Oninit® DrWatson — Security & Privacy
DrWatson handles operational data — configuration files, log tails, performance counters, AF dumps — that is typically considered confidential by the customers who run the underlying database. The summary below describes how we treat that data. Detailed operational and compliance documentation is available on request through your engagement contact.
What we receive
A submission consists of an ifxcollect tarball or an
af.<hex> assertion-failure file. These are
captured by the customer on the customer-controlled host and
uploaded through the portal. DrWatson does not run on the
customer database, does not connect to the database remotely, and
does not collect anything not present in the submitted artefact.
Confidentiality
Submissions and the reports generated from them are treated as confidential information of the submitting customer. Each submission is scoped to its own ULID-addressed storage area; cross-customer access is structurally prevented (cross-tenant reads return HTTP 404 rather than 403, so a stranger cannot even verify a submission ID exists). Reports are not shared with third parties, apart from the optional language-model features described under DrWatson Q&A grounding and Narrative wording below.
Authentication & access
Portal access is gated by a magic-link sign-in. Magic links expire in fifteen minutes; sessions are short-lived and stored in HMAC-signed cookies that the server can revoke. Paid-tier features (named snapshots, customer-facing DrWatson Q&A, the per-rule reference pages) sit behind an additional authorisation check.
Transport & storage
The portal expects to run behind TLS for any production customer deployment; uploads, downloads, and session cookies travel only over the encrypted channel. At-rest data lives on operator-managed storage with normal Unix permissions; encryption-at-rest options follow the deployment host's standard practice.
Pre-upload obfuscation
Customers who want a tighter trust boundary can scrub identifiers out of captures before upload. See Obfuscation for what DrWatson reads by default, what's safe to redact (hostnames, table and database names, SQL literals, OS user IDs), and the account-level toggles available on request.
DrWatson Q&A grounding
The optional question-and-answer panel on a delivered report is grounded strictly on that report's own findings. DrWatson is not shown the underlying fact base, other customers' data, or historical reports; it has no tool surface for fetching anything beyond the question and the report it was asked about. Customers who do not want any third-party model involvement can request the feature be disabled for their tenant.
Narrative wording
WebSphere reports have their derived narrative reworded by a third-party language model. DrWatson detects WebSphere content from the uploaded logs and applies this automatically; it is not a per-customer setting. The model is shown only the derived narrative text, not the underlying fact base or your uploaded files, and the root cause analysis itself does not use a model. Every quantity, date and identifier in the reworded text is checked against the source, and if that check fails the report falls back to the standard narrative wording and says so.
Retention
Customer submissions and the rendered reports are retained on DrWatson's storage. By default uploaded files are deleted after 30 days and the whole submission, including the rendered report, after 90 days; different terms can be agreed in your engagement. DrWatson ships a daily sweeper that purges submissions past their retention window. Deletion-on-request is supported — contact your engagement representative.
Audit trail
Each finding in a report carries the ID of the rule that produced it and the evidence it rests on, so a conclusion can be traced back and checked. Reports are retained on the same schedule as the submission they came from.
Results are repeatable: running the same evidence through the same rule version gives the same findings, recommendations and evidence trail, because the core analysis is rule-based with no language model in the loop.
Reporting a security concern
For security questions, vulnerability reports, or compliance documentation requests, contact support@oninit.com or your usual engagement representative. Detailed operational posture (specific cryptographic primitives, key-management practice, data-flow diagrams, retention timers) is available under NDA.