Your data, first
It reads your retail sales before it says a word. On this run it touched the data three seconds in.
The Bench is the layer of Bedrock that turns one question into a briefed team. An orchestrator reads your data, locks the scope, hands each specialist seat one job, settles their disagreements, and hands back an answer, a deck or a custom app. Ask one thing, and every seat that has something to add gets to work.
It’s not a team of analysts. But it’s not-not a team of analysts.
Here’s what happens when every seat works the same question.
It reads your retail sales before it says a word. On this run it touched the data three seconds in.
Category Analyst, Category Manager, Trade Analyst and more. Each seat knows its job. The orchestrator only calls the ones the question needs.
Brand, retailer, window and share basis are pinned before any specialist starts, so four seats read one set of numbers.
Rost Raisin Bran is winning share at Fred Meyer, on one item and on base volume.
You choose what comes back: a written answer, a deck for the buyer meeting, or a custom app your team keeps using. This run built an eight-chapter category review with live charts that exports to PowerPoint.
Every question falls through the same eight layers in the same order: sources, harness, platform, tools, taxonomy, intelligence, agents, solutions. Only what the question touches lights up. What it skips stays dim, so you can see what wasn’t needed.
The layer order is the run’s real order. On 21 Sep the Bench touched data at 0:03, locked scope at 0:19, briefed its specialists at 1:01, published the deck at 9:31 and answered at 10:16.
The Bench has learned a few new tricks.
Stages, not a roster. Each step hands the next a typed piece of work, so nothing passes by accident.
In developmentEvery seat scores every option against its own goal. Where the scores split, you see the argument instead of a smoothed-over consensus.
In development · illustrative scoresTurning this finding into an ask is your call. The work is done up to the line.
It stops where judgment belongs to a person, where the data isn’t there, or where no goal was set. Then it says what would unlock the next step.
In developmentThe same relay has read four customers’ data in four categories, and caught the traps each one set.
In developmentOne sentence in. Blair, the orchestrator, locked the scope, briefed three specialists at once, weighed what came back, then handed the story to Daisy, the Deck Builder. Every line here is from the run’s own log.
Session of 21 Sep 2026 on Bedrock’s demo cereal database. Rost is a demo brand.
When two seats read the same brand two ways, the orchestrator doesn’t pick one quietly. It names the conflict, says which figure it kept and why, and carries both into the deck.
Nothing in the deck rules on profitability or payback.
No planogram, facings or reset date. The deck states the absence instead of estimating.
TPR, feature and display fields are empty, not zero, so no TPR-only cut.
Every answer ends with what was missing and what would open it up. On this run: 12 of 13 inputs present, clean evidence on the brand, moderate on the competitive ranking, and a list of what to connect next.
Ask the way you’d ask your analyst, lowercase and all. Name the brand, the retailer and what you want. The Bench works out the rest and brings back something you can take into the buyer meeting.
The headline, with its basis printed
See how it got there →The ask, sized and capped
See what it kept →The deck, ready for the buyer
See the stack →