
Before you do, check these issues in the inputs, since they will affect the repair:
– **Sources in the content plan that aren’t in the research notes.** The plan’s must-cover claims cite apoteket.se (the 189 kr BioSalma price, the 17-product count, the ”most studied” category text), tillskottstest.se (the 23-study meta-analysis and +4,43 kg result), starktillsammans.com, and biohackbalance.se (the testwinners). None of these appear in `facts`, `verified_facts`, or `must_cite_sources`. Under Gate 3, claims from those sources can’t be validated, so they’ll need to be downgraded or dropped if the article uses them.
– **Conflicting prices for the same product.** Star Nutrition is listed at 179–249 kr in the research, and the 189 kr BioSalma figure is only in the plan. If the article gives one price for a product, the source has to match it.
– **Quality gates that failed in the research output.** `min_unique_sources_18`, `tier_1_2_share_min_60pct`, and `verified_facts_min_8` are all false. The contract’s `must_cite_min_count: 10` and `must_cite_tiers: [1, 2]` can’t be met from the research as given: only about 8 of the `must_cite_sources` are tier 2 or 3, and only 5 verified facts exist.
– **Internal links are weak.** Neither link dataset entry is about kreatin. Gate 6 and the plan’s `relevance_filter` may reject both, which would leave the article with zero internal links.
– **Research confidence is low.** Under Gate 10, the rumor-list should be at least as long as the confirmed-list.
If the P2 article is unavailable, I can draft a new article from the research notes that uses only the facts with tier 2 or 3 sources, leaves out the unsourced plan claims, and hedges the low-confidence items. That would be a new draft, not a repair, so it would need your sign-off.







