You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Question: intended operator path to curate an overgrown entity page (4.5k fenced facts; repair duplicate-facts reaches only a fraction)
#8313
Moved from #8120, opened by @TBWMT on 2026-10-31. Feature requests now live in Discussions > Ideas.
Question first: for an entity page whose facts fence has grown into the thousands, what is the intended operator path to curate it — is there (or is there planned) a bulk-retire verb (e.g. by notability), a fence-shrink path, or is capture-time capping (facts.page_cap.*, #6877) the intended boundary with existing overgrown pages expected to be curated by hand?
Observed (CLI 0.60.529.0, self-hosted Postgres, single managed git-canonical source):
One person-entity page carries 4,528 live facts, 4,355 distinct (so ~170 same-claim extras). Effectively the whole page is one ## Facts fence; the page is over 1 MiB and is stored as 2 segments. The rows accumulated from months of automatic extraction (email/chat capture lanes) — no single event produced them.
For the overgrown page's own rows, the repair is only a small lever: 13 items are same_entity (retireable), 148 are segmented_page (kept by design), 117 are cross_entity_held (owner-reviewed), 60 are fence_kept_elsewhere — the tool itself says fence rows "cannot name a fact outside their page, so they are kept (forget or edit them in their page)."
So the page's mass is mostly distinct auto-extracted rows, many plausibly low-notability, and no bulk verb reaches them: forget strikes one fact at a time, and forget --purge removes all same-claim rows plus tombstones the claim — neither is a curation tool for thousands of rows. We read memory-boundaries.md and repair.md and found no shrink/trim path for an existing fence.
Is a bulk curation verb (retire/expire rows below a notability threshold on one page, preview-first) on the roadmap, or is per-row forget the intended (if slow) path?
Is there a sanctioned way to shrink an existing facts fence, or is fence-append-only by design (in which case: should very large entity pages be split, and is there a supported split)?
For cross_entity_copies_held rows (1,277 source-wide here): is owner-by-owner review the intended path, or is a preview listing them with a disposition action planned?
Happy to supply more measurements (per-day accrual series for this page, notability distribution) if useful. This is a question, not a defect report — the tooling behavior above all matches the documented repair scope as we read it.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Moved from #8120, opened by @TBWMT on 2026-10-31. Feature requests now live in Discussions > Ideas.
Question first: for an entity page whose facts fence has grown into the thousands, what is the intended operator path to curate it — is there (or is there planned) a bulk-retire verb (e.g. by notability), a fence-shrink path, or is capture-time capping (
facts.page_cap.*, #6877) the intended boundary with existing overgrown pages expected to be curated by hand?Observed (CLI 0.60.529.0, self-hosted Postgres, single managed git-canonical source):
## Factsfence; the page is over 1 MiB and is stored as 2 segments. The rows accumulated from months of automatic extraction (email/chat capture lanes) — no single event produced them.repair duplicate-factsdry-run over the source reports:affected: 1014, residualssame_entity_groups: 1129,cross_entity_groups: 1054,copies_to_retire: 1788,cross_entity_copies_held: 1277,fence_kept_elsewhere: 466,segmented_page: 210.same_entity(retireable), 148 aresegmented_page(kept by design), 117 arecross_entity_held(owner-reviewed), 60 arefence_kept_elsewhere— the tool itself says fence rows "cannot name a fact outside their page, so they are kept (forget or edit them in their page)."forgetstrikes one fact at a time, andforget --purgeremoves all same-claim rows plus tombstones the claim — neither is a curation tool for thousands of rows. We readmemory-boundaries.mdandrepair.mdand found no shrink/trim path for an existing fence.What we'd like to know:
forgetthe intended (if slow) path?facts.page_cap.*(feat(facts): opt-in capture-time fact filter, specific-entity routing, and per-page fact cap #6877, currently an unmerged contributor PR folded into fix waves at maintainer discretion) the intended long-term boundary — i.e., existing over-large entity pages are expected to remain as-is unless hand-curated?cross_entity_copies_heldrows (1,277 source-wide here): is owner-by-owner review the intended path, or is a preview listing them with a disposition action planned?Happy to supply more measurements (per-day accrual series for this page, notability distribution) if useful. This is a question, not a defect report — the tooling behavior above all matches the documented repair scope as we read it.
All reactions