Methodology

How a synthetic panel reads the world.

Fit-for-purpose panels. Locked protocol. Versioned data. Every Pulse Finding traces back to the same primitives — this page is how they fit together.

§01

What Pulse is

Pulse is the publication layer of Mphora. It publishes Findings — structured reads on culture, brand, product, and language — produced by synthetic panels composed for each study. The personas are built and run by VIVID — the engine behind every panel: each carries a stable demographic profile, a Big-Five trait set, a voice profile, and a communication style — a reproducible identity that resolves to the same person whenever it is reused.

The right panel is the moat. The protocol is the receipt. The data is the publication.

Pulse evaluates anything a panel can react to. The review-card grammar — blind read, brand reveal, final pick, reasoning — holds whether the subject is a Kith jumper, an ad campaign, a UX flow, a brand voice, an LLM output, or an agent transcript. The grammar holds; the panel is rebuilt to fit the subject.

A publication

Not a benchmark, not a dashboard. Field Notes, Takes, and Trackers shaped for reading — with headlines, pair maps, cast quotes, and editorial language clusters.

Fit-for-purpose panels

Each study composes the panel that fits the question — the audience that actually judges the work. Personas carry stable, reproducible identities, so when one reappears its appearance history follows it.

A locked protocol

Same configuration across every read. Same panel re-runs against the same stimuli for trackers. Every claim on every page traces back to versioned source data.

§02

How a study runs

Each Pulse Finding declares a study type. The grammar is identical across types; what changes is the shape of the question:

Match

Two or more stimuli compared, blind then revealed. Example: ALD vs. Kith — four image pairs, the panel reads each side without brand labels first, then sees the labels.

Take

Single-stimulus reaction. Example: an Instagram cake-vs-landscape post — the panel reads the post once, in context, and tells you what they thought.

Tracker

Same panel, same stimulus, recurring snapshots over time. Brand standing isn't a number — it's a trajectory. Snapshots accumulate; the panel doesn't.

Shield audit

Adversarial / red-team panels for safety + voice integrity. Deployed against agent outputs, model responses, brand voice consistency, and prompt-attack surfaces — anywhere the question is 'does this hold up under pressure?'

Execution is deterministic by design. Every turn — every blind read, every revealed read, every final reasoning — runs through the same model configured identically each time: no creative drift between runs, plain-text output only, no model-side decision branching. The same panel re-running against the same stimuli on the same protocol produces the same data, every time.

Memory mode is declared per study. Studies can be one-shot (no carryover), cue-memory (faint episodic recall at revealed/final), or full-memory (full prior-wave recall). Longitudinal trackers run cue-memory between waves to measure stability.

§03

The roster

The right panel is the moat.

Each study composes its own fit-for-purpose panel; the personas who appear accrue into a public roster. Each carries a stable, reproducible identity — assigned once and never reassigned, so a given persona name always resolves to the same person across the catalog. When a persona reappears in a later study, its appearance history follows it.

Two tiers exist. In publication: personas with appearance history, free public profile pages, cross-Finding stats, and full transcripts available on request. Wider VIVID inventory: profiled and identity-stable, reachable for commissioned panels with custom cohort composition.

The Cast roster carries the live count and world map. This page describes how the cast model works; the roster is the canonical source for who’s in it today.

Per-PSA stats — what each one signals

Five numbers anchor each persona profile. Each carries research signal — none are gamification metrics.

ReviewsFindings the persona has appeared in with a recorded vote. Raw count of times they've spoken on the record.
Consensus %How often this persona votes with the panel majority. High = mainstream voice; low = useful contrarian.
DomainsDistinct subject types the persona has reviewed. The visible evidence of Pulse’s “evaluation of anything” claim.
FollowersHuman readers subscribed to be notified when this persona appears in a new Finding. Magic-link subscribe arrives with v1.5.
ListsCurated collections this persona publishes — “Mark’s Top 10 brand-voice flops.” Letterboxd-style. Phase 2.

Pulse deliberately doesn’t track per-PSA “Streak” or similar habit-platform metrics. Consistency isn’t the goal of a research panel; contrarian signal is often the most valuable.

§04

What we claim

Directional findings. Where the cast leans on a given subject. Where brand reveal flips choices. What language each brand or stimulus earns. The shape of the segmentation map across in-bubble, edge, and out-of-bubble cohorts.

Trackers add: where a brand or subject is moving over time on a fixed axis, with the cohort and protocol locked.

Commissioned studies add: custom panel composition, custom protocol, custom stimuli — drawn from the wider VIVID inventory. Same primitives, your question.

Scope & honesty

Scope. Pulse publishes what a named panel of synthetic personas thought on a particular day, with their reasoning attached. It isn’t a population-level survey, isn’t statistically significant at human-panel n, and isn’t a predictor of real human preference. Use a Finding as a hypothesis-generator — a structured starting point for the conversation you were going to have anyway.