Kairos.
Try the agent
FINAL RUN — 2026-09-13

250 REQUESTS.
EVERY ROW VALIDATED.

The deterministic engine read every dataset row, resolved every conflict, simulated the 90-day horizon for every candidate plan, and wrote one decision per request — then an independent 22-point validator checked the whole table.

250
requests decided
25,342
financial events read
275
user profiles rebuilt
215
messages interpreted
3
languages understood
16
receipt images extracted
4
currencies converted (dated FX)
90
days simulated per decision
22
final validation checks passed
0
model calls needed (deterministic core)

AFFORDABILITY STATUS MIX

affordable_now
29
affordable_with_plan
32
affordable_later
65
not_affordable
124

RECOMMENDED METHOD MIX

full_payment
30
installments
31
wait
4
not_recommended
185

EVIDENCE THE AGENT READS — 16 images, OCR + vision extraction (click to inspect)

LIVE DECISION EXPLORER — the real output.csv, one decision per request

THE 22-POINT FINAL VALIDATION

  1. exactly 250 output rows plus header
  2. every request id appears exactly once
  3. no unexpected request ids
  4. column order exact
  5. no unintentionally blank decision fields
  6. 0 ≤ amount_safe_to_pay ≤ requested_amount
  7. allowed enum values only
  8. payment-plan syntax valid
  9. payment dates chronological
  10. partial plans contain exactly two payments and sum correctly
  11. installment plans match supplied offers exactly
  12. recommended payments all payable safely
  13. minimum balance preserved throughout the forecast
  14. spending changes target only eligible flexible events
  15. no more than three spending changes
  16. earliest full-payment dates are genuinely the first safe dates
  17. affordable_now uses request_date
  18. not_recommended plans use none
  19. every explanation agrees with the structured decision
  20. usage report corresponds to the final run
  21. chat transcript maintained per AGENTS.md
  22. no API keys or secrets in any artifact

SUBMISSION ARTIFACTS

output.csv250 predictions, 8 columns
code.zipsource · prompts · tests · evaluation
log.txtfull build transcript
usage_report.mdmodel usage + cache stats