Algo Console

Ship ranking experiments without waiting for the ML backlog.

The Console is where your team builds, previews, validates, deploys, and monitors ranking logic — as visual stages, an AI prompt, or code, kept in sync the whole way.

console.mbd.xyzdraft
🧩 Compose · scratchpad
BLOCK
Search
INDEX
polymarket-items
MODE
Filter & Sort
BOOST
group_boost · wallet tags
mbd.search() .index("polymarket-items") .boost().groupBoost(…)
Insert into DSL ↵
Algorithm · Studio DSL
▶ Run draft
Preview · runs against the current DSL
CONVERGENCE · HIGH
3 top-ranked traders converged on Bitcoin Up or Down (Polymarket)
#17Trader · +$5.9M PnL · 28% win$16
#57Trader · +$8.5K PnL · 51% win$10
$41 inflow · still holding65¢ → 73¢ (+12.3%)
WHALE MOVE
Top-50 wallet opened $314.20 on Counter-Strike: M80 vs NIP (BO3)
#3Trader · +$12.4M PnL · 61% win$314
48¢ · 654.59 shares8 minutes ago
weights 70/30fallback ×2deployed
3 ways
to build: visual, AI prompt, or code
100+
algo recipes tested
10s
sandboxed test runs, pre-deploy
1–2 weeks
from kickoff to production
Build

Three ways in, one source of truth

PMs sketch it visually, anyone can prompt it, engineers refine the code — the form and the DSL are two views of the same algorithm.

visual
Compose with stages

Search, Hydration, and Ranking stages as forms: pick an index, add filters and boosts, choose what context to attach, set the sort. Every change is mirrored in code, so nothing is a black box.

ai
Describe it in English

“Liquid election markets this wallet would trade” — the AI builder turns the prompt into a working algorithm, and a conversational assistant helps you refine it from there.

code
Write the DSL

Drop to the Studio DSL when you want full control. Validation checks syntax and security and signs the exact code that passed — only reviewed code reaches production.

Deploy & operate

Production controls, not a demo toy

Staged rollout

Draft for experiments, Group for a beta cohort from a CSV whitelist, Live for everyone — promote when the numbers look right.

Tiered warmup

Your most active users get feeds rebuilt every few hours; dormant ones on demand. Compute follows attention.

Config without redeploys

Algorithm weights, dual fallback tiers, cache TTLs, and exclusion rules live in the feed config — tune them without touching code.

Preview before it ships

Run any algorithm or config against live indices and inspect real results as cards, a table, or raw JSON — before saving anything.

Notification streams

Wire algorithms to webhooks with daily budgets, cooldowns, and priority filters. Send a test payload, preview the last 100 matches, then activate.

Your data, ingested

Provision a data source, push items, users, and engagement events, and reference them from any algorithm or stream.

Measure

Watch what a weight change does

Every feed reports engagement with 7-day deltas, so a config tweak on Tuesday shows up in the numbers by Friday — no analytics sprint required.

DAUimpressionsengaged itemsengaged usersunique authorsunique items

Your ranking team, without the hiring plan.

Build a feed this afternoon, put it in front of a beta cohort tomorrow, and promote it to production when the metrics agree.