[kalwei]

What

• Rate order. A plant can hold one standing rate order: a name, and for each item a target rate in units a minute. It has no quantity and no due date.
  • New tables rate_orders and rate_order_lines, with an alembic migration chained to a9c4e1f7b2d3.
  • GET/PUT/DELETE /rate-order: any role reads it; writing needs masterdata.write.
  • A rate_orders kind in the master-data pack. The Satisfactory lab's starter, labs/satisfactory/masterdata/rate_orders.json, is seeded once and then left alone.
  • No order means no target, never a target of zero.
• Parts mix panel under Energy on /dashboard/analysis.
  • Columns: item, target /min, actual /min, target share %, actual share %, gap /min, and the game's own figure (getProdStats) /min as the check.
  • Rows with a target come first, shortest first. Items with no target come after and say "no target".
  • The panel is absent when the plant has no order and no PARTS.* readings.
  • The machine OEE tab gets one line: what the building makes, its rate, the plant's rate and the target.
  • oee_breakdown carries parts_mix, so the assistant sees it.
  • The rate order never enters an OEE figure. OEE and Energy are unchanged.
• "Last played" window. The reader writes FEED.Stretch on each watched poll. On a plant that has those readings, the analysis page now opens on the latest unbroken stretch, shown as "last played 04:41–05:23 today". The hours choices are still there, and ?hours= and ?shift= still win. The 80 % rule is unchanged.
• Reader fixes.
  • A building missing from 5 consecutive unpaused polls is retired as torn down. The count is configurable as satisfactory_torn_down_polls, and the building is un-retired when it is listed again.
  • A line that arrives within 5 s after its poll is folded into that poll instead of being skipped. A genuinely empty poll is still skipped.

Why

The handoff satisfactory-parts-mix: Scott wants to see whether the factory made the mix he asked for the last time he played, beside the OEE figure, without the OEE figure changing.

How verified

• Tests.
  • New: tests/test_the_parts_mix_stands_beside_oee_against_a_standing_rate_order.py (18 prose-named tests, one or more per DoD case) and tests/test_the_parts_mix_panel_and_the_last_play_window_in_a_browser.py.
  • The full fast tier and the whole -m browser tier pass locally, and ruff is clean.
  • The browser test holds /analysis/oee back by 1.5 s with a Playwright route and waits on the panel's rows with wait_for_function.
• Replay on a scratch plant (loopback, port 8197) of the 2026-10-30 fixture: 43 polls, 0 skipped (the late getProdStats lines are now absorbed), 11,580 units booked, 5 buildings retired as torn down. The page opened on "last played 04:41–05:23 today · 41 min", and 41.1 min were watched.

What the screenshots showed

• Parts mix panel, with the starter order "60.0 units/min over 3 items":

  Item	target /min	actual /min	target share	actual share	gap /min	game /min
  Screws	24	22.8	40 %	36 %	−1.2	26.9
  Iron Rod	12	13.1	20 %	21 %	+1.1	13.7
  Iron Plate	24	27.6	40 %	44 %	+3.6	29.5

  Below those, the items with no target: Wire, Iron Ingot, Concrete, Copper Ingot, Biomass, Cable. The note line lists the ores the game made that the MES books none of.

• Window picker: "last played" is selected by default, and the hours choices are below it.

• Machine OEE tab for a constructor: "Parts: makes Screws · 22.3 units/min here over 42.13 min watched (938 units) · the plant 22.8 units/min · rate order 24.0 units/min".

Arithmetic hand-checked against the raw feed and getProdStats:

• Units booked match a recount from the raw file: Screws 938, Plate 1,137, Rod 538, Concrete 1,232, Wire 3,544. The MES's figures differ by under one unit, because it books whole units.
• Concrete runs steady on 2 × 15/min: the MES says 29.96/min and the game 29.99/min.
• The game figure for Screws is 1,101.3 summed over 41 polls with statistics, which is 26.86/min, matching the panel.
• The MES is lower than the game on Screws because the game's CurrentProd is a rolling figure. It decays for about 8 minutes after a building stops (IsProducing 0, e.g. 04:54–04:58 and 05:12–05:22). The MES books only producing intervals (Satisfactory: book only while running; one save on the floor #212's rule), so the gap is the game's smoothing, not lost units.

Notes

• The design's "60 reinforced plates/min, mix 40/40/20" is read as 24 Iron Plate, 24 Screws and 12 Iron Rod a minute.
• Line coverage in the last-play window is about 61 % on the constructors, because several buildings were placed at 05:19–05:22. That is the unchanged coverage rule, not the feed: the feed was watched for all 41 min.
• Out of scope, as the handoff says: anything predictive, writing to the game, charts, the 80 % rule, and the energy figures.

🤖 Generated with Claude Code

[github-actions]

Thanks for the first pull request. Checks: every commit signed off (git commit -s), ruff and pytest green, and the house-rules checklist in the template. The maintainer reviews within a few days.