Every answer in ChatLima carries three controls: mark it right, correct the term it read differently from you, or save it as the verified answer for that question.
Fourteen suppliers delivered after the confirmed date. Delta Fasteners Ltd has the highest late rate at 7.4%.
Three controls, one record for each, and one owner who decides when a correction serves everyone.
An answer arrives with its working on show: the ontology terms it used, the systems it read, the time it took and the query. That working is what makes feedback precise. A person can point at the term that reads differently in their company, in place of judging the whole answer.
Marking an answer right takes one click. The question, the answer and the working are kept as a candidate for the verified set, and the same question from a colleague brings the same result the same way.
A correction is a sentence in plain language: what the term means here, which period the review uses, which plants count. Juno stores the correction beside the question that produced it, so a reviewer sees the reason as well as the change.
Your domain owner reviews corrections in a queue. Approving one adds the meaning to the vocabulary and publishes a new version. From that moment every person asking the same question receives the approved answer, and the answer records the version it used.
Each control stands on its own, and each leaves a record with a name and a time.
One click keeps the question, the answer and the working as a candidate for the verified set.
Point at the term that reads differently here and state the meaning your company uses.
The domain owner approves a question and its answer, and that pair serves everyone from then on.
| Control | What it stores | Who may use it | What changes next |
|---|---|---|---|
| Mark it right | Question, answer, terms used, sources read, query, time | Anyone asking a question | The pair joins the candidate list for verification |
| Correct a term | The term, the stated meaning, the scope, the person, the time | Anyone asking a question | A correction enters the owner’s review queue |
| Save a verified answer | The approved question and the answer it must produce | A domain owner or a data lead | The pair serves everyone and joins the accuracy set |
| Restate the question | The original wording beside the wording that worked | Anyone asking a question | The phrasing joins the index of your words |
A correction becomes a stored fact with an owner, a scope and a version.
The review step is what keeps one person’s wording from becoming everyone’s definition by accident. Corrections from several people on the same term arrive side by side, each with the question that produced it, so the owner can choose one meaning or give each one a scope.
The questions a team repeats every week are the ones worth verifying first.
| Question | What the answer must contain | Approved by | Served |
|---|---|---|---|
| Which open PO lines can I still cancel this week? | Open lines inside the cancellation window, value by supplier | D. Okafor · Thu | 41 times |
| Which suppliers were late last quarter? | On-time delivery by supplier over the last eight weeks | D. Okafor · Thu | 28 times |
| What is our excess above eight weeks of demand? | Excess value by material and plant, with days of supply | J. Ridley · Fri | 17 times |
| Show the NCNR lines in the excess list. | Lines carrying the non-cancellable, non-returnable flag, with value | D. Okafor · Fri | 12 times |
A verified pair is a question your company asks often, with the answer it must produce. Juno keeps the wording, the terms, the systems read and the shape of the result. When the same question arrives from anyone, the verified route runs, and the answer carries a mark showing that a named person approved it.
Verified pairs also become the backbone of your accuracy set. The accuracy testing page describes how the set is scored and re-run after every change.
Corrections are kept, attributed and versioned, so the gain compounds over the weeks.
Feedback changes what words mean and which answer a repeated question receives. That is a large part of accuracy, and it is the part the people asking the questions are best placed to fix.
The other parts keep their own reviewed routes, each with its own screen and its own version history. Source rows stay in your systems. Column mappings change on the mapping review screen. Policies and agent rules change in the versioned rule store, with simulation before release. Access rights come from your identity provider and your security roles.
| What a correction touches | Where it lands |
|---|---|
| The meaning of a term | The vocabulary, as a dated version |
| The period or filter a question assumes | The question’s verified pair |
| The wording people use | The index of your words |
| A column that carries the wrong meaning | Mapping review |
| A policy or an agent limit | The rules engine |
| Who may see a column | Security and access |
MIT NANDA, July 2025: 95% of pilots show a profit effect too small to measure, and the tools that retain feedback from their users do better than the tools that start each cycle fresh.
Showcase estate.
L. Tan asked which suppliers were late last quarter. The answer used the requested date over three months, and in her company “late” means later than the confirmed ship date, over the last eight weeks. She pointed at the term and stated the meaning in one sentence.
D. Okafor approved the correction at 11:02, and vocabulary v3 was published. The Monday shortage review, which fourteen people follow, has used the new meaning ever since. Four weeks later the showcase estate held 62 verified pairs, and nine of the ten questions in that review were served from one of them.
The four registers that tell Juno what your policies say, what your agents plan, and how far its knowledge reaches.
Anyone who can ask a question can raise a correction. A domain owner or a data lead approves it, so the meaning your company publishes stays in the hands of the people who own the term.
They keep the current published vocabulary. The correction stays visible to its author and to the owner’s queue until it is approved, and then it serves everyone.
The owner sees both corrections with the questions that produced them. One meaning can win, or each can receive a scope, so purchasing and logistics keep their own sense of the word.
Feedback changes meaning: terms, wording and saved answers. Source rows stay in your systems, column mappings change on the mapping review screen, and rules change in the versioned rule store.
Yes. Each vocabulary version lists the corrections it carries with the person and the date, and every answer records the version it used, so an old answer stays explainable.
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.