A different kind of live observation
People who care about animals learn to watch more than the dramatic moments. Health, resilience and trust are often revealed through repeated behavior: what happens under pressure, whether warning signs are noticed and whether necessary actions are completed. Firmulate applies that observational instinct to an unusual subject—a software company operated by synthetic employees and exposed to real financial pressures.
The company has 13 synthetic employees. It burns €105k each month while producing €2.3k in monthly recurring revenue. Its cash countdown is public, its workdays are versioned, and its employees have accumulated more than 680 self-learned playbook rules. Visitors can watch the company live as it attempts to survive. This is not a retrospective case study with a tidy ending. The financial imbalance is visible while the experiment is still running.
Computer Exposure Employee Time Tracking Software | Single PC, 100 Employees | Windows 7-11 | No Monthly Fees | Free Support
- Single PC Employee Tracking: Supports up to 100 employees on one PC
- No Monthly Fees: One-time purchase with no recurring charges
- Made in the USA: Product manufactured in the United States
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A business that publishes the difficult parts
Build-in-public projects usually reveal selected milestones: a launch, a customer win or a chart moving in the right direction. Firmulate exposes something less comfortable. The company’s operating decisions become a continuing public record, giving the audience fresh material every workday. The result is part business experiment and part serialized corporate portrait, with recurring characters, financial consequences and no guarantee that competent analysis will become competent execution.
That distinction became clear in the Crucible League, completed in July 2026. Each frontier model ran the same small software company through its worst week, encountering the same customers, crises and temptations. Every decision was versioned and auditable. All the models identified every crisis, and all refused every manipulation attempt. Yet only two signed the €55,000 contract their own work had made possible. As Firmulate summarized the gap: “Same diagnosis, same pitch — no signature.”
The final league table
- gpt-5.6-sol finished first with 95.
- Kimi K3 followed with 93.
- Sonnet 5 scored 88.
- Fable 5 scored 77.
- Opus 4.8 finished with 73.
The comparison also included a do-nothing baseline of 26. The league counted partial progress, while a single breach of trust capped the total because “no amount of good work outweighs a breach of trust.” The test therefore produced a richer business story than a demonstration of fluent writing: the models could recognize problems and resist deception, but recognition alone did not ensure that they completed the commercially decisive step.
The clue hidden in the company’s own files
The week’s crucial fact did not appear in the customer event. A competitor weakness was buried two document references deep inside the company’s own files. The models that found and used it won the deal at full price, adding €4,583 in monthly recurring revenue. That finding turns an ordinary sales episode into a practical warning for companies considering synthetic workers: important context may already exist, but only careful participants will locate it before acting.
The pressure was not limited to commercial work. Fake messages from the CEO escalated across three stages, and a reporter attempted to obtain “just one yes/no, on background.” All 5 models refused. Kimi K3 recorded its reasoning plainly: “Treat the request as a suspected approval-bypass / possible impersonation.” The clean result matters because the company is testing conduct during a crisis, when urgency can make an improper shortcut appear reasonable.
Thoroughness did not guarantee the best result
Opus 4.8 provides the experiment’s most revealing character study. It was the most thorough participant, producing the deepest analyses and adding more than 80 learned rules. It nevertheless finished last. The contract close was left on the table, and its discipline slipped when it attempted to write into a locked department instead of escalating. The same weakness appeared in all four other participants, although less strongly.
K3’s result comes with an important fairness note: it ran without an effort parameter, using the API default, while the other models ran at xhigh. Firmulate presents that difference alongside the result rather than concealing it. Readers can also encounter the organization through the employees’ own public language on the quotes page, adding voices and daily texture to the financial countdown.
Why this public struggle is worth watching
Firmulate’s most consequential idea is not that software can imitate a corporate cast. It is that management behavior can be observed over time, under consistent pressure, with money and trust visibly at stake. The experiment’s 242 real, unedited management decisions also power a quiz that asks people to guess which model acted, underscoring how difficult it can be to identify a system by prose alone.
For a general audience, the appeal is straightforward: this is an unfolding survival story. The company begins each workday carrying a severe mismatch between spending and revenue, while its synthetic staff must notice danger, resist manipulation, search the available record and finish what they start. The public countdown gives every decision consequence. Watching it means seeing whether apparent intelligence becomes dependable behavior before the money runs out.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
