[FBumann]

Prompt: Dont include nested windows yet. But do it in a stacked draft PR

Note

The following content was generated by AI.

EachWindow may stand anywhere in a tuple of axes. One stitch spans every window axis, so the answer is over every windowed dimension and keeps only what each window owns at every level. The API is unchanged from #1888.

sweep = sps.solve_over('pathway.yaml', sources, (
    sps.EachWindow('period', steps=1, lookahead=1, into='y', carry={'existing': 'capacity'}),
    sps.EachWindow('snapshot', steps=24, lookahead=24, into='t', carry={'soc_initial': 'soc'}),
))
sweep.primal('soc')  # (period, snapshot, value)

What changed

• The windows-last refusal is gone. Two windows over one local index are refused instead, since one local index cannot stand for two dimensions.
• Stitch spans several window axes: locals, dims, the window keys it drops and the outer keys it keeps. owned holds one row per coordinate every window owns: an inner window's owned rows are joined to the outer window's owned rows, so an hour the decade window solved only as lookahead is not in it. restore joins on every owned column but the dimensions.
• Slice.owns holds one count per axis, and a window's carry reads its own axis's seam: the decade window hands on the capacity at the last decade it owns.
• The spill manifest names the stitch's locals, dims, keys and outer keys (inside ANSWER_LAYOUT 5 from feat(api): a sweep can be cut by several axes at once, each carrying its own state, such as a rolling horizon per scenario or a pathway whose decades roll through their hours #1888, which is unreleased).
• The archive's uncut sources drop every window's local index, and a number over a window's local index is spread over that window's dimension for each window axis.
• Docs: the "Several axes" rules say any axis may be windows, the stitch rule covers every level, and the windows-last message gives way to the pathway example.

Tests

• Windows inside windows: decades with a one-decade lookahead, hours rolling inside each. The answer is over (period, snapshot) with each pair once (the lookahead decade dropped), and per window it keeps both keys and both locals.
• A decade window hands its fleet on: the oracle is the loop by hand. Each decade window is solved as its own hourly horizon over its decade and the next, with existing set to the capacity the window before ended its own decade with; every window's objectives match.
• While building it, each decade's owned cost (the spend expression summed over its hours) matched the decade solved alone, when no carry couples them: 2804.0, 6947.5, 11380.0. Primals did not match exactly, because the window also solves the lookahead decade and this LP has several optimal vertices for the store.
• The refusal case window-outside became one-local-index-for-two-windows; Slice.owns assertions read one count per axis.

Mutation table

python -m tools.mutate for the deletion, scoped to tests/test_nested_sweep.py and tests/test_strategy.py; the hand mutations by sed, restored with git checkout --, __pycache__ dropped on both sides. Taken on 507123e, before main was merged in, so the line numbers are that commit's; tree clean after every run.

guard	result
two windows over one local index (axes.py:354-358)	caught
by hand: an inner coordinate is owned only under the outer window that owns its decade (join dropped)	caught (2 failed)
by hand: a window carry reads its own axis's seam (owns[-1])	caught (1 failed)

Verification

• Full suite (pytest -n 4, HiGHS, Gurobi and Xpress) on 06d149d, with main through the 0.8.1 release merged in through feat(api): a sweep can be cut by several axes at once, each carrying its own state, such as a rolling horizon per scenario or a pathway whose decades roll through their hours #1888, under polars 2.0.0: 5632 passed, 566 skipped, 4 xfailed; feat(api): a sweep can be cut by several axes at once, each carrying its own state, such as a rolling horizon per scenario or a pathway whose decades roll through their hours #1888 at 7edf615 gives 5628 passed. python -m tools.changelog check: releases nothing on merge.
• This PR's own diff against feat(api): a sweep can be cut by several axes at once, each carrying its own state, such as a rolling horizon per scenario or a pathway whose decades roll through their hours #1888 is unchanged by that merge but for the changelog's context lines.
• ruff check ., ruff format --check ., pyrefly check: 0 errors. zensical build --strict: pass.
• Not run: test-bench (nothing under bench/ changed in this PR).

Deliberately not done

• Reading one outer window as a whole. per_window=True returns the innermost windows, keyed by every window key; there is no read per decade window.
• A number over two local indexes at once in an archive's uncut sources is spread over each window's dimension in turn, not over both; no test covers it.

🤖 Generated with Claude Code

https://claude.ai/code/session_014wLrXpeSGDqoBFxyrs8NEz

[read-the-docs-community]

Documentation build overview

📚 specsolve | 🛠️ Build #35079040 | 📁 Comparing 06d149d against latest (1487031)

  🔍 Preview build  

56 files changed · + 1 added · ± 55 modified

+ Added

• howto/small-models/index.html

± Modified

• index.html
• examples/index.html
• sweep/index.html
• tables/index.html
• about/architecture/index.html
• about/benchmarks/index.html
• about/changelog/index.html
• about/linopy/index.html
• about/roadmap/index.html
• examples/dispatch/index.html
• and 45 more...

[codspeed]

Merging this PR will not alter performance

✅ 93 untouched benchmarks
⏩ 307 skipped benchmarks¹

---

_{Comparing claude/wizardly-lamport-e3g6om-windows (06d149d) with claude/wizardly-lamport-e3g6om-chains (7edf615)}

Footnotes

1. 307 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[FBumann]

Note

The following content was generated by AI.

Folded into #1888: its branch now holds this PR's commits at e1a9ebb, with main through 9c13352 merged in. The description, tests and mutation rows moved there.

---

Generated by Claude Code