feat/atlas-scan

seg_count never accumulated on upsert — always stays at 1

scripts/atlas/distribucion/test_falsable.py:59

The ON CONFLICT clause updates `division_code` but not `seg_count`; the INSERT hardcodes `1`. dist_enumerate.py correctly accumulates with `seg_count = circuit.seg_count + excluded.seg_count`. Circuits spanning multiple tiles (common at 0.1° step over a 0.5°×0.35° bbox like Monterrey) will always show seg_count=1, making any downstream query that uses seg_count as a circuit-size proxy unreliable.

libre() probes a different WFS layer and bbox — can give false 'disponible'

scripts/atlas/distribucion/test_falsable.py:26

libre() queries `sigdis:subestacion` in the Los Mochis bbox ([-109.0, 25.7, -108.9, 25.8]) to check reachability, but the main loop fetches `sigdis:linea_mta_hc` in the Monterrey/Guadalajara bbox. CFE may allow the subestacion layer while blocking linea_mta_hc, so the liveness check returns True and the script proceeds, only to break on the first tile with a 403 having done zero useful work. The probe should query the same layer (or at least the same bbox) the main loop uses.

None division codes silently excluded from analysis after being inserted

scripts/atlas/distribucion/test_falsable.py:55

`p.get('division')` returns None when the property is absent; None is added to `divisiones` and inserted into the DB as `division_code=NULL`. The downstream loop guards `if not dv: continue`, silently excluding those circuits from transformer and capacity analysis. Filter at collection time (`if cc and p.get('division')`) to make the exclusion explicit and avoid polluting the DB with null rows.

Float tile accumulation drift — loop counter not rounded

scripts/atlas/distribucion/test_falsable.py:31

`x += step` and `y += step` accumulate floating-point error; the round() on yielded values doesn't affect the loop counter. For Monterrey's ~25 tiles this is negligible, but using `x = bbox[0] + n * step` with integer `n` is the conventional fix. Consider for larger bboxes.

SSL verification disabled (inherited from dist_common) — data integrity not guaranteed

scripts/atlas/distribucion/test_falsable.py:1

All WFS calls go through dist_common's CERT_NONE context. Data from CFE GeoServer cannot be integrity-verified. Accepted and documented in dist_common.py as a workaround for CFE's cert chain; no action needed unless auth is ever added to the endpoint.

CLI arg parsing is fragile — IndexError and KeyError on invalid input

scripts/atlas/distribucion/test_falsable.py:91

`sys.argv[sys.argv.index('--zona') + 1]` raises IndexError if --zona is the last token; `main(zk)` raises KeyError if the zone key is unknown. One guard: `if zk not in ZONAS: sys.exit(f'zona desconocida: {zk!r} — válidas: {list(ZONAS)}')` prevents both cryptic failures without adding any dependency.

DB connection never explicitly closed on early exit

scripts/atlas/distribucion/test_falsable.py:47

`con = C.db()` is never wrapped in `with` or closed in a finally block. On a 403 break the process exits and SQLite's WAL recovers, but an explicit `con.close()` in a finally block is cleaner and avoids file-lock issues if the caller ever reuses the process.

No pre-committed numeric threshold for 'apretado' — outcome is ambiguous

scripts/atlas/distribucion/test_falsable.py:1

The docstring anchors the green baseline at 4.1% (Los Mochis) but never states what utilisation level counts as 'notoriamente más apretado'. Without a pre-committed threshold, any observed value above 4.1% can be argued either way post-hoc. For a true falsifiable test, define the decision boundary before running (e.g. '>15% median utilisation = corroborated').

tiles() is duplicated from dist_enumerate.py

scripts/atlas/distribucion/test_falsable.py:27

An identical tiles() generator exists in dist_enumerate.py. The only difference is the default step (0.1 here vs 0.12 elsewhere — 44% more WFS calls). Consider importing from dist_enumerate or moving to dist_common.py. If 0.1 is intentional for urban density, add a constant (STEP = 0.1) with a comment.

libre() is duplicated from division_run.py

scripts/atlas/distribucion/test_falsable.py:19

The same probe pattern appears in division_run.py (libre()) and pilot_run.py (disponible()). A shared C.probe_libre() in dist_common.py would eliminate three copies. For a research spike, even a direct `from division_run import libre` would reduce drift.

step=0.1 vs 0.12 in sibling scripts — silent WFS load increase

scripts/atlas/distribucion/test_falsable.py:27

Other scripts use step=0.12; this uses 0.1, producing ~44% more tiles. For urban bboxes (Monterrey: ~25 vs 18 tiles) the difference is small but a comment or named constant (STEP = 0.1) would signal this is deliberate.