Boston's Permit Feed Has Been Silent for 21 Days. It Holds 658,919 Permits Back to 2006 — the Deepest Dataset in the Corpus. All 87 Cells Are Now Stale.
The Setup
Boston's building_permits feed has not received a new issue_date since 2026-07-31 — 21 days as of this writing. That silence lands on the single deepest feed in the corpus: 658,919 permits reaching back to 2006-09-26, easily the longest and largest municipal permit history tracked anywhere in the platform. Weekly volume ran a steady 589 to 830 permits a week from late April through the week of July 27, with no seasonal taper visible in that run. Then it stopped. The three full weeks since — August 3, August 10, August 17 — show zero permits each.
The Chain
The scorer has already reacted. All 87 of Boston's tracked cells now carry staleness_tier = 'stale' and needs_refresh = true in cell_scores. The development_pipeline value sitting on those cells averages 59.3 — a real, differentiated number, not the 33.33 no-evidence sentinel this corpus has previously documented for cells with an empty permit signal. That distinction matters: Boston's score isn't degenerate, it's frozen. It reflects genuine permit activity as of late July, held in place because the input that would move it has gone quiet.
The Implication
A frozen-but-real score is a harder failure mode to catch than a degenerate one. A cell stuck at 33 announces itself — the number is a known artifact. A cell reporting 59.3 looks exactly like a cell that was scored yesterday, because as far as the scoring logic can tell, it was: the last input it saw was legitimate. Every one of Boston's 87 cells is currently presenting three-week-old ground truth with the same confidence as a cell whose feed updated this morning, and nothing in the score itself signals the difference. staleness_tier and needs_refresh carry that information, but only if something downstream reads them.
What to Watch
Whether this is a source-side stoppage — the City of Boston permit portal itself going quiet — or a collector-side failure that simply stopped picking up rows that exist upstream. Whether Boston's staleness_tier escalates past stale if the gap runs past three or four weeks. Whether any of the other 21 tracked metros show the same shape of gap starting in early August.
Limitations
This is a coverage read, not a root-cause diagnosis — no scout run logs or source_uri checks were inspected here, so source outage versus collector failure is not distinguished. It says nothing about whether 59.3 was an accurate development_pipeline reading before the freeze, only that it has not been able to change since. Boston only; no cross-metro freeze scan beyond the one comparison query in this note.
--- Data as of 2026-08-21. Source: `building_permits` (issue_date), `cell_scores` (staleness_tier, needs_refresh, development_pipeline), Axiom Locus.