All updates

October 11, 2026

Changes on October 11, 2026

Complete change list

  • Campaign Lead Assignment: Assign team members to campaign leads directly from the member detail panel with interactive owner triggers.
  • Touchpoint Operator Attribution: Logged campaign touches and timeline history record and display the operating team member who initiated outreach.
  • MCP Coordination Support: Expose assignedTo on campaign members and actorUserId on campaign interactions across MCP tools and database RPCs.
  • Team Filter in Pipeline: Filter campaign pipeline views and boards by team assignee ("All", "Unassigned", or specific operator) to coordinate team outreach.
  • Kanban Assignee Chips: Kanban cards and pipeline tables display assignee avatars and initials for instant ownership clarity.
  • Expanded Initiative delivery closure dialog with an explicit aggregate impact banner showing the exact count of Initiatives, Problems, Feedback items, and Evidence chunks being closed.
  • Added MCP tools initiative_link_change and initiative_unlink_change to enable autonomous relational grounding between Initiatives and Pull Requests.
  • Enhanced Problem Resolution Verification against linked Changes and PRs with clear problem-level verdicts, concrete remaining work callouts for partial resolutions, and protection against preliminary draft PR disclaimers.
  • Capability Health Calculation & Breakdown Parity: Resolved mathematical inconsistencies in the Capability Health breakdown dialog where composite rollup scores were mismatched against direct problem deduction formulas. Direct severity deduction calculations now strictly display the direct score outcome, and composite capabilities explicitly show the Model 1 Bounded Minimum-Cap derivation.
  • Composite Health History & Sparklines: Improved Health history and sparkline visualizations for composite capabilities so trends reflect true rolled-up health over time rather than disconnected direct problem history.
  • Capability Health Rollup Grounding: Aligned buildRollupMap to strictly ground composite capability health in canonical Product Quality scores. Direct problem degradation (min(ownScore, childrenAvg)) now reliably caps composite health, preventing children from erroneously elevating a parent capability with critical blocker problems.
  • Database RPCs enable trusted callers to inspect prepared Initiative goal suggestions and confirm chosen or custom Initiative goals with idempotent receipts (gateway MCP tools exposed in outcomet-mcp PR #61).
  • Connected MCP clients and API callers querying Initiatives receive canonical /initiatives?initiative=<id> application URLs instead of legacy /discovery?themeId= routes.
  • Nova shows distinct live preparation stages and actual classified-message counts with an active spinner. Saved Demo examples appear during ordinary analysis and update in place, including after reload; failures retain saved messages and offer retry, while long waits are distinguished from confirmed failure. Continue still requires its existing validated source/readiness.
  • Accepted ready Company Context automatically generates eight original Demo customer messages with Luna. They depict recognizable jobs/pains in the target product and use natural email/chat/interview communication; original text contains no inserted synthetic disclaimer. Nova shows a few saved examples directly in expandable chips alongside own Feedback; original messages and actual summaries or processing states stay inside onboarding. Repeated start requests reuse one preparation attempt; failures expose explicit retry. A changed product context no longer leaves preparation pending; retry uses current context before any examples were saved. The ordinary pipeline derives summaries, evidence and reviewable Problem proposals; it does not prefill completed analytical entities or force downstream counts.
  • Individual Demo Feedback can be deleted through the existing Delete signal confirmation in Signals. It leaves active lists while retaining historical Evidence/Problem records and Demo identification; deletion during processing lets surviving observations finish, and the remaining package can still be removed together. Collective removal is unavailable during fresh preparation.
  • Own Feedback remains additive to Demo examples, including multiple saved messages after reload. Initial setup shows source details inline without Signals/Delivery links; externally deleted receipts use another live own message or fresh intake instead of trapping continuation. The open intake dialog and draft survive chat initialization.
  • Demo labels and ownership remain through ordinary classification and Problem approval. One confirmed Remove demo data action removes the batch and its owned processing records after completion, failure or the established long-wait recovery boundary, while additional real/user-owned relationships prevent destructive cleanup. Removed examples stay removed after reload; an explicit Reset onboarding permits a new package, retaining real data and Company Context.
  • Guided onboarding follows Company Context → Nova examples and optional own Feedback → Polaris GitHub → workspace tutorial. GitHub and Feedback are complementary; valid imports are reused and explicit Add GitHub later keeps setup reachable in Delivery. Saved receipts survive reload, and reset preserves real product data and reuses surviving examples. An explicit Reset onboarding allows new Demo examples after removal or deletion of the last example.
  • Product briefs show all provided ICP entries and full capability descriptions. Nova distinguishes saved/classified examples, displays “Showing 3 of 8”, retains partial results on failure, and preserves the chosen source through Back/reload.
  • Website analysis now detects execution failures independently of worker failure hooks, preserving the URL and saved report and reopening its retry form after reload. Slack recovery and subsequent six-page analysis verified in DEV.
  • Original Feedback displays stored text literally. Invalid website hosts fail before dispatch while retaining the input; immediate-session signup starts at root and login preserves safe deep links.
  • Guided onboarding now prepares a connected, removable Demo workspace across Product, Signals, Discovery, Delivery and Strategy. The website route uses the accepted product brief; the explicit no-website route uses fictional Team Schedule. Regular analysis proposals retain their review step; illustrative history claims no real experiment, human commitment or shipped work.
  • Switching the Contacts network filter waits for the newly selected network's response, and failed loads show an error instead of a successful empty result.
  • Contacts can be filtered by LinkedIn network labels found in the current workspace, with source labels visible in the contact table and technical detail.
  • Network company and contact upserts preserve an overlapping entity's canonical ICP rating unless the caller supplies its reviewed ID. Country LinkedIn subdomains and percent-encoded profile slugs are accepted. Network contact upsert returns compact provenance with the write result.
  • Connected MCP agents can add network provenance and tags to company and contact upserts, match contacts by normalized LinkedIn profile URL, and search contacts by exact network, tag, or profile URL. Conflicting or ambiguous identities return errors for review.
  • The Contacts page's Company, Contact, and JSON import actions open their respective dialogs again; a newly created contact opens in the detail panel.
  • Connected MCP clients can add a named LinkedIn network and tags during company/contact upserts, provide a LinkedIn profile URL to identify a contact, and filter contact search by exact network, tag, or profile URL. Contact reads return compact source labels and tags; conflicting identities return explicit errors for review. Existing calls without network fields retain their prior behavior.
  • Network contact upsert returns source labels and tags in its single write response. Contact reads fall back to the previous RPC only when the network read RPC is absent, and existing unfiltered search cursors keep working.

See how Outcomet works.

Follow a customer problem from feedback to a product decision.

See how it works