Skip to content

Latest commit

 

History

History
102 lines (76 loc) · 7.08 KB

File metadata and controls

102 lines (76 loc) · 7.08 KB

Kashf vs Tabaqa — the "should we switch back?" decision (2026-07-11, 5 days out)

Written because the user asked: validate Kashf against the 5 AMAD criteria the way we did Tabaqa, and tell me honestly — is Kashf actually better? (worry: Tabaqa feels "like Sentry," a 2025 winner.) This file is the answer. Verdict: do NOT switch. Stay on Tabaqa.


The one-paragraph answer

Kashf is a good idea that was already killed for three structural reasons that are more true today, not less. It is not built (M0 skeleton) with 5 days left, while Tabaqa is live and demo-ready. And the "resembles a past winner" fear is pointed at the wrong product: Tabaqa does not really resemble Sentry — but Kashf strongly resembles QuantumBoard (2025 1st place), which is exactly why the team retired the agent-debate framing back in May. Switching to Kashf to escape a resemblance would run into a stronger one.


The 5 official criteria (AMAD 2026, no published weights)

الابتكار · التطبيق التقني · تحليل البيانات · تجربة المستخدم · قابلية التنفيذ الفعلي

Scored 1–10, judged the way the rubric is judged. Kashf scored as it stands today (skeleton), with its ceiling-if-fully-built in brackets.

# Criterion Kashf (today [ceiling]) Tabaqa Why
Innovation & Creativity 5 [6] 7.5 Kashf's "7 agents debate a loan live" is (a) the QuantumBoard 2025-winner recipe, (b) already shipped in substance by Abwab.ai (Saudi AI SME underwriter, Arabic, SAR 100bn target), (c) a published 2025 mechanism (arXiv PMADS/CreditXAI). Novelty is theater, not substance. Tabaqa's applicant-facing transparency + bank/wallet fusion survived ~15 refutation passes — "almost nobody on earth."
Technical Implementation 4 [7] 8.5 Kashf = M0 skeleton; a half-working multi-agent demo in front of an Apple-trained UX judge is a real risk. Tabaqa = live app + live API w/ sandbox keys + ALLaM in production + offline smoke test; judges drive it themselves.
Data Analysis (underrated axis) 5 [6] 9 Kashf's "analysis" is LLM debate narration — soft on hard metrics. Tabaqa ships an in-app Model Card: Berka +0.203 AUC lift, UCI negative control, AlfaBattle 963k real apps, TSTR 96%. Likely the strongest data story in the room.
User Experience 4 [7] 8.5 Kashf's debate view is visually cool but B2B credit-officer-facing and not built. Tabaqa = bank-grade bilingual RTL, 40-sec live reveal (4,000→10,000, DECLINE→APPROVE), mobile-safe, decision-memo cockpit, printed Arabic attestation + QR. Apple-judge bait.
Practical Feasibility (Alinma-judge axis) 4.5 [6] 7.5 Kashf dies on WRONG-BOTTLENECK: the binding KSA SME constraint is collateral / risk-appetite, not scoring capability — a better score yields a more confident "no," not a new loan. Plus "why not just call Lean+Synapse / Abwab?" Tabaqa: no-new-license route, SAMA-DBR in-engine, compliance receipt in hand, real inclusion KPI (11M unscorable).
TOTAL /50 22.5 [32] 41 Decisive gap even at Kashf's built-out ceiling.

The three kill-shots on Kashf — now worse than when we dropped it (2026-06-10)

  1. Wrong-bottleneck (structural, unchanged). Scoring isn't the Saudi SME-lending constraint; collateral and risk-appetite are — that's why the SAR 123bn Kafalah guarantee book exists. SIMAH already ships SME scoring. A sharper committee produces a more articulate rejection.
  2. Already solved / no moat (now much worse). In June this was "Lean+Synapse shipped the pipe." Today it's also Abwab.ai — a Saudi-built AI SME credit engine, Arabic-native, integrates Lean+Tarabut, ~$400M processed, eyeing SAR 100bn in 2026, just partnered with Tuum for end-to-end GCC SME lending. Kashf's entire "Saudi + Arabic + multi-source + AI underwriting" moat is occupied by a funded incumbent. Kashf's only unshipped bit is the debate theater.
  3. Not built, 5 days out. kashf-api is an "M0 skeleton" commit. Tabaqa is live on Vercel with a metered API, ALLaM narration, frozen model card, and a rehearsed judge script. You cannot rebuild that surface area for a different product in 5 days without torching your best asset.

The Sentry worry — the premise is backwards

What Sentry was (2025, likely 3rd): Yousef Al-Mutairi's project — a monitoring / sentinel agent, fraud/compliance-flavored (RegTech shape). A watchdog.

What Tabaqa is: a credit-inclusion score for the 11M Saudis a bank can't rate — Open Banking track, a completely different user, problem, and output. The only thing that even rhymes with Sentry is Tabaqa's secondary "درع طبقة" fraud-flag. That's a minor feature, not the headline. If it's causing confusion, de-emphasize or drop it — the score-turned-toward-the-applicant is the whole pitch.

Three things follow:

  • Tabaqa ≠ Sentry in any load-bearing way. Different track (Open Banking didn't exist in 2025), different problem, different product shape.
  • Resembling a past winner isn't disqualifying anyway. 2 of 3 2025 podium teams were agentic; judges rewarded the recipe. Different year, different (likely new) judges, new track.
  • The irony: the product that actually resembles a 2025 winner is Kashf, not Tabaqa. Kashf's hero moment — "watch the agents debate live on real data" — is QuantumBoard (2025 1st). The team already spotted this in May and retired the investor-agent version for exactly this reason ("QuantumBoard but Arabic"). Switching to Kashf to dodge a resemblance walks into a bigger one.

So "it looks like a past winner" is an argument to keep Tabaqa, not to leave it.


If what you actually miss is the agentic wow

Kashf's real temptation is the theatrical demo. You don't need to switch products to get it: graft a lightweight "committee narration" on top of Tabaqa's existing reveal. ALLaM already narrates the decision — frame the reason-codes as 2–3 named voices (cash-flow, affordability, risk) converging on the score. You get the QuantumBoard-style spectacle without the wrong- bottleneck, the Abwab collision, or a 5-day rebuild.


Recommendation

  1. Stay on Tabaqa. It wins 4 of 5 criteria outright and ties on none that matter.
  2. Do not re-open the Kashf question again — it's now been killed twice on the same structural grounds, and the market moved against it (Abwab at SAR 100bn).
  3. If you want more demo drama, add the committee-narration layer on Tabaqa's reveal.
  4. Keep the planned Jul 14–15 competitor re-check (Abwab / Tamam×FICO / Qarar / Tarabut-Servable).
  5. Consider dropping or hiding درع طبقة if it's what triggered the "this is Sentry" read.

Sources: Abwab.ai (thefintechtimes.com, sharikatmubasher.com, abwab.ai); Lean×Synapse (openbankingexpo.com, wamda.com); AMAD official criteria (docs/research/01); Kashf validation (project_kashf_validation, 2026-06-02); Tabaqa NORTHSTAR + COMPETITION.