Some questions are about your data. Others are about your rules, your agents, or how far Juno’s knowledge reaches.
The sort takes one step and decides which of four places holds the answer.
The first step of every answer is a sort. Juno reads the question and places it in one of four kinds. A question about what your company has decided is a policy question. A question about what an agent will do is a plan question. A question about coverage, sources or scope is a scope question. Everything else is a data question.
The sort matters because the four kinds have four different right answers. A question about your safety stock rule is answered by quoting the rule and its version, and a query over stock tables would give a number that misses the point.
Each register entry carries a name, a plain-language statement, an owner and a version. The answer quotes the entry and names it, so the reader can go to the owner and check. Entries stay in the append-only version history, so an answer given last month stays explainable.
Data questions leave the registers and go to your systems. The registers still shape the answer, because the policy and the plan supply the thresholds, the classes and the windows the query uses.
Four registers, one shape: a name, a plain sentence, an owner and a version.
Your named business policies in plain language: classes, cover, service levels, approval limits.
The published planning tables of your agents, each with the rule version it runs today.
The registered sources, tables and mapped columns, and how far the mapping reaches.
The subject areas training has covered, so a reader can see where answers are strongest.
| Register | Entry | What the entry states | Owner | Version |
|---|---|---|---|---|
| Policies | ABC and XYZ classes | A items carry the top 80% of annual value; X items have steady demand | J. Ridley | v2 · 12 Aug |
| Policies | Safety stock cover | AX items hold three weeks of cover at plant 1210; CZ items hold one week | L. Tan | v5 · 2 Sep |
| Policies | Service levels | Northbay Logistics holds a 98% line fill; Helvetia Retail AG holds 95% | J. Ridley | v1 · 4 Jul |
| Plans | PO cancellation | Cancel lines with excess above eight weeks, inside the window, under $25,000 | L. Tan | v3 · 18 Sep |
| Plans | Expedite | Pull in lines where the shortage lands inside the lead-time fence | L. Tan | v2 · 11 Sep |
| What Juno knows | Registered scope | 23 tables from five systems, 412 columns mapped: 371 proposed, 338 confirmed | A. Novak | v7 · 19 Sep |
| Data territories | Purchasing | Purchase orders, schedule lines, suppliers and supplier confirmations | A. Novak | v3 · 18 Sep |
Seven entries of the showcase estate’s registers. The expedite plan belongs to a designed role from the agent blueprint, shown as it would be registered. Showcase estate. blueprint
Each answer states the place it came from, so a disagreement has an address.
The sort is shown to the reader. An answer from a register says which entry it quotes and who owns it; an answer from your data shows the terms, the systems and the query. A reader who disagrees with an answer therefore knows exactly where to go: to the owner of an entry, or to the mapping behind a column.
The registers hold statements; your systems hold the rows.
A register entry is therefore a by-product of work that already happens: writing a policy, releasing an agent, registering a source, running training. The registers turn that work into something a question can reach.
When a question falls outside the entries a register holds, Juno says which register it read and who owns it, and offers the nearest entry it has. Those questions collect in a list your owners review, so the gap is visible and someone can close it.
Analysts ask about this directly, because it is the difference between an answer that rests on an approved statement and an answer assembled on the spot.
Gartner, 17 June 2025: organisations that prioritise semantics in AI-ready data will raise the accuracy of generative AI models by up to 80% and cut cost by up to 60% by 2027.
Showcase estate.
With the cancellation policy released at supervised level, D. Okafor asked what the agent would do with a line worth $40,000. That is a plan question. Juno quoted the published planning table of rule version 4: lines of $25,000 and above are drafted and passed to the buyer.
J. Harrison then asked which customers hold a 98% line fill. That is a policy question, answered from service levels v1 with J. Ridley named as its owner. A third question, about excess above eight weeks of demand, went to the data and came back in 4.2 s with the query on show.
Save a line of questions as a walkthrough, share it with the team, and schedule it as a weekly digest.
Policies are written by their owners. Plans arrive from the rules engine when an agent is released. Registration and sync scans fill what Juno knows, and each training run updates the data territories.
Juno states which register it read and who owns it, and offers the nearest entry it holds. The question joins a list your owners review, so the gap stays visible.
A policy states intent in plain language. The Planner Agent turns it into a reviewable planning table, and the rules engine executes that table. The planner agent page shows the step.
The named owner of the entry. Every version keeps the owner, the date and the change, and restoring an earlier version creates a new version on top.
The registers hold statements and scope: policies, plans, sources and subject areas. Your rows stay in your systems and are read in place when a data question arrives.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.