API —
Audit chain —
Reciprocity: Silver · illustrative
Tier —
Queries today
0
of daily quota
This hour
0
of hourly quota
Success rate 14d
—
scored / total
Errors 14d
0
denials incl. rate limits
Needs your attention
—Quota usage
Fixed UTC windowsThis hour—
Today—
Over-quota requests return 429 with a Retry-After header — build your backoff on it. Idempotent replays never consume quota.
Outcomes 14d
—
success
— scored
— errors
— total queries
Reciprocity Illustrative — shadow tiers
give data, get the networkSILVER
Silver contributor — sample tiers pending commercial terms
Mature bureaus run on reciprocity: you see the network you feed. Your contribution ratio (records contributed ÷ scored queries consumed) sets your tier; Gold unlocks off-us early-warning alerts, full benchmark depth and 10× Watchlist capacity. The live ledger switches on with the contribution API at onboarding.
API activity — last 14 days
Auto-refresh
Successful queriesErrors / denials
Recent queries
| Time (UTC) | Subject | Purpose | Status | Score |
|---|
Try the API — shared public sandbox
Live call against the sandbox institution——
Monitored subjects
0
borrowers you're tracking
Active alerts
0
need attention
Avg change since added
—
across your watchlist
Portfolio risk
—Alert rules
applied live to your watchlistWatchlist Illustrative
| Subject | Band | 6-mo trend | Change | Alerts |
|---|
Monitor borrowers you hold and get flagged when their risk changes — a default reported by another lender, a score drop, or a new inquiry. Only a bureau sees across lenders, and you only get default alerts because lenders contribute. Illustrative preview on sample subjects — live monitoring re-queries each subject on the monthly cycle and is billed per scored query.
API logs
| Time (UTC) | Subject | Purpose | Status | Result | Score |
|---|---|---|---|---|---|
| Loading… | |||||
Every row is one immutable entry in Sheba's hash-chained audit log — the same record the NBE can inspect. Click a row for full detail.
Data entries — subjects you've queried
| Subject | Queries | Last queried (UTC) | Last status | Last score |
|---|---|---|---|---|
| Loading… | ||||
Subjects are pseudonymous references — Sheba never exposes names, ID numbers or phone numbers through this dashboard.
Errors & denials
Self-diagnosis: consent, quota and validation failures| Time (UTC) | Subject | Purpose | Result code |
|---|---|---|---|
| Loading… | |||
consent_required — the consent token is missing, revoked, expired or out of scope. rate_limited — you hit your tier quota; honour the Retry-After header. invalid_request — malformed body or a raw identifier was sent (only pseudonymous subject_ref is accepted).
API keys
——
Full keys are shown once, at issuance, and never stored by Sheba — we keep only a SHA-256 hash. Keys authenticate server-to-server calls only; this dashboard never uses them.
Limits & controls
Rate limit (hour)—
Rate limit (day)—
IP allowlist—
Key last used—
Environmentssandbox production — at go-live
Production partners additionally get mTLS and an enforced IP allowlist before live subjects are queryable. Key rotation stays operator-run by design — no self-serve key minting on a credit rail.
Team
Access is provisioned by Sheba during onboarding| Role | Since | |
|---|---|---|
| Loading… | ||
Need to add or remove a teammate? Contact us — self-service team management ships after the pilot phase. Every dashboard view by any member is written to the access audit.
No charges
Your pilot is free. Everything on this page is a shadow statement — what your usage
would cost at the illustrative pilot price card, so a production agreement holds no surprises.
Amounts become payable only under a signed commercial agreement, invoiced in birr and settled by bank
transfer. Every figure is derived from your own hash-chained audit log and can be independently verified.
Adverse-action reason codes
Every score response carries the factors that drove it, each with an English and Amharic label. Map these into your own adverse-action notices — the code is authoritative; the labels are provided for display. NBE CRB/02/2019 requires a local-language decline reason.
| Code | English | Amharic (አማርኛ) | Category |
|---|
Test persona library Sandbox
every integration path, documentedLike Stripe's test cards, but for credit: each persona documents one integration path and the exact response contract your integration must handle. "Load in console" pre-fills the sandbox console; the console always shows the live response.
Personas marked partner sandbox are seeded in the private integration sandbox you get at onboarding (the shared public sandbox exposes only the happy-path subject). Note thin-file: a thin file returns a low-confidence Building band — never a fabricated "good" score; your UI must surface that honestly. All personas are synthetic; no real people.
Audit-trail integrity
—…
Verifying…
Every credit query — approvals and denials — is one row in an append-only, hash-chained log. This verdict is a live recomputation of the entire chain, genesis → head, run at the moment you opened this page. Any edited, deleted or inserted row would break it.
Consent & query governance
aggregates only — zero borrower dataModel governance
—Thin-file policyunknown is never scored as good
Default orderingknown-default < unknown < clean
Weights—
Built for NBE examination: the same governance telemetry we will present to the National Bank of Ethiopia — computed live from the rails that serve your queries, not staged for a demo. Built to comply with NBE Directive CRB/02/2019, pending NBE licensing.
Month to date
—
queries
Billable (scored)
—
denials & throttles are free
Forecast run-rate
—
at the current daily pace
Indicative cost Illustrative
—
shadow pricing
Query volume — last 30 days
Scored
Denied
Throttled (429)
—
Purpose split maps to CRB/02/2019 art. 7.14 permitted purposes. Idempotent replays bind to the original audit row — retries never appear here and never consume quota.
Quota & headroom
—Load profile — hour of day
30-day total · UTC—
Usage alerts Saved on this device
server-side email/webhook alerts — roadmap
Hourly quota utilisation above %
warns before your integration starts seeing 429s
Monthly queries above
volume guard for the month (checked against the run-rate forecast)
Indicative monthly cost above ETB
based on shadow pricing — illustrative until a signed price card
Every figure on this page is derived from the immutable, hash-chained audit log — the same records that drive your statements, so Usage and Billing can never disagree.
Score migration — your monitored book Illustrative · synthetic sample
Q2 → todayRows = band a quarter ago, columns = band today. The diagonal is stability; below-diagonal cells are deterioration your credit team should see first. Live version computes this from your monitored subjects' recomputation history.
Your book vs market composite
synthetic benchmarkAverage score691 · market 668
Share ≥ Good (670+)58% · market 47%
Default-flag rate4.2% · market 6.1%
Blue bar = your monitored book; black tick = market composite. Only a bureau can show this — the composite comes from the whole contribution network (synthetic until the network scales; the label flips the moment it's real).
Vintage curves
avg score by months on bookJan cohortMar cohortMay cohort
Newer vintages starting higher = acceptance quality improving. Illustrative sample data.
Sheba Score — model card sheba-score-v1.0.0
segments: urban · farmer · smePayment history35%
Account activity25%
Savings history20%
Identity verification12%
Engagement / literacy8%
Urban-segment weights; the farmer segment reweights toward savings, repayment and cooperative standing. Weights are declared priors pending actuarial validation on repayment outcomes — published precisely so this claim is checkable.
Fairness & testing
Adverse-impact ratio · ≥580 cut0.924 (≥ 0.80 pass)
CalibrationProvisional
Fairness / calibration harness29 / 29 tests
Score regression suite21 / 21 tests
Golden vectorsub_9f3a → 724
Policy, stated plainly: unknown is never scored as good; known-default < unknown < clean; no output may drive a binding lending decision before validation. Enforced in code and tested — not aspirations.
Version history
v1.0.0 — live
deterministic 5-factor model · 300–850 · reason codes EN/አማ · confidence field
v1.1 “B-lite” — staged, pending sign-off
reported-default floor (known-default < unknown) + bank/MFI payment provenance · fairness re-run complete
v2 — roadmap
ML ensemble + agri signals · only after outcome data and the NBE conversation
Every model change lands here before it scores anything — for your model-risk team.
Evidence pack — audit export
watermarked “pre-licence · shadow data”Ethiopian lenders answer to the NBE themselves. This pack is evidence that every credit check you made through Sheba was consented, purpose-bound and tamper-evidently logged.
Webhooks Roadmap — ships with notification infra
watchlist.alert— default reported / score drop / new inquiry on a monitored subject
statement.generated— monthly shadow statement ready
usage.threshold— quota / spend alert crossed
model.version— model card changed (for your model-risk team)
Designed now so you can plan integrations; delivery needs signed payloads, retries and a dead-letter queue — an honest “not yet”, not a fake toggle. Until then, alerts live in the Command Center inbox.
Pilot onboarding checklist
Progress is saved on this device. The full 14-day onboarding guide is in your pilot pack.