After the document, the project goes on: the invoice comes in and is compared with the projection, assumptions are re-evaluated on every environment snapshot, and the track record shows recommended against measured, including where we were wrong.
The problem
Traditional consulting delivers the report and disappears; the tool delivers the answer and forgets. Six months later, nobody compares what was promised with what happened, and the projection error never teaches anything.
The track record confronts every promised number with the measured one, month by month. A field without a measurement shows a dash, never a zero, and the cost deviation becomes calibration for the next projection.
How it works
Cost, latency, ingestion and assumptions in the same table: projected, measured and deviation. The project’s accuracy stays visible, with the bias declared.
"Not measured" exists as a first-class state. It is the difference between honesty and a dashboard that only shows what went well.
Recommended against measured
Analytics layer · 4 months since rollout
| What was claimed | Projected | Measured | Deviation |
|---|---|---|---|
| Monthly cloud cost | R$ 2,600 | R$ 2,840 | +9.2% |
| Main dashboard query | < 2.0 s | 1.4 s | within |
| Daily ingestion | < 15 min | — | not measured |
| Assumption: volume below 1 TB | — | 61 GB | holds |
Cost accuracy on this project: within plus or minus 20% in 3 of 4 months, biased low.
The connected agent’s periodic snapshots feed the watch: a new table, a removed index and anomalous growth become warnings, and the agent’s silence does too.
A broken assumption and an open objection show up in the same place, because the track record asks a single question: is the decision still standing?
Receives the investigations, runs them in your environment in read-only mode and returns the reports on its own.
Connect in Claude Code
claude mcp add EVODA --transport http …/api/mcp
Index removed from public.pedidos: idx_pedidos_cliente
Anomalous growth in public.eventos: ~1,200,000 → ~2,400,000 rows (+100%)
Silence warns too: if the agent stops reporting, the watch flags the absence.
Under the hood
The rules that keep the follow-up from becoming propaganda.
What comes next
12- and 24-month TCO computed by arithmetic, a declared source per price, and the deviation checked later against the real invoice.
See it from the inside →MemoryMemory and assumptionsFacts, decisions and assumptions outlive the conversation, and each assumption is watched with the condition that would invalidate it.
See it from the inside →MCPConnected agentThe Claude Code on your side connects over MCP, runs read-only investigations and returns reports and environment snapshots on its own.
See it from the inside →Paste the structure, get the most serious findings in about two minutes, and decide later whether an engagement is worth opening. No account, no card.