Lab · research instruments

Can taste be measured?

Partly. Two instruments, each built on published research and my own recorded judgements — kept as separate scores, because a design can be kind to readers and still not mine, or mine and still unkind.

Paarvai · பார்வை

A critic that shows its sources.

Paarvai renders a page in a real browser — desktop and phone, light and dark — and measures it: Hasler–Süsstrunk colourfulness, edge density as a complexity proxy, calm-surface ratio, hue families, contrast of every text node, line length, leading, type families, declared motion and whether reduced-motion is honoured. Then it scores two lenses.

This site, scored by Paarvai

100Evidence38 principles
100Tastemy recorded picks

Verdict: kind to readers, and unmistakably his. Measured 25/9/2026.

Seven pages critiqued; worst: an essay page 93/92 (complexity just above the strong-reference p75).

What it cannot see: imagery, copy tone, JavaScript-driven motion, anything below the first screen for pixel metrics. A high taste score is partly circular here — I built this site knowing Paarvai's rules. The honest check is you.

The evidence lens

38

principles, each graded by the strength of its evidence. 2 were downgraded as weak or failed to replicate — including the famous red/blue study, which did not survive a larger replication.

colour · 6complexity · 8space · 5layout · 6type · 3motion · 2philosophy · 8

My taste, as axes

  • cool ↔ warmhigh
  • decorative-first ↔ accessibility-firsthigh
  • ornamented ↔ restrainedhigh
  • playful ↔ seriousmedium
  • solid ↔ translucentlow
  • static ↔ motion-richmedium
  • monochrome ↔ meaningful colourhigh
  • sharp ↔ roundedmedium
  • one typeface ↔ serif prose, sans UIhigh
  • one density ↔ density per contextmedium

never

Never use pure white or pure black as a surface/ink colour -- always a warm, slightly toned paper/ink pair.

pure white reads sterile/clinical against every other warm token; pure black has no warmth and breaks the palette's coherence.

prefer

Keep brand/identity colour and functional/semantic colour in architecturally separate layers, and let only one file own literal colour values.

so a rebrand of the identity layer (e.g. purple wordmark -> marigold) never touches meaning-bearing colour (Viyugam stays purple, alarm stays red), and drift is caught by lint rather than eyeballing.

prefer

On brand/marketing surfaces, use exactly one chromatic accent; on information-dense software, use a full meaningful semantic palette -- don't confuse the two registers.

a marketing surface needs one thing to look at; a dashboard needs colour to carry state.

never

Never rely on colour alone to signal meaning -- pair with shape, icon, or label.

accessibility (colour-blindness) and because hue-only signalling is a decoration, not a real UI contract.

prefer

Treat WCAG AA (and 7:1 for anything projected) as a hard design constraint to design INTO the token file, not an audit run after the fact.

contrast fixes made after the fact are patchy; baking the ratio into the token comment makes every future colour change self-checking.

prefer

Cut any panel, legend, or log that is either read-only-and-unread or duplicative of another queue, even if it was the 'correct' thing to build originally.

clutter has a real cost even when each individual piece is defensible; the test is whether it's actually used, not whether it's conceptually justified.

A sample of the evidence

  • Score colourfulness and complexity for first-impression predictionstrong

    Compute colourfulness and complexity on any candidate design and use them as the primary automatic first-impression score.

    Reinecke, K. et al. (2013). 'Predicting users' first impressions of website aesthetics with a quantification of perceived visual complexity and colorfulness.' CHI 2013. n=450 websites, 548 raters.

  • Default to positive polarity for sustained-reading surfacesstrong

    Default text-dense, sustained-reading surfaces to light background/dark text; offer dark mode as an ambience preference layered on top, not as a legibility claim.

    Piepenbrock, C., Mayr, S., Mund, I., & Buchner, A. (2013). 'Positive display polarity is advantageous for both younger and older adults.' Ergonomics, 56(7), 1116-1123.

  • Honour prefers-reduced-motion unconditionallystrong

    Any design that ships large-amplitude motion by default without honouring prefers-reduced-motion should be flagged as an accessibility defect, not a taste choice.

    W3C WCAG 2.2, Understanding SC 2.3.3 'Animation from Interactions.'

  • Constrain body-text measure to roughly 55-75 characters per linemoderate

    Constrain the measure of body text to ~55-75 characters per line for perceived ease, even though longer lines are not objectively slower to read.

    Dyson, M.C. & Haselgrove, M. (2001); Dyson, M.C. & Kipping, G.J. (1998). Reviewed in 'Optimal Line Length in Reading -- A Literature Review.'

  • Score whitespace ratio as a luxury/quality lever, without invented percentagesstrong

    Increasing whitespace ratio around a focal element is a real, recently-confirmed lever on perceived quality; report it directionally, never with a specific unverified percentage.

    Iseki, M. et al. (2025). 'Perception of Luxury and Product Quality in Package Design: Examining the Effects of White Space, Typeface, and Visual Texture.' Journal of Sensory Studies. n=1193, 2x2x2 design.

  • Use rasa as the engine's top-level mood ontologynot empirical

    Replace an open-ended, vague 'vibe' vocabulary with a small closed set of named rasa-like registers, each bound to a concrete bundle of colour/contrast/motion/whitespace parameters.

    Bharata Muni, Natya Shastra; ninth rasa (shanta) added by Abhinavagupta, 10th century CE.

Oli · ஒலி

A room that tunes itself to a mood.

Built on the valence–arousal circumplex. Tempo carries arousal, scale carries valence, and the two are kept apart. Steady sound for focus, because changing-state sound is the strongest distractor in the research. No speech, ever. Loudness held between −23 and −27 LUFS — every mood checked with an EBU R128 meter before it ships.

Calm focus

Yaman-like, 62 bpm, 30% silence, pink air

Deep work

modal drone, 70 bpm, brown air

Wonder

Marwa-like, bell voice, 5 s room

Warm

Bhoopali-like, pluck, 74 bpm

Playful

Desh-like, pulsed, 112 bpm

Reflective

Bhairavi-like, 40% silence

Energised

132 bpm, dense, short room

Dusk

Kafi-like, bittersweet

Night

Malkauns-like, 50 bpm, 6.5 s room

Open the full studio →

Raga names are melodic inspiration, not emotion claims — the research does not support a fixed raga-to-mood mapping, and the time-of-day (samay) default is labelled as tradition, not evidence.