Omniscio

PM Data Density (earn-the-right-to-speak gatekeeper)

An in-development gatekeeper that decides when a Mission Control board has accumulated enough usage data to be spoken to intelligently, tracking six domains separately and holding fresh-start users to higher thresholds than people who imported from another platform. Its behavior-pattern engine watches item history for six recurring habits, algorithmically and at no model cost.

On this page

What it is

An in-development density tracker and behavior pattern detection engine, listed in Settings → Lab as PM Data Density & Behavior Patterns, that determines when the PM has enough usage data to speak intelligently. It is the gatekeeper behind every suggestion Omniscio makes about your boards: nothing gets recommended from a board until that board has earned it.

Where to find it

There is no surface of its own -- it is a backend engine that other PM features consult before they say anything. It is in development, gated behind the pm-data-density unreleased feature (pmDataDensityEnabled), so it is off in a normal install.

How it behaves

It monitors per-domain usage accumulation across six domains (task-management, workflow, collaboration, scheduling, board-structure, integrations) and determines when each domain has enough data to generate insights. It distinguishes fresh-start users (higher thresholds, 3-day minimum) from import users (lower thresholds, faster ramp).

The behavior pattern detection engine watches item history for six patterns:

  • Repeated manual status changes (threshold 10)
  • Recurring subtask creation (threshold 3)
  • Reassignment flows (threshold 5)
  • Batch behavior (5+ in a 30-min window)
  • Time-of-day patterns (top 2 hours >= 40% of activity)
  • Feature usage frequency (2+ column types required)

All detection is algorithmic -- zero LLM cost. Observations are gated by domain density readiness: a pattern detector will not run until its domain has sufficient data.

When the feature is disabled, isDomainReady() returns true so downstream consumers degrade to ungated behavior.

Per workspace and per account

Readiness and tips are worked out separately for each workspace, and they belong to one Omniscio account:

  • Each workspace earns readiness from its own boards' activity and its own age. A brand-new workspace starts with no tips, even inside an older, busy account, until it earns them.
  • Every tip is saved with the signed-in account and the workspace it was found in. Two accounts on the same computer never see, or dismiss, each other's tips.
  • The board Insights window shows only the tips for the open board's own workspace.
  • Whether you imported from another tool (which picks the lower "import" thresholds) and the integrations count are account-wide, because imports are not tied to a workspace.
  • Tips saved before this change had no account attached; they are hidden and expire on their own within 7 days.

What counts as a person's activity

Every count in this engine -- and in the behavior engine that gates off it -- asks one shared question, isPersonActivity() in src/main/services/pm/shared/pm-person-activity.ts. It counts human, sync and api rows from pm_item_history, and excludes automation and import.

That list is not a guess about who is a person. It is what the column actually holds. The Mission Control screen writes straight to the cloud backend, so a person's edit arrives back on this computer through the sync, which records every row it writes with a fixed actor (syncCtx = { actorType: 'sync' }); a CLI write carrying the install key is recorded as api. Only an in-app-session CLI caller and two IPC handlers write human. Measured on a live install: api 4186, sync 1720, import 158, automation 19, and human 2.

A row says who wrote it here, not who caused the change upstream. An automation running server-side leaves no local automation row, so its change arrives as sync and counts as a person. Closing that needs the backend to say who made each change; until it does, the engine over-counts a server-side automation rather than under-counting a person's work -- the direction the original bug was filed in. Reported as a known limit, not papered over.

Do not re-state the actor list at a call site. tests/unit/lint/3a/person-activity-chokepoint.test.ts fails the build if any reader filters on actor_type itself, because eleven copies that agreed on the wrong actor is exactly what this bug was.

The boards whose usage is being counted are described in mission-control.md. The snapshot layer that turns the same board data into cross-board reasoning is pm-cross-board-intelligence.md. The forecasts and risk views further up that chain are in pm-predictive-analytics.md.

Last verified 2026-10-15