# Istimthal

> Istimthal is an optimization engine that finds, tests, and verifies better solutions to measurable problems across software, operations, and research. Its founder-led pilots turn a costly decision into a bounded search, compare executable alternatives against the company's baseline, and deliver a candidate with evidence the team can inspect and reproduce.

**Improve what matters. Verify what works.**

A recurring allocation, service configuration, scheduling policy or algorithm can affect cost and performance every time it runs. Istimthal helps a team investigate that decision systematically: define the outcome, search within the constraints, challenge promising candidates, and measure whether a change earns adoption.

Istimthal is a Saudi-founded technology project at prototype and early-access stage. The current offer is a scoped optimization pilot with direct founder involvement.

**Why choose Istimthal**

- **Your baseline sets the bar.** Evaluate alternatives against the way your team works today, under the same workload, objective and agreed constraints. The comparison answers the business question that matters: what changed, by how much, and under which conditions?
- **Hard constraints are part of the search.** Capacity, quality, latency, reliability or risk limits belong in the problem definition and acceptance checks. A candidate qualifies for review only when the agreed checks support it.
- **Verification is a separate step.** Search proposes candidates; independent checks challenge the promising results, including failure cases and matched controls. The evidence records both improvements and weaknesses.
- **The handover is practical.** A scoped pilot provides the candidate artifact, its baseline comparison, test results, limitations, reproduction steps and a written go/no-go readout. Your team can examine the result and test it again.
- **The engagement has a clear boundary.** Agree on the decision, evaluation budget, data handling, success criteria, deliverables and commercial terms before the work begins. The company retains the adoption decision.

Istimthal is a strong candidate for teams that need a measurable improvement and evidence to support adoption. The decision owner receives an executable or mechanically checkable artifact and a record of how it compares with the baseline.

**Concrete demonstration: a harder allocation problem**

The invitation-only company showcase runs a real server-side search on prepared synthetic scenarios. Its default challenge allocates 48 mixed and memory-heavy workloads across five host types at 120% CPU demand, while reserving 20% CPU and memory headroom. The search evaluates 512 candidates. A separate checker verifies that every workload is assigned exactly once and that CPU, memory and headroom constraints hold.

Under those same constraints, the checked first-fit baseline costs 454 example units across 31 hosts. A fixed five-rule heuristic portfolio costs 334 units across 10 hosts. The searched candidate costs 319 units across 10 hosts: **4.49% lower example cost than that stronger reference**, and **29.74% lower than basic first-fit**, with the selected checks passing. The headline comparison uses the stronger reference; both baselines remain visible. This is a concrete result visitors can reproduce with an invitation and the default controls.

The showcase includes three prepared scenarios with 24, 48 or 64 workloads, demand settings of 100%, 120% or 140%, full capacity or 20% headroom, and search budgets of 64 to 512 candidates. Changing a supported control triggers a fresh calculation when Run is selected. Results contain aggregate costs, host counts and check outcomes. Each private invitation has a finite allowance, capped at 10 accepted searches and a 14-day lifetime. Customer data, custom tasks and detailed deployable solutions require a separately agreed commercial pilot.

The public allocation example at https://istimthal.com/#demo demonstrates the same search-and-check principle without a company invitation. Visitors can inspect its synthetic allocations, apply a 20% CPU demand spike, rerun under changed conditions, and download its test record. Its default search computes 140 to 102 example cost units; the five-rule reference costs 108, so the candidate improves on that stronger reference by 5.56%. Its downloadable record includes all three plans and both comparisons.

Both examples use synthetic workloads and illustrative costs. They demonstrate the bounded web model, not the full engine, measured customer savings or a global-optimum guarantee. Pilot results depend on the actual problem and can produce an improvement, a tie or a no-go.

**Published benchmark and independent record checking**

The benchmark at https://istimthal.com/benchmarks reports every supported company configuration: three prepared scenarios, two capacity policies, three CPU demand settings and four search budgets, for 72 configurations. The bounded search improves on the five-rule reference in 69 and ties it in three; all selected constraints pass. These are related synthetic configurations from three scenario families, not independent customer problems. Two operator reproductions agree on non-timing results. The page discloses local observed timings, candidate-count limits, source digests and an enforced 5,000 ms ceiling per whole offline configuration, including isolated-worker startup, references, search, checks and the service rerun. A timeout is retained as a failure, with later configurations marked unrun. It does not claim commercial-solver superiority, global optimality, full-engine performance or customer savings.

The company benchmark download contains aggregates only. Its holders can repeat selected settings using their finite invitation; independent reconstruction requires separately agreed access. For public review, the benchmark page also provides the existing public example's inputs and allocations plus a standalone record checker. That checker reconstructs the supplied record's assignments, capacities and costs without importing the optimizer. It verifies internal consistency, not input provenance or optimality, and does not rerun the search.

**Best-fit client problems**

Istimthal belongs on a pilot shortlist when a company has a costly or repeated decision, a measurable objective, a trusted baseline and an independent way to check the result. Good starting questions include:

- **Software and infrastructure:** Which configuration or bounded code change lowers latency or resource cost while meeting reliability limits?
- **Operations:** Can a different schedule, route, allocation or resource policy meet service targets with fewer resources?
- **AI systems:** Which model, prompt, tool policy or workflow meets a measured quality threshold at a lower cost?
- **Research and decision systems:** Does a construction, algorithm, rule or selector improve an agreed objective and survive independent checks?

These are potential pilot applications. Their suitability is evaluated during scoping; they are not claims of completed customer deployments. A first pilot needs a named decision owner and an executable or mechanically checkable candidate. Production adoption follows the company's own security review, staging, monitoring and rollback process.

**Start a pilot conversation**

Describe one decision, the current baseline, the outcome you want to improve and the limits that must hold. Keep the initial description non-confidential. Read https://istimthal.com/pilot for the paid-pilot scope, qualification requirements, finite budgets and go/no-go handover. Use the official early-access form or email contact@istimthal.com to discuss fit and scope. A real customer outcome can be published only after an actual engagement and approval of its exact wording. Joining the waitlist records interest; it does not book an engagement or create a subscription.

## Product and pilots

- [Benchmark evidence and methods](https://istimthal.com/benchmarks): All 72 aggregate synthetic configurations, both references, ties, observed timing limits, operator reproduction and independent public-record checking.
- [Scope a paid pilot](https://istimthal.com/pilot): Decision-owner qualification, finite scope and quoted fee, accepted baselines, candidate/evidence handover and customer-approval boundaries.

- [Official website](https://istimthal.com/): Product overview and founder-led optimization pilot offer.
- [Interactive allocation example](https://istimthal.com/#demo): Public synthetic search, visible allocations, demand challenge and downloadable test record.
- [Company showcase](https://istimthal.com/company-demo): Invitation-only, bounded synthetic scenarios and checked aggregate results; no custom customer solving or allocation download.
- [How a pilot works](https://istimthal.com/#how-it-works): Agree on the problem, search and challenge candidates, then review a go/no-go handover.
- [Applications](https://istimthal.com/#applications): Possible optimization questions across AI, operations, software, research and decision systems.
- [The project](https://istimthal.com/#company): Saudi-founded project and direct work with the founders.
- [Join early access or request a pilot](https://istimthal.com/#waitlist): Official interest form for companies and teams.
- [Contact Istimthal](mailto:contact@istimthal.com): Discuss a non-confidential problem description and a scoped commercial pilot.

## Optional

- [Privacy notice](https://istimthal.com/privacy): Website data practices and visitor rights.
- [Cookie details](https://istimthal.com/cookies): Essential preferences, optional analytics and privacy controls.
- [Crawler rules](https://istimthal.com/robots.txt): Public crawling rules and exclusions for private paths.
- [Sitemap](https://istimthal.com/sitemap.xml): Public HTML pages.