Fix a crash rebuilding yesterday's plan from a slot-edge-only status

[springfall2008]

The crash

File "/config/output.py", line 3428, in calculate_yesterday
  plan_html_yesterday, plan_json_yesterday = self.publish_html_plan(...)
File "/config/output.py", line 1181, in publish_html_plan
  charge_window_n = self.in_charge_window(self.charge_window_best, try_minute)
File "/config/plan.py", line 1138, in in_charge_window
  if minute_abs >= window["start"] and minute_abs < window["end"]:
TypeError: '>=' not supported between instances of 'int' and 'NoneType'

This propagates out of calculate_plan() and kills the whole update_pred() cycle — a detail of the History view takes the plan down with it. Reported from a user log at 13:30:35, two minutes after #4872 merged.

Cause

calculate_yesterday() rebuilds each History slot with two walks over the same per-minute status list:

• one decides whether the slot produces a charge/export window (the dominant-state tally)
• one decides where that window starts and ends

#4872 widened only the first to trust a slot's first/last 5 minutes when they hold one state throughout. The second stayed on range(edge_minutes, plan_interval_minutes - edge_minutes).

So a state living solely in a trusted edge was counted by the tally but never seen by the boundary scan, and the window was appended as {"start": None, "end": None}.

The two walks now cover exactly the same trusted minutes, so a counted state always has a start minute by construction. This also makes a short plan_interval_minutes safe: at 10 or below the old interior range was empty, so any activity would have hit this.

A second defect in the same loop

Fixing the first surfaced this one. The charge/export handoff — whichever side starts second ends the one that started first — also fired when both sides started at the same minute, which is exactly what cross-charging is. It closed the export window at its own start minute.

Every cross-charging slot therefore rebuilt a zero-width window: present in the list, invisible to in_charge_window() (start <= m < end can never hold), and worth nothing in the simulated cost. Probed on main:

export windows: [{'start': 0, 'end': 0}, {'start': 30, 'end': 30}, ...]
charge windows: [{'start': 0, 'end': 30}, {'start': 30, 'end': 60}, ...]

The export half of cross-charging was thus still missing from History long after #4466 restored it — because _test_cross_charging_reconstructed_as_both_windows only asserted the list was non-empty, and a list of 60 zero-width windows passes that happily.

An equal start is an overlap, not a handoff, so both sides now run the full slot.

Important

This changes reported savings for anyone whose inverters cross-charge — those export limits now actually apply in the yesterday simulation. In the direction of being correct, but the numbers will move.

Tests

• _test_edge_only_state_still_gets_real_window_bounds — both slot edges, and checks the bounds survive the real in_charge_window() lookup, since that is where the crash landed.
• _test_cross_charging_export_window_covers_the_slot — asserts full-slot coverage, that the export window matches the charge window, and that in_charge_window() can see it.
• _test_cross_charging_reconstructed_as_both_windows strengthened to require start < end. This is the assertion whose absence let the zero-width bug hide.

Both new tests were confirmed to fail against the unfixed code before the fix went in.

./run_all --quick passes in full, including the debug_cases golden plans (22.66s, no shift). ./run_pre_commit clean.

🤖 Generated with Claude Code

[Copilot]

🟢 Approval recommended

The production fix addresses the reported crash and cross-charging reconstruction defect with targeted regression tests added to prevent recurrence.

Pull request overview

Fixes a calculate_yesterday() crash that could terminate the whole update_pred() cycle when reconstructing yesterday’s plan from History slot statuses, and corrects cross-charging slot reconstruction so export windows have real duration and are visible to in_charge_window().

Changes:

• Align the slot “boundary scan” with the same trusted-minute range used by the dominant-state tally, preventing {"start": None, "end": None} windows that crash in_charge_window().
• Fix charge/export “handoff” logic so equal start minutes are treated as overlap (cross-charging) rather than ending one side at its own start.
• Add/strengthen regression tests around edge-only states and cross-charging export windows; document the debugging lesson in the triage journal.

File summaries

File	Description
apps/predbat/output.py	Aligns History slot scans with tally trust range and fixes cross-charging handoff to prevent None/zero-width windows.
apps/predbat/tests/test_calculate_yesterday.py	Adds new regression tests and strengthens assertions to catch zero-width export windows.
.claude/skills/issue-triage/references/debug-journal.md	Adds a debugging note capturing the “two scans must agree” lesson and the cross-charging zero-width pitfall.

Review details

Suppressed comments (1)

apps/predbat/tests/test_calculate_yesterday.py:1662

• This test docstring says the rebuilt window bounds “lie inside its own slot”, and the loop also carries slot_start, but the assertions never validate slot alignment/range. Adding explicit bounds checks (and using slot_start) would better lock in the intended behaviour and avoid the slot_start variable being effectively unused.

    for label, windows, slot_start in (
        ("charge", captured.get("charge_window_best") or [], charge_slot_start),
        ("export", captured.get("export_window_best") or [], export_slot_start),
    ):
        if len(windows) != 1:

• Files reviewed: 3/3 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.