فراسةfirāsa · reading what you can’t see from what you can

Read the signs.
Test the reading.

Firasa helps organisations put AI to work, measure what matters, and find out honestly whether it worked.

Psychology-trained. AI-native. Sydney.

01 / Sixty seconds

How much can you read from a face?

Two portraits, five personality traits each. Rate what you see, or say when you cannot tell. Then find out which answers you could have defended. It takes a minute, and it is the quickest way to see what I do for a living: check a confident reading before anyone acts on it.

A woman in her late twenties by a window, one hand at her collarbone, brows drawn together, looking just past the camera.
Waiting for newsPortrait 1 of 2 · generated, not a real person

How would you rate this person?

Slide a trait to rate it, or say you cannot tell.

02 / Which sounds like your week?

Pick the sentence closest to yours.

  • Something else.Tell me what you’re working on →

03 / How I work

Four steps, every time.

  1. Name the claim. Say exactly what you want to be true, and who needs to believe it.
  2. Read the signs. Find what can actually be observed, and what each sign could mean.
  3. Grade the warrant. Judge how far the evidence carries the claim, and where it stops.
  4. Design the next test. Choose the cheapest observation that would change the decision.

04 / Projects you can open

Every project, with its real status ↗

Start with one of these.

Two you can try in a browser, one you can run, one you can explore and one still being built. Each page says what has and has not been tested.

Featured

The Tell

A blind judging game: pick the work you prefer, guess which was hand-made, and see how well you can really tell.

Read more
Active

ConstructLens

An interview that turns what someone says into evidence-backed readings, with their own words kept attached.

Read more
Active

PsychForge v2

A team of AI agents that designs a custom benchmark for an AI system, checked the way a psychological test is checked.

Read more
Featured

Gyre

An open playground of psychology-based patterns for controlling AI agent loops, and for measuring whether they help.

Read more
Featured

Stratos

Country governance profiles on five dimensions, compared within regime type, with uncertainty shown and no league table.

Read more

05 / AI evaluations

A benchmark is a measuring instrument. Test it like one.

A benchmark, an AI judge and an interpretability probe all claim to measure something about a model. Psychology has spent a century learning how to check claims like that. I bring those checks to AI evaluation.

06 / Latest perspective

All perspectives ↗

The Organisation Reads You Back

Hiring, promotion and reward as firāsa practised with imperfect tools

Read the essay ↗

A small measurement experiment

One sign. More than one reading.

Add evidence to a plausible first impression. The point is not to reveal the answer. It is to see what the answer can survive.

Observed
18 files changed. All existing tests pass.
changed path passing tests untested failure path

Two plausible readings

Reading A

A careful improvement

The passing suite and smaller surface suggest a well-scoped change.

Reading B

Risk outside the frame

The suite may simply be unable to see the failure that matters.

What evidence should resist the first reading?

No additional evidence selected. Both readings remain plausible.

Current positionHold both readings open

Synthetic example. No codebase was assessed, and no reliability or validity claim is being made.

The website is an instrument, too.

You can watch us
jump to conclusions.

A few seconds on a page. A click on a paper. Enough to build a story about you?

Try a small experiment in how observation becomes inference, and how easily inference becomes marketing.

○ OBSERVATION OFFLOCAL EXPERIMENT

Opt in to count section visibility, selected clicks and scroll depth. The readings stay in this page’s memory and disappear on reload. No cookies or analytics uploads are used by this experiment.