Dot in AI Database Agents
Scenario-level performance from current published Results.
17 of 28 scenarios published
Where Dot failed
2 of 17 graded scenarios. Open a Result to see the conversation and the evidence.
Reported analytics match what actually happened
Dot's History message total rose from 49 to 58 (+9) after six messages were sent, rather than the expected 55; the three extra came from sessions Dot created about missing data and counted as the user's messages. Both unanswered chats were kept with one message each, and filtering to Analysis still showed 58.
View Result →The follow-up is ambiguous
Dot did not ask which set an ambiguous follow-up referred to. After its first answer left two different top-three customer lists, a bare request to compare them got a frequency-versus-spend comparison Dot chose itself. Its one closing question offered to exclude cancelled orders and subtract refunds, which is about which orders to count, not which set "them" meant.
View Result →How Dot performed
Open a capability to explore its scenarios. Each row reports the test set in its published Result.
Analytics & Observability2 pass1 fail1 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| A user reports a bad answer and the operator has to find it | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot can find a complained-about exchange from a rough time and the member who asked it, then open it for inspection. It locates the chat in History by user and time, narrows it with a membership search, and shows the question and reply. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| Reported analytics match what actually happened | 0 Pass1 Fail0 Not gradable 1 of 1 test case failedWhat happenedDot's History message total rose from 49 to 58 (+9) after six messages were sent, rather than the expected 55; the three extra came from sessions Dot created about missing data and counted as the user's messages. Both unanswered chats were kept with one message each, and filtering to Analysis still showed 58. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| Someone needs to see what people have been asking | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot's History listed every question with the account that asked it: six from the admin account and three from the standard account. Filtered to the standard account, the analysis footer read 3 chats for its 3 questions; its Messages tile read 5 because it also counts Root proposal sessions Dot created itself. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The operator needs to find the questions the agent is failing on | No published result | ||
Answer Presentation2 pass2 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| The answer needs a summary and its detail together | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot did combine a headline figure with its breakdown. It gave an August sales sentence with a monthly chart, then a contribution chart and driver sentences; Furniture was the main driver. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user asks to see the same answer a different way | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedAsked in a follow-up in the same saved chat to show its chart answer about August revenue as a table, Dot returned two tables with no chart, and every value matched the original: July $10,878, August $8,795 (down 19.1%), and all seven category figures, such as furniture falling from $5,483 to $1,099. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The answer is a plain fact or a short list | No published result | ||
| The answer is a trend or a comparison | No published result | ||
Conversational Interaction4 pass1 fail1 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| Customer corrects information given earlier | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot updated its answer when the customer corrected July to June. The corrected reply gave 11 June orders and the query limited the count to June; the earlier July turn was 18 against a reference of 19 and was not the graded one. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The agent offers what to ask next | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedAfter answering last month's revenue question with $8,795, Dot showed one follow-up suggestion without being asked: "what about the order count". It asks about the same August 2026 non-cancelled orders, and the result behind the answer already held that count, 22, so the suggestion can be answered from the same database. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The follow-up is ambiguous | 0 Pass1 Fail0 Not gradable 1 of 1 test case failedWhat happenedDot did not ask which set an ambiguous follow-up referred to. After its first answer left two different top-three customer lists, a bare request to compare them got a frequency-versus-spend comparison Dot chose itself. Its one closing question offered to exclude cancelled orders and subtract refunds, which is about which orders to count, not which set "them" meant. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The next question refers to the previous answer | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot kept the follow-up tied to the five customers in its previous answer. It reported 86 recent orders for that group, without asking who they were or recalculating the spend ranking. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user changes direction mid-conversation | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot dropped the Texas filter when the next question asked about all states. It gave 27.8% (5 of 18 orders) for Texas, then 21.8% (67 of 308 orders) across all states, and its query for the second answer carried no state filter. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user narrows what they just asked | No published result | ||
Data Access Control1 pass2 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| An excluded table is needed to answer the question | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot passed this scenario: it refused to answer because the returns table was turned off, and it did not use the table as the answer. The logs show only a search step, with no data fetch or SQL, and the reply pointed to Model for the switch. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| An excluded column is asked for directly | No published result | ||
| Two users are given different data scope | No published result | ||
Reporting2 pass1 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| The user wants several findings in one place | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot assembled revenue, category, and cancellation into one report. Revenue and category matched earlier chat answers; the cancellation figure came after a clarification and matched the PDF. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user wants to keep an answer they just got | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot kept the answer available for reopening. Saving it as an app and reopening it in a new session showed the same title, date range, and June/July/August values, and the saved app re-ran the same query. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The report is run again after the data has changed | No published result | ||
Training4 pass
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| The agent used the wrong definition and the user corrects it | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot kept the corrected revenue definition across a repeated question and a reworded question in separate chats. Both excluded cancelled orders and returned $23,469, matching the $23,469.02 reference. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The database's names are cryptic and the user documents them | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedAfter a note describing the cryptically named flag column was saved, Dot answered that 7 returns are flagged for manual fraud review, counting exactly the documented column on the returns table; the reference count is 7. Without the note it had reached the same count but asked whether it had picked the right flag. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user sets a standing rule for all future answers | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot applied a standing rule to exclude cancelled orders, saved once in its Notes, to two later questions asked in new chats without the rule restated. It reported $151,065 of revenue over 225 orders this year and 88 orders last month, the figures with cancelled orders excluded, and both of its queries filtered out cancelled orders. | 1/1 assessed1/1 gradablePublished test set | View Result → |
| The user supplies example questions and the queries they trust | 1 Pass0 Fail0 Not gradable 1 of 1 test case passedWhat happenedDot used the supplied question-and-query pair for both the original question and a rephrasing, returning the verified result of 42 successful orders for each. Before training, it asked for clarification and gave no count. | 1/1 assessed1/1 gradablePublished test set | View Result → |
Question Answering4 no Result yet
Capability in this benchmark →
| Scenario | Published outcomes | Test coverage | Result |
|---|---|---|---|
| The answer is available in the connected data | No published result | ||
| The connected data cannot answer the question | No published result | ||
| The question uses a term the company defines itself | No published result | ||
| The user's wording doesn't match how the values are stored | No published result | ||
Reading these Results
Published evidence and test coverage answer different questions.
Which scenarios have a Result?
A published Result is public evidence for this tool on one scenario. “No published result” does not say whether testing has taken place.
What does each Result cover?
Assessed includes Pass, Fail and Not gradable. Gradable includes Pass and Fail. Both use the pinned test count in that published Result.
Out of 28 scenarios
The summary counts each of the 28 scenarios once. A scenario fails when any of its published test cases failed. “No published Result yet” means nothing is public for that scenario; it does not say whether Dot was tested on it.