How we fact-check our quick-read articles

The whole system rests on the loop, not on promises:

1. 1

   Each quick-read article has its own fact-check

   Every quick-read article gets its own report at /quick-reads/<slug>/fact-check.

2. 2

   Readers verify by clicking through to the parent

   A reader who wants the full case citation or the unrounded statute number clicks the 'Read the full-length version' affordance — the parent article carries the primary sources the quick-read article compressed.

3. 3

   Readers who find drift tell us

   If a quick-read article's compressed claim doesn't match the parent's stated rule, that's a single click to flag.

4. 4

   We re-check what's flagged

   And update the quick-read + report, advancing the verification date.

It is: a reproducible, dated check that each verifiable claim in a quick-read faithfully compresses what the parent article / research actually says, with no drift in meaning.

It isn't: primary-source verification of the underlying statutes (that lives upstream at the research surface), proofreading, or endorsement of the parent's editorial framing.

Quick-reads are translations: from the full article's ~3,000 words to the quick-read's ~250. Translation can preserve meaning or distort it. The fact-check asks: does this compression preserve the parent's meaning?

A faithful summary is still a new claim. “California requires meal breaks” sounds like a compression of the parent's “California requires a 30-minute unpaid meal break after 5 hours” — but the shorter version drops the duration, the pay status, and the trigger threshold. Three different readers walk away with three different mental models, none matching the law.

The verification asks each of these about every compressed claim. Each drift mode names a specific way a faithful-looking summary can be unfaithful.

• Dropped qualifier. "Generally," "in most cases," "after X hours" — compression that drops a qualifier turns a conditional rule into an absolute one.

• Conflated rules. California meal-break and California rest-break are two distinct rules. Compressing them as "California breaks" turns two rules into one.

• Inverted hedge. The parent says "may," the quick-read article says "must." Or the parent says "in some states," the quick-read article says "in most states."

• Hallucinated specifics. A statute number, a case name, a dollar amount that sounds plausible but isn't in the parent. Compression pressure invents fillers; the fact-check catches them.

• Scope drift. The parent's claim applies to non-exempt employees in California; the quick-read implies it applies everywhere. The scope shrank or grew during compression.

Each verifiable claim in a quick-read is categorized under one of these. The translation- fidelity check walks each one against the parent.

For each compressed claim in the quick-read:

1. Locate the parent statement. Find the paragraph in the parent article / research that this claim compresses.
2. Walk the five drift modes. Did the compression drop a qualifier, conflate rules, invert a hedge, hallucinate a specific, or shift the scope?
3. Trace to the primary source. The parent's fact-check has already verified the primary; we're checking the compression matches the parent's statement, which traces to that primary.
4. Record the result. Claim text, parent-source URL, status, verification date.

The trust signal is the artifact itself.

1. Mark every unverifiable claim. List them with ⓘ status; don't quietly omit.
2. Show every parent-source URL. The reader should be able to click through to the parent for the full citation.
3. Date everything. Per-claim, per-quick-read, per-methodology.
4. Mark drift as Issue. ✗ Issue means the quick-read needs revision before shipping the next iteration.

• Rubber-stamping. A check that always returns ✓ Verified. If a process can't produce ✗ Issue outcomes for compression drift, it's not a check — it's a stamp.
• Re-verifying primaries. The quick-read fact-check is a translation check, not a primary-source check. Primary verification lives upstream at research. Doing both here is duplicate work that pushes the publish gate slow without adding trust.
• Hand-waving compression. 'It captures the essence' is not a verification. A faithful summary preserves every load-bearing qualifier; the check is concrete, not impressionistic.
• Silent omission. Pretending an unverifiable claim doesn't exist. Reports list every claim — including the ones we couldn't verify, with ⓘ Unverifiable status.