the setup call.

• A data inventory and data-flow description: this page and the Trust page, which lists every outbound flow with its condition.
• A security and data-handling summary with its limitations register, in the form our internal evidence review produced it.
• The consent text each engineer sees before joining, per mode and per purpose.
• A sample Report with illustrative figures, so reviewers can see the shape of what the organisation receives.
• The collectors' source code, event schema and signed releases, so what is sent can be verified before anything is.
• A setup call with the founders, including the engineer and the psychologist who designed the measures.