- Home
- Skills
- Research & Analysis
- Traceable Numbers: Verified Figures in Data Answers
Works with the AI Agents you already use
Traceable Numbers: Verified Figures in Data Answers
Makes your agent compute every figure in code, record its source, check up/down direction and reconcile totals before answering, so it never reports +18% growth when revenue actually fell 6%.
$5
Traceable Numbers: Verified Figures in Data Answers
Example session with this skill installed
Using sales.db, how did net revenue change in October compared with September?
- Read your context and instructions
- Compiled the traceable numbers
Summary
Net revenue fell 7.7% in October: 1,380,395.00 TRY vs 1,495,585.00 TRY in September.
Definition
Net revenue = sales - refunds - discounts, by business date (Europe/Istanbul), tax and tips excluded. Both months complete.
Checks
- Join fan-out: order count unchanged after joining payments (split bills grouped first).
- Reconciliation: POS net vs processor payouts within 0.3%.
- check_numbers.py: PASS, every number traced.
How these numbers were computed
- oct_net_revenue = 1380395.00 (pos.db: orders, refunds)
- sep_net_revenue = 1495585.00 (pos.db: orders, refunds)
- mom_pct = -7.7 ((oct / sep - 1) * 100)
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
AI agents are confident with numbers they never computed. They mix gross and net, double-count rows after a join, or write "up" next to a negative change. One wrong figure in a report ends trust.
What this skill does
- Defines the metric first: what is counted, the period, timezone and grain.
- Computes in code only: every figure comes from SQL or Python and is recorded as evidence with its source and query.
- Sanity checks: join fan-out, duplicates, nulls, partial periods, percent vs ratio, averages of averages.
- Reconciles: compares against an independent total (e.g. POS sales vs processor payouts).
- Verifies the draft: a bundled script lists every number as traced, untraced or wrong direction, and fails until all are fixed.
- Shows its work: a short "How these numbers were computed" section the reader can check.
Example
Draft: "Revenue was up 18% last month and up 7.7% vs September."
Check result: UNTRACED 18, WRONG DIRECTION 7.7 <- mom_pct (-7.7). The agent fixes the draft before anyone sees it.
Good for
Sales and finance reports, POS and accounting exports, KPI summaries, dashboards, any answer where a number will be acted on.
Requirements
Python 3.8+, standard library only. No network access, no dependencies.
Requires Python 3.8+. Small counting numbers (0-10, 12, 100) and years are skipped by default; use --include-small to check them. Numbers are matched by value, so a coincidental match with an unrelated recorded value counts as traced. Sanity checks and reconciliation are steps the agent performs with its own SQL or Python; the script does not connect to data sources.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 12 days ago
Requires
Permissions
Terminal: runs the bundled scripts/check_numbers.py with Python. Read files: reads the draft answer and the evidence file. Write files: appends computed values to .evidence/evidence.jsonl in the working directory. No network access.