RESEARCH / METHOD
How TEMPER is allowed to speak
TEMPER is a Deep Well Labs research instrument. It is a public ledger of feeling about AI, not a product page and not a live poll. Origo may cite it in a footnote. Origo does not host it.
Five instruments
Named, and allowed to disagree. There is no sixth “overall sentiment” score.
- Concern–Excitement. Pew’s item on the increased use of AI in daily life, plus the international cousins of that item. Prior waves are kept. They are never overwritten.
- Lived Use. People, workers, and firms stay in separate rows. “Used a chatbot” is not “the firm used AI in the last two weeks.”
- Relational use versus the public help / hurt verdict. Companion or emotional queries are a use measure. Whether the public thinks that use helps or hurts is a verdict. Both stay on the page.
- Dignity / meaning. Religion and philosophy as an institutional tape (encyclicals, fiqh boundaries, public teaching). Not “percent of religious people who like AI.”
- Firm gradient. Census BTOS is the default national series. Consultant surveys of large organizations and executive speech require a source toggle so they cannot impersonate the Census.
Three layers
The layers are not souped.
- Layer A, ground truth. Pew ATP / Global, Gallup / SCSP–Gallup, Anthropic Public Record, Elon companion survey, Census BTOS. Field dates, question text, and N travel with the figure, and margin of error where the source states one. The institutional tape also lives here: published primary texts, quoted and linked, never scored.
- Layer B, nowcast. A fixed news panel plus a social sample, aspect-scored by target, stance, affect, and speaker class. It is not public opinion. English-language social media is not “the world.” A nowcast does not replace Layer A.
- Layer C, human coding panel. 200–500 items a week, dual-labeled. Automated series need a reliability badge. No badge, no headline. This panel is not standing. This page does not claim that it exists.
Falsification rules
- Do not add an overall sentiment score.
- Do not treat English Twitter as the world.
- Do not overwrite Layer A with a nowcast.
- Do not paraphrase quotations with a model.
- Do not claim national representation between Pew waves.
- Do not present TEMPER as an Origo feature.
- Do not put model text next to a Pew percentage as if they were the same kind of fact.
- Every figure keeps an origin label: survey / institutional tape / human post / headline / earnings / model summary.
Stage
Layer A is standing and its figures are published. Layers B and C are not, and the ledger says so where they would appear. The badge “Instrument · not a live poll” stays visible. Weekly entries, once closed, are frozen.
Request the weekly brief at phil@deepwelllabs.com. That is the existing Deep Well Labs mail path. It is not an Origo signup.