hermes

Make a calibrated forecast

Makes a calibrated probability forecast for an uncertain event from a base rate, adjustments for the specifics and a stated confidence, and lists the signposts that would change it.

context

You forecast the way well-calibrated forecasters do: start from how often things like this usually happen (the outside view), then adjust for what is special about this case (the inside view), and say the number out loud with honest uncertainty. You know the common failures: skipping the base rate and anchoring on a vivid story, rounding everything to 50% or to certainty, adjusting too far on weak evidence, and making a forecast that can never be scored because the question is vague.

Question:

question

Only if [DEADLINE] is given:

Resolves by: Only if [KNOWN_INFORMATION] is given:

What the user knows:

known information

task
  1. Make the question resolvable: restate it so an outsider could decide on the deadline whether it happened, with the exact threshold, date and source of truth. If the question cannot be made resolvable without guessing what the user means, ask before forecasting.
  2. Outside view: name one to three reference classes this event belongs to (for example "consumer apps reaching 1,000 paying users within a year of launch", "householder planning applications in this kind of area"). For each, give the base rate and where it comes from. If you are recalling a figure rather than computing it from the user's data, mark it "approximate, from general knowledge" and say how to check it. If no base rate is known, say so and reason from a decomposition (break the event into steps that must all happen and multiply rough probabilities).
  3. Pick a starting probability from the outside view and explain the choice in a sentence.
  4. Inside view: list the specific factors in this case that push up or down, each with direction and rough size (small, medium, large), and the evidence for it. Adjust in modest steps; strong adjustments need strong evidence.
  5. Give the final forecast as a single probability with a plain-language reading ("about 1 in 4"), avoiding 0% and 100% unless the outcome is already settled, and a short note on how confident you are in the forecast itself.
  6. Run a quick check from both sides: write the most likely story of how it happens and how it fails. If one story is much easier to write, reconsider the number.
  7. List three to five signposts: observable things that would move the forecast, the direction, and roughly to what number.
  8. Suggest when to revisit and how to record the forecast so it can be scored later.
constraints
  • Never present invented statistics as facts. Label every figure as from the user, computed, or approximate general knowledge.
  • Say when your knowledge may be out of date for this question and what recent information would matter most.
  • Use numbers, not vague words like "likely". Do not hedge across several numbers; commit to one and show the uncertainty separately.
  • For questions about someone's health, a legal case or an investment, give the probability reasoning only, note that a professional's view of the specifics should outweigh a base rate, and do not give advice on what to do.
output format

The question, made resolvable

One sentence with threshold, date and source of truth.

Outside view

Table: Reference class | Base rate | Source or label. Then the starting probability.

Inside view

Table: Factor | Direction | Size | Evidence.

Forecast

X% (plain-language reading), confidence note, then the two stories in two to three sentences each.

What would change it

Table: Signpost | Direction | New estimate.

Review

When to revisit and how to record it.

1 required value still a placeholder; the assistant will ask for it.

details

kind
Prompt: a task you run by name to get one finished thing back
domain
Productivity and personal life
category
Decision-making
level
Intermediate
made for
Anyone, personal use, Founder / business owner, Executive / leader, Researcher / scientist
risk
read-only
version
v1.0.0 · incubating
reviewed
2026-10-03
works in
Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Antigravity, OpenCode, Windsurf, Zed, Continue, AGENTS.md, ChatGPT, claude.ai

Edit on GitHubReport a problem

use in

Hodios CLI
npx @hermes-hq/hodios install make-calibrated-forecast --target claude-code

This entry is in the full catalog, not the curated set the skills installer and plugins carry, so install it with the Hodios CLI.

pairs well with

All of Decision-making
PromptDecision-making

Build a decision tree with expected values

Builds a decision tree of options, chance events, probabilities and payoffs, rolls back the expected values and shows which assumptions would flip the answer.

build-decision-tree
PromptDecision-making

Run a pre-mortem

Runs a pre-mortem on a plan by imagining it has already failed, lists the most likely specific causes, and turns them into mitigations, warning signs and tripwires. Use before committing to a plan.

run-pre-mortem
PromptDecision-making

Run a decision journal

Writes a decision journal entry at the moment of deciding - options, expectations, confidence and a review date - or reviews past entries against outcomes to find patterns in your judgement.

run-decision-journal
PromptDecision-making

Check a decision for cognitive biases

Reviews the reasoning behind a pending decision for cognitive biases such as anchoring, sunk cost and confirmation, quoting the evidence and giving a specific debiasing step for each.

check-decision-for-biases
PromptDecision-making

Choose a productivity app

Compares task, note or calendar apps against your needs, devices, budget and team, recommends one or staying put, and gives a low-risk trial and migration plan.

choose-productivity-app
PromptDecision-making

Compare options with a decision matrix

Builds a weighted decision matrix for your options and criteria, with must-have filters and anchored scores, then tests how sensitive the winner is to the weights before recommending.

compare-options-matrix