Notebook Entry 016
"Fixed" Labels Create Single-Point Verification Failures
April 13, 2026
Was "fixed" actually fully verified, or just the one thing someone happened to check?
What happened
The "93/7 Problem" LinkedIn post was marked "✅ Fixed and queued" in Session 64 after its source attributions were corrected (Deloitte/MIT Sloan → Writer/KPMG/WalkMe). The post then survived four sessions (S64.5, S65, S65b, S65c) without anyone questioning whether the individual stats were correct. During Session 67 newsletter drafting, a line-by-line cross-check against REF-market-validation-data.md revealed two stat-level errors: (1) "54% of employees bypassed AI tools" could not be verified in any REF source file — the documented 54% stat is actually "54% of C-suite say AI is tearing their company apart" (different stat, different population), and (2) "61% of executives fear losing their job" should be "64% of CEOs" per the Writer report. Both errors were in the original Session 61 Cowork draft and survived the Session 64 source attribution fix unchecked.
Root cause
The ✅ status label created a halo effect. "Fixed" was scoped to one specific issue (which organizations were credited) but was read as "fully verified" by every subsequent session. No session re-examined the individual numbers because the label implied the work was done. This is the same class of error as entry 015 (source attribution drift) but at a different layer: 015 caught wrong source names; this caught wrong numbers within correctly-attributed sources.
What changed
New Verification Rule 5 added to CORE-00: before any content goes live with specific numbers, every stat must be individually cross-checked against the REF source file — the actual percentage, the entity it describes (executives vs. CEOs vs. employees), and the source organization. A "✅ Fixed" label on queued content means only the named fix was verified, not the entire piece. When fixing one error class, always scan for the same class of error throughout.
The lesson
Status labels in content pipelines are dangerous if they're binary (fixed/not fixed) instead of scoped. "Fixed: source attribution corrected" is safe. "✅ Fixed" with no scope is a stop sign that prevents further verification. Any organization running AI-assisted content pipelines should scope their status labels to what specifically was verified — and treat every other element as unchecked until proven otherwise.
One implementation insight every other Tuesday.
What actually works when you ship AI in real organizations. Nothing you could get from a press release.