SYNXLEVEL 6
SYNX L5 · AUTONOMOUS CONTENT FACTORY

It has to argue for the product
before it is allowed to build one.

L5 runs unattended. It scouts a market, costs it against a live marketplace, and states a case for one specific product — with citations. If the citations do not resolve against the data it actually scraped, nothing is built. Most runs do not ship, and that is the design.

FACILITY OVERVIEW — SYNX L5 CONTENT FACTORY
TO THE CUSTOMERINTAKEthe market, as it isSCOUT & COSTINGprove it is worth buildingGATEFEASIBILITYcan it, will it, is it wantedBUILD HALLmake itQUALITY GATESany one can refuseGATEPACK & LISTonly if it reconcilesthe line a run takes when it shipsconcept refused — rebuilt or abandonedsubject refused — the run ends, and the reason goes back to the scout
Live market data arrives and is queued for costing

The two red lanes are the ways out that are refusals rather than products, drawn as prominently as the main line because they are the common case. This shows how a run moves, not what any particular run is doing right now.

Intake
Pulls live market data — what is selling, at what price, and what is being taken up.
Scout & costing
Prices a niche against real listings, then argues for one product with citations you can open. If the case does not hold, nothing is built.
Feasibility
Three questions before anything is settled: can it be made, will it sell, is it wanted. Any one of them can end the run here, before a line of work starts.
Build hall
Writes the packing list, hands each item to the specialist for it, and inspects what comes back. Anything wrong goes back once, with the defect named.
Quality gates
Four independent refusals: safe to sell, different enough from what already exists, holds together, and contains what the listing claims.
Pack & list
Reconciles the packing slip against the listing. If they disagree, the product is not sold.
WHAT THE DECISION RESTS ON

Measured, not asserted

Nothing here is allowed to claim demand. It has to point at something real and public, and every citation is checked against the data actually gathered before a build is authorised.

What it sells for
Every niche is costed against what comparable products actually charge. A proposed price has to land inside that measured band — with a floor as well as a ceiling.
Whether anyone wants it
How many people chose a thing and kept using it. Held apart from price on purpose: take-up shows people want the problem solved, never that they will pay for the answer.
Every citation is checked
A source that is not in the data actually gathered fails the whole brief, and the quote is taken from the item rather than retyped.

A market that returns nothing is reported as unmeasured, never as empty. Zero competitors reads as a wide-open opportunity and would argue for building — the opposite of what an empty result actually means.

THE GATES

What we check before anything is offered for sale

The case for it
How do you decide what to build?
By arguing for it before it exists, with evidence you can check yourself.
  • Every claim points at a real, public listing or post you can open and read.
  • At least one citation has to come from somewhere people actually buy or install things — a question thread proves someone was stuck, not that anyone would pay.
  • Three independent citations at minimum, and the price has to land inside the measured band.
The legal read
Is it safe to sell, and safe to use?
We read the whole thing before it goes anywhere near a listing.
  • Trademark, copyright and liability, assessed across the entire document rather than a sample of it.
  • Where the risk belongs to the subject rather than the wording, we stop. Some things should not be sold as a download, and rewording them does not change that.
  • The finding is recorded alongside the decision, so what was caught and why is part of the record.
The value check
Is it worth what you are asking?
It has to be a different PRODUCT from the nearest thing already on sale — not a different description of the same one.
  • We compare what the buyer receives — formats, how many files, structure, options. Rewording a competitor changes nothing, because none of it is measured on wording.
  • The comparison is against the closest single product already for sale, not an average of the market. That is the one a buyer would otherwise choose.
  • Priced above the top quarter of a market, the listing has to name what it carries that the comparables do not — and the packing check then enforces those claims.
  • Four assets means four distinct things. Each is measured against the others before the package closes.
The packing check
Does it contain what the page says it contains?
Every claim on the listing is reconciled against the actual files before it can be sold.
  • A worksheet, spreadsheet or media file named on the page has to be in the archive.
  • Formats are checked against the files that shipped, not against what was planned.
  • If the listing and the package disagree, the product is not listed.
THE RUN RECORD

454 runs, 114 of them finished

29 Aug 2026 to 22 Sep 2026 — roughly 180.2 hours of production, and the record marks which runs got as far as a listing. 435 of the 454 are hash-chained to the run before — 434 links, none of them broken — and the record says why each run that did not finish stopped.

Some listings were later pulled, after a check found the package did not contain everything the listing named — which is the rule the packing gate now applies before a listing goes up rather than after. On 22 Sep 2026, 64 products were on the store; on 29 Sep 2026 it lists 86, free to download, and you can count them.

The run-by-run table below is an earlier export of the same ledger: 171 runs, 29 Aug 2026 to 7 Sep 2026. It has not yet been refreshed to the full count above.

130COMMERCE_ABORTEDa gate refused the product before it could be packaged — an empty file, a promise the package did not keep, a thumbnail that was a flat colour, or a legal screen that could not be reached and so was not treated as passed
94SCOUT_REFUSEDthe case for the product did not survive checking — a citation that did not resolve, too few independent ones, or a price outside the measured band. The factory has to argue for a thing before it is allowed to build it
78LLM_UNREACHABLEno model provider answered, so the factory stopped rather than guessing. Historic: it now fails over across providers and verifies inference itself rather than trusting an availability endpoint. It has not been the leading cause since
15FEASIBILITY_REJECTthe demand case or the deliverable did not hold up, so the run ended before any work started
14GATE2_EXHAUSTEDno concept survived the buildability check
5PIPELINE_EXHAUSTEDfive concepts attempted, none cleared
2GATE3_KILLEDstopped at the third gate
1PACKAGE_INCOMPLETEthe package was short of what the contract named
1UNKNOWNthe run stopped without recording why. One in 454, and it is listed rather than quietly dropped

Published because a factory that showed only its successes would be advertising. Every listing carries a timestamp and a hash chain, so the record can be checked rather than believed — the same reason the verifier exists.

See a governed run end to end

L5 is the factory. Level 6 is the governance — the proof chain, the routing, the constitutional validation. Today Level 6 governs the work it sends to L5 the same way it governs any pipeline it does not run: L5 accepts governed work only with a credential only Level 6 holds, Level 6's jobs take priority in L5's queue at each stage boundary, and if Level 6 stops waiting, the build is abandoned — never published. A job is sent to L5 at most once.

Talk to us about L5The governance underneath