A prediction misses → find its parent claim in the left column.
Open every dependent link; mark which inherit the miss.
Log the framework correction in ledger / RESEARCH per scoring rules - one upstream fix, not twenty silent edits.
Worked example. Suppose the 2029 behind-the-meter prediction in energy misses because permitting reform makes tariff supply competitive again. The parent claim is "energy / permits bind before chips" in the substrate table. Its dependents split on inspection: Uncertainty 2anticipated exactly this skew, so it scores as a called shot, not a miss; the bipolar page loses its "US self-constrains" paragraph but keeps its structure; the 2026–2028 timeline moves left. One miss, three different inheritances. That asymmetry - dependents inherit a miss differently - is why the index maps pages rather than propagating a single truth value, and why step 2 says mark which, not mark all.
What the index deliberately leaves out. Low-centrality claims are excluded even when pages cite them, because an index that lists everything ranks nothing - the value here is the ranking, and a maintenance burden that grows with corpus size would eventually make the index wrong, which is worse than incomplete. The "updated through round 13" stamp above is therefore a feature to check, not an apology: when a new high-centrality claim appears (as complement erosion did in round 21), the Review step owes this page a row before the round closes.
Assets; prices; Game 3 distributional read; U5 as accelerant
Open-weight lag ~3–6 months (measured)
Game 1 prize; Game 2 leakage; Part V row 4; bipolar coupling; global-south lease
Gray-zone Taiwan (soft assumption fail)
Compute risk premium; timelines in quarters; A6 not sufficient alone
These rows do not replace the stamp; they are the debt the stamp records. A future round that rewrites the full table should fold them into the main index and reset the stamp.