SIG-6 Signals & decisions

Reading the counterfactual and confidence basis

Last verified

The counterfactual is the line beginning “If you do nothing:” — Deloosh’s account of what the condition costs you if it goes unaddressed. The basis underneath it labels the quality of that estimate instead of dressing it up as a confidence percentage.

”If you do nothing”

Every Signal proposes a move, so every Signal owes you the alternative. The counterfactual states the projected consequence of leaving the condition alone: the cost of inaction, in the terms the condition is measured in.

Read it as a projection, not a prediction. It is Deloosh’s argument for why the Signal was raised, and it is the thing you are weighing when you dismiss, snooze, or approve.

The basis line

Directly below the counterfactual is a line like:

basis: heuristic_uncalibrated

heuristic_uncalibrated means the projection comes from a governed heuristic that has not been calibrated against your outcomes. It is a reasoned estimate, and Deloosh is telling you exactly that.

This is a deliberate choice. Deloosh could print “87% confidence” next to that projection and it would look more authoritative — but the number would be invented, and an invented number is worse than an honest label. There is deliberately no numeric confidence field in a Signal; the basis label is the whole of what Deloosh claims to know about its own estimate.

The same honesty runs through the product: the live data and design-target badges make the same distinction about a card’s underlying data. See OS-5, What do “live data” and “design-target” badges mean?.

What sits alongside it

  • Evidence — a count of the real ledger rows behind the condition, such as 12 real rows. The condition is not a hunch; it is drawn from your sending records.
  • Severityneeds attention (high) carries the note that the condition is “Large enough to plausibly move your sending health”; advisory (low) is “Small today - worth knowing, not urgent.”

Together these tell you three separate things: what Deloosh found (evidence), how much it matters (severity), and how well it can estimate the consequence (basis). They are worth reading separately — strong evidence and an uncalibrated projection can, and often do, appear on the same card.

For the full anatomy of the card, see SIG-1, What is a Signal?. For why Deloosh refuses to invent figures anywhere, see TRUST-2, Why the AI won’t make things up.

More in Signals & decisions