GOAL: for each assigned row R, fetch one USGS elevation sweep and save it.

WORKDIR (bash tool = mcp__workspace__bash): /sessions/upbeat-quirky-clarke/mnt/outputs/sfmap/

STEP 1 - get the URL for row R:
  awk -v r=R '$1==r {print $3}' /sessions/upbeat-quirky-clarke/mnt/outputs/sfmap/urls200_final.txt
  (field 2 is k, the expected sample count.)

STEP 2 - fetch it with mcp__workspace__web_fetch ONLY.
  curl/wget/python-requests are STRICTLY PROHIBITED. Bash is for local file I/O only.
  If the response body comes back EMPTY (just the URL echo, no JSON): retry the same URL once.
  If still empty, cache-bust by moving the "&f=json" parameter to a different position in the
  query string (e.g. put it right after "getSamples?") and fetch again. Never change the
  geometry, sampleDistance, or geometryType values.

STEP 3 - read the JSON. It looks like:
  {"samples":[{"location":{...},"locationId":0,"value":"38.582172394",...}, ...]}
  CRITICAL FACTS:
   - The samples come back OUT OF ORDER (grouped by rasterId). Never rely on array position.
   - Some locationIds are sometimes ABSENT (dropped at raster tile seams). That is expected.
   - The ONLY thing that establishes alignment is the "locationId" field. Use it verbatim.
     Do NOT renumber, reindex, sort-and-count, or fill in gaps.

STEP 4 - save. Build a space-separated list of "<locationId>:<value>" tokens, value rounded to
  1 decimal place, keeping the sign. Include EVERY sample present in the response, in any order.
  If a value's magnitude is absurd (abs > 10000, e.g. -3.4028235E38 = NoData), write -9999 for it.
  Then run exactly:
    cd /sessions/upbeat-quirky-clarke/mnt/outputs/sfmap/ && python3 save200.py R "0:38.6 1:44.3 2:49.4 ..."
  The script validates ids against k, writes erow_<RR>.txt, and prints
  "row R: <saved>/<k> saved, missing ids [...]". If it prints an assertion error, fix your token
  list and re-run. Do one row per save call.

STEP 5 - move to the next assigned row. Save each row immediately after fetching it; never batch
  several rows' data before saving.

DO NOT: create any other files, write .md reports, edit save200.py/plan200.json/urls200_final.txt,
or touch rows outside your assignment.

FINAL ANSWER: one line per row -> "row R: saved/k, missing [ids]". Nothing else.