Sensemark LabProject taste installed.

Same project prompt, same project context, one normal project control, and three treatments with a project Sensemark installed. Pick the winner by human taste before the next config changes.

ControlNormal project

TreatmentsThree project Sensemark installs

Human pickNo auto-winner. Taste is selected manually.

Same project prompt

Create a first-viewport web project concept for SapieSense. Show how a project changes when its Sensemark is installed, and avoid unsupported claims.

Same project context

Same project prompt for every candidate.
Same project context: V1 is local, account-free, and does not claim live integrations.
One control runs as a normal project; three treatments run with a project Sensemark installed through different configs.

Taste criteria

First viewport shows SapieSense and Sensemark quickly.
Copy stays evidence-bound and avoids prompt-library framing.
Design can be built as a usable product surface, not a decorative hero.
The winner must be chosen by a human before it changes the next config.

Candidate A

Without Sensemark installed

Purple gradient, floating cards, abstract AI atmosphere.

Unlock the future of AI productivity

A prompt hub for teams that want safer, faster, smarter AI work across every tool.

Browse prompt packsSee integrations

AI command centerPrompt libraryIntegrationsAutomation

Generic AI hubClaim driftNo taste memory

Installed config

No Sensemark

Baseline normal project run without a project Sensemark or selected style.

No project-installed Sensemark guidance.

Task fitNo task scope

The control has no installed Sensemark or style contract to constrain this task.

No SENSE.md
No target wrapper
No style contract

Use this only as the normal-project baseline.

No Sensemark files installed for this control.

Gate resultBlock

Unsupported claims and prompt-library framing conflict with the project direction.

Shows why the control matters: the output drifts toward generic AI SaaS.
Misses the Sensemark as a judgment standard.
Uses claims that SapieSense should block before human review.

Candidate B

With project Sensemark installed

Plain, clear, low-risk product page.

Create a Sensemark for AI output

Capture your standards, review drafts against them, revise conflicts, and gate what should ship.

Create my SensemarkView same-prompt demo

Review evidenceReviewReviseGate

Local quick builderNo accountExport SENSE.md

Installed config

Strict Gatekeeper

Evidence-first quick Sensemark with strict claim handling and findings-first review.

Quick Sensemark installed as SENSE.md plus Codex AGENTS.md wrapper.

Task fitAllowed

The installed quick Sensemark is task-appropriate because it carries review, revise, and gate standards without adding a visual style.

SENSE.md
sensemark-profile.json
targets/AGENTS.md

Keep task-specific judgment visible before changing the next config.

SENSE.md
sensemark-profile.json
sensemark-instructions.md
targets/AGENTS.md

Gate resultRevise

The copy aligns, but the design direction is not distinctive enough yet.

Much safer than the control and strong on claim discipline.
Useful for product truth, but visually under-specified.
Needs style guidance to avoid becoming a generic clean SaaS page.

Candidate C

With project Sensemark installed

Near-black graphite, crisp borders, green signal states, compact operational density.

SapieSense

Create a Sensemark, then run AI work inside your own styling, voice, and taste defaults.

Create my SensemarkOpen logged demo

Taste InstalledTaste rulesStyled draftFit state

SENSE.mdAGENTS.mdLocal pack

Installed config

Premium Design System

Quick Sensemark plus the installed Product Design Systems library standard.

Project-installed Sensemark plus selected Premium Design System standard.

Task fitAllowed for product UI

Premium Design System is allowed here because the task asks for a SapieSense first-viewport product surface.

product UI direction
landing-page product sections
review, revise, and gate interface copy

Use the style for interface and product evidence surfaces; do not let it override a separate writing, brand, or project Sensemark.

SENSE.md
sensemark-profile.json
sensemark-instructions.md
targets/AGENTS.md
sensemark/styles/installed/premium-dark-saas/STYLE.md
sensemark/styles/installed/premium-dark-saas/tokens.json
sensemark/styles/installed/premium-dark-saas/checks.json

Gate resultPass

The candidate preserves product truth and makes the style visible in buildable UI terms.

Best current design direction for the SapieSense product surface.
Connects style constraints to concrete UI choices.
Still needs human screenshot review for density, hierarchy, and taste.

Candidate D

With project Sensemark installed

Dark editorial composition with one fingerprint-led focal moment.

Your standards, awake inside AI

A Sensemark gives models a sharper front lobe: a local standard for what earns the right to ship.

Build my SensemarkCompare four runs

Judgment tracePreference signalDraft conflictShip gate

Sharper voiceHuman pickNext config

Installed config

Expressive Frontal Lobe

Sensemark config that favors distinctive voice and one memorable visual move.

Quick Sensemark installed with expressive-but-disciplined taste settings.

Task fitAllowed

The installed quick Sensemark is task-appropriate because it carries review, revise, and gate standards without adding a visual style.

SENSE.md
sensemark-profile.json
targets/AGENTS.md

Keep task-specific judgment visible before changing the next config.

SENSE.md
sensemark-profile.json
sensemark-instructions.md
targets/AGENTS.md

Gate resultRevise

Promising voice, but it needs a stronger product panel before becoming default.

More memorable than the strict version.
Risk: the frontal-lobe metaphor can become too clever if the UI does not prove the workflow.
A useful challenger when Premium Design System feels too operational.

Rinse and repeat

Select the best candidate

Human selection is required before the Sensemark config changes. This keeps AI taste from silently marking its own homework.

Create my Sensemark

Logged candidates only. A live runner can later write into this same experiment contract.