About the data & the math

What's real, what's simulated

Synthetic demo. Every account, developer, signal, and agent query is fabricated. No real company is profiled. The scoring math is real; the data is illustrative.

The pipeline · click a stage

What's whatReal engineruns in code, liveSimulated datafabricated seed dataRoadmapnot built yet

IdentityReal engine

A pure resolver clusters raw identifiers with deterministic bindings and probabilistic scoring, a human-review band, and Art. 17 erasure - all enforced in code. Nobody is pre-tagged as resolved; the status is computed. Runs live on every account page.

The glass box · scores are computed, never hand-written

Signal type

commit · docs · agent

Intensity

how strong

Recency

time-decay

Frequency

×count

Breadth bonus

multiple devs

Surge ×

recent burst

Saturate → 0-100

concave · noise can't inflate

A concave saturation curve means the first real signals move the needle and repeated noise can't game the score. Same math on the demo - drag the weights and watch it re-rank.

Identity · when two fragments become one person

0.000.500.751.00

< 0.50 reject

logged, never merged

0.50-0.75 review

routed to a human

≥ 0.75 merge

auto-cleared

Governance · enforced in the engine, not in copy

No lawful basis → no name

GDPR Art. 6 required

Office NAT ≠ identity

+0.30 · below 0.75 bar

Two verified never merge

hard constraint

Review band routes to human

0.50-0.75 held

Art. 17 erasure recomputes all

score re-runs live

The honesty ledger · every claim, tagged

Real engineGlass-box scoring enginepure, tunable, replayable · no DB, no randomness

Real engineIdentity resolution enginedeterministic + probabilistic merges, review band, Art. 17

Real engineLLM outreach & Q&A (OpenRouter)labeled deterministic fallback without a key

Simulated dataForge DB, Acme Corp & all 12 accountsevery developer, raw event, agent query - fabricated

Simulated dataEnrichment directorynames, roles, seniority - seed data

Simulated dataFit scores, ACV bands, precision/recall labelsillustrative constants, not measured outcomes

RoadmapIngestion connectorsGitHub · npm · Docker · docs

RoadmapThe MCP agent gatewayagent signals in the demo are simulated events

RoadmapLive CRM / Slack activationlabeled payload preview only

This is an independent teardown of Reo.Dev (the AI-native developer-GTM platform, $4M seed led by Heavybit), attacked on three flanks: Glass-box scoring, Agent-native capture, Execution-grade UX. Any vendor metrics cited are their own self-reported positioning, not verified facts. Identity-resolution match rates in this space are notoriously hard - this demo doesn't claim to have solved them, and it isn't legal advice.

The thesis

Developers are the new buyers. They evaluate by doing - and now by sending their agents to do it for them - weeks before a buyer ever appears in your CRM. Traditional intent data sees none of it. Plume clears the pipeline fog.

Two structural shifts make this a category: developers adopt bottom-up (the buyer signs the check weeks later), and AI agents now do the evaluation - querying docs and calling MCP tools that never show up in a CRM. Plume is built for both.

See it in action