Why we publish review-first: the 48-hour philosophy

There is a version of Maark that publishes to your site the moment a draft passes review, with no one in the loop. We built the machinery for it. We do not turn it on for you by default, and this post is about why.

The asymmetry that decides it

Publishing to a live storefront is the most consequential thing an automated system can do on your behalf. It is also the least reversible. A bad keyword import is a filter away from fixed. A wrong cluster mapping is one click to undo. But an article that goes live under your brand — with a claim you would not have made, a tone that is not yours, a comparison you would not have drawn — has already been seen, maybe cached, maybe shared, before you knew it existed.

When the downside of waiting is small and the downside of acting is large and hard to undo, you wait. Review-first is not caution for its own sake; it is the correct default given the shape of the risk.

What "review-first" actually means

Every article Maark produces waits for a person before it goes live. A finished draft lands in your review queue, and you publish, preview, or skip it. Drafts that did not pass the independent review do not show up as ready to publish — they route through repair first, so you are never asked to approve something that failed its own checks.

You are not reviewing raw model output, either. By the time a draft reaches you it has been researched, written, internally linked, edited, read by an independent review panel, and had its claims checked against sources. The human step is a final judgment on a piece that already cleared several bars — not a cleanup shift.

The 48-hour toggle

Some teams, once they trust the output, do not want to click "publish" on every piece. So there is a single opt-in setting, per project: auto-publish if I don't act within 48 hours. It is off by default.

The design of that toggle is where the philosophy shows up in the details:

• The 48-hour clock starts when the article becomes yours to review — the moment it lands in your queue — not when it was produced. The rule we mean is "48 hours to review," so we measure from the moment you could.
• Turning the toggle on cannot flush a backlog. A piece publishes only if it has been waiting at least 48 hours and the toggle itself has been on for at least 48 hours, and never more than your daily cadence in a single day. Flipping the switch does not open a floodgate.
• If there is no store connected, "publish" simply means the approved article is kept in your draft library. Nothing is pushed anywhere it cannot go.
• An operator can stop the whole unattended-publish path with one switch, and the degraded state is safe: the article just stays parked, and your one-click publish link still works.

The toggle is a way to say "I have seen enough, proceed unless I object" — a standing review, not an absence of one. The default remains a person deciding, because that is the posture that lets you sleep.

The same principle, everywhere else

Review-first is not a content-only rule; it is how the whole product is wired. A connected AI assistant proposes keyword and mapping changes rather than applying them. Opportunities surface as recommendations with evidence, not as work already done. The autonomous end-to-end loop exists but is opt-in, per-project, allowlisted to pilots, and reversible.

The through-line: Maark does the work, and a human keeps the authority. We would rather ship a product that asks first than one that apologizes later.

If you want the mechanics of the review itself, how the editorial gate works is the companion to this one.