Paghoria commissions original imaginative work — counterfactuals, conceptual fusions, short fiction — and certifies, as it is being written, that a human mind produced it. You license data made in the author's own language, signed at source, with a provenance that withstands scrutiny rather than asking to be taken on trust.
A working platform, certifying server-side, in production.
Neither a social scrape nor a synthetic generator can produce the equivalent.
The open web is saturating with machine-generated text, and models trained on that output degrade — the failure mode researchers call model collapse. As it spreads, data demonstrably written by a human becomes the scarce input, and its value rises with the scarcity. Generating fluent text was never the hard part; establishing that a person actually produced it is.
So we don't run an "is-this-human" classifier over the finished text — those are gameable, and they flag genuine authors as false positives. We certify the act of composition itself: written by hand, in our app, by someone responding in real time. The guarantee attaches to how the work was made, not to how it happens to read.
The certification covers how the work was produced — composed live, in the app — so there is no output classifier to defeat, and no real author wrongly flagged.
Each unit carries a cryptographic signature issued solely by us, which we confirm on request. It cannot be forged or recomputed by anyone else.
Every contributor licenses their own work, so the chain of rights is explicit and documented — precisely what you are expected to demonstrate downstream.
Authors write in their own language. Genuine creative data outside English is scarce; here it arrives labelled by language and ready to use.
Mid-session, an unscripted twist; the author responds on the spot, against the clock. We score that improvisation separately — creative behaviour under constraint, demonstrably live.
Each sample ships with a datasheet — collection method, quality distribution, provenance, rights — in the form a compliance function expects to receive.
From 2 August 2026 the EU AI Act acquires teeth for providers of general-purpose models: the penalty regime applies, and the Article 50 transparency obligations take effect. Providers are already required to maintain technical documentation, publish a summary of their training content, and operate a copyright policy.
None of this obliges you to buy certified data. But once you must account for where training data originated and whether its rights are clean, data signed at source and licensed by the people who produced it is considerably easier to defend than data whose provenance you can only assert.
For context: the separate deadline for high-risk deployment systems was deferred to December 2027 under the Digital Omnibus. The general-purpose model obligations and the Article 50 rules were not — they remain on the August 2026 calendar.
1,000–2,000 certified units in our strongest formats, composed to your brief: exclusive licence, indemnity, a datasheet per sample. Priced per unit, calibrated to your use case.
Fund a brief, and the majority of the budget goes to the people who write. You set the prompt and retain the certified output, rights clean.
Because each unit is mapped in idea-space, we can elicit certified human data in precisely the region you lack — something neither an expert marketplace nor a synthetic generator can target on demand.