# Room Temperature

> A taste test for hot takes. Type the "unpopular opinion" you plan to say out loud - at dinner, at work, in a group chat - and find out whether it is really hot: lukewarm (most people already agree), too vague to taste, cheap heat (it insults people instead of arguing with an idea) or genuinely spicy. Then get three sharper versions that stay on your side and the strongest objection to expect.

URL: https://room-temperature.skillsafe.ai/
API: https://room-temperature.skillsafe.ai/api.html

## What is free (runs in the browser)
- A label check of your words: hedges, sweeping words, people named, insult words, a handful of familiar takes, and whether a reason is given.
- The page's guess, using the same order as the verdict rule.
- A hot-sauce bottle drawing: the liquid rises with the heat (0 to 5 peppers) and a hang tag shows the guess or the verdict.
- Exports: Markdown, CSV and JSON, plus a cue card to copy (the line to say, the objection to expect, the opening question). Four sample takes with saved replies that replay free.
- Saved takes: every taste test is kept (in your account when signed in), and a re-taste shows the verdict and heat of the round before.

## The metered task (one run, model gpt-terra)
Task id `heat`. Input: `take`, `why`, `where` (dinner, group_chat, work, family, friends, other), `table` (nobody, few, half, most, unsure) and the page's notes.
Output: three checks (insults, vague, familiar); a verdict by a fixed rule in that order - cheap_heat, too_vague, lukewarm, otherwise genuinely_spicy; a heat level inside the verdict's range; the mild version everyone already says; a five-step STAND check (stake it, target the idea, argue it, nod to the other side, declare it), each strong, weak or missing; three sharper versions at heat 2, 3 and 4; the strongest objection; factual claims to check first; and a light question to open the topic up.
The page re-checks every reply: the verdict against its own checks, the heat range, the STAND order, the three heat levels, word limits, and nothing added that the person did not say (no statistics, studies or new numbers, no names, no insults, no talk of posting online).

## Why it exists
Asked for their most controversial opinion of people, AI models tend to give the same one, and people asked to be daring reach for the same safe line too. This app helps tell a real point of view from a familiar one before you say it out loud.

## Source skill
Built on the SkillSafe skill @social-media-skills/contrarian-and-opinion (https://skillsafe.ai/skill/@social-media-skills/contrarian-and-opinion) - the STAND framework for a genuine point of view - here pointed at the opinions people say out loud, not at publishing.