Coalesce concurrent polls to prevent burst responses by tomquist · Pull Request #542 · tomquist/AstraMeter

tomquist · 2026-07-02T20:05:59Z

Summary

Fix a critical issue where batteries and the CT002 meter emulator would receive multiple response deltas in quick succession when a meter read was delayed, causing the battery's control loop to overshoot its target. The fix coalesces concurrent polls from the same battery/consumer into a single response per meter reading.

Problem

When a meter read is delayed — either via WAIT_FOR_NEXT_MESSAGE, a throttled meter, or a slow HTTP source — the battery keeps polling (~1/s) and each datagram spawns its own handler task. Without coalescing, all parked handlers wake on the same fresh reading and each sends a response, creating a burst of deltas that the battery's firmware adds to its output. For Shelly batteries with a zero-export controller, this feeds the same stale error signal multiple times before the plant can respond, causing overshoot. For CT002, it steps the stateful balancer multiple times per real sample, winding up the battery.

Solution

Track in-flight handlers per battery/consumer using a set (_inflight_consumers for CT002, _inflight_batteries for Shelly). When a poll arrives:

If a handler is already parked for that consumer/battery, drop the duplicate poll (the state was already refreshed by the report update)
Otherwise, add it to the in-flight set, process the request, and remove it from the set in a finally block

This ensures exactly one response per meter reading, regardless of how many polls arrive while the read is in flight.

Changes

CT002 (src/astrameter/ct002/ct002.py)

Added _inflight_consumers: set[str] to track consumers with parked handlers
Wrapped the meter-read and response logic in a try/finally block
Early-return duplicate polls with a debug log
Clear the in-flight flag in the finally block to unblock the next reading

Shelly (src/astrameter/shelly/shelly.py)

Added _inflight_batteries: set[str] to track batteries with parked handlers
Wrapped the meter-read and response logic in a try/finally block
Early-return duplicate polls with a debug log
Clear the in-flight flag in the finally block

Tests

Added GatedPowermeter test fixture that parks reads on an asyncio.Event
Added comprehensive test suite for both CT002 and Shelly covering:
- Concurrent polls from one consumer/battery coalesce to a single response
- Coalescing is per-consumer/battery (two batteries each get their own response)
- Polls are answered normally after a burst is coalesced (no permanent lock-out)
- In-flight flag is cleared even when meter reads fail

Notes

The fix is transparent to the battery/meter — it sees the same response it would have gotten without the burst, just once instead of N times
The state update (report parsing) happens before the coalesce check, so dropped polls still refresh per-consumer metadata
Both implementations use identical logic and naming for consistency

https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

Summary by CodeRabbit

Bug Fixes
- Fixed duplicate or repeated responses when meter readings are delayed or slow.
- Requests from the same battery/consumer are now combined so only one reply is sent per reading.
- Improved behavior under bursts of concurrent polls, including after temporary errors, so service resumes normally afterward.

When a meter read is delayed — WAIT_FOR_NEXT_MESSAGE awaiting the next push, THROTTLE_INTERVAL sleeping out its window, or any slow source — the battery keeps polling (~1/s) and each UDP datagram spawns its own handler task. Every handler parked in before_send woke on the single fresh reading and sent its own response, so the battery received a burst of instructions milliseconds apart. Because each CT002 instruction is a delta the firmware adds to its current output, the burst wound the battery up by N times the intended correction and stepped the stateful balancer N times per real sample. Track an in-flight consumer set: the first handler owns the read-and- respond section for its consumer; concurrent polls from the same battery are dropped after refreshing per-consumer state, so exactly one response goes out per reading. Keyed per consumer, so distinct batteries stay independent. This sits at the request-handler level, so it fixes both the WAIT_FOR_NEXT_MESSAGE and THROTTLE_INTERVAL causes at once. Python-only: the ESPHome mirror drains UDP synchronously (pump_udp_ -> handle_request_) with no async before_send, so it has no pile-up to coalesce and needs no counterpart change. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

The Shelly Pro 3EM emulator has the same concurrent-handler pile-up as CT002: each UDP datagram spawns its own task, and when the meter read is delayed (WAIT_FOR_NEXT_MESSAGE awaiting a push, THROTTLE_INTERVAL sleeping, or a slow source) the battery keeps polling so handlers pile up parked in the read; one fresh reading wakes them all and each answers. Although a Shelly response carries an absolute grid reading rather than a CT002-style delta, the harm is the same: the battery runs a closed-loop zero-export controller, so the reading is the error signal it integrates against its own output. Feeding it the same pre-adjustment reading several times before the plant can reflect its response makes it over-correct past target — effectively a burst of deltas. Track an in-flight battery set keyed on source IP (matching the existing dedup identity): the first handler owns the read-and-respond section; a concurrent poll from the same battery is dropped after _track_battery_seen has refreshed liveness/poll-interval state, so exactly one response goes out per reading. Keyed per battery, so distinct batteries stay independent; the flag is cleared in finally so a failed/hung read never locks a battery out. Replaces the old test_multiple_requests_with_throttling (which asserted the pre-fix all-answered behavior) with an end-to-end UDP coalescing test plus deterministic in-process tests for single-response coalescing, per-battery independence, no lock-out, and flag-clear on meter failure. Python-only: the Shelly emulator has no ESPHome mirror. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

coderabbitai · 2026-07-02T20:06:07Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: db4c72be-85a6-45d4-9252-cd16891c9cd2

📥 Commits

Reviewing files that changed from the base of the PR and between 58ee49e and 63ae437.

📒 Files selected for processing (5)

CHANGELOG.md
src/astrameter/ct002/ct002.py
src/astrameter/ct002/ct002_test.py
src/astrameter/shelly/shelly.py
src/astrameter/shelly/shelly_udp_test.py

Walkthrough

CT002 and Shelly emulator request handlers now track in-flight requests per consumer/battery using new sets, dropping duplicate concurrent polls while a meter reading is pending. Both handlers use try/finally to clear tracking state. New tests validate coalescing behavior, per-key isolation, and recovery after bursts or failures.

Changes

Poll coalescing fix

Layer / File(s)	Summary
CT002 request coalescing `src/astrameter/ct002/ct002.py`	Adds `_inflight_consumers` set and wraps `_handle_request` to drop duplicate polls per `consumer_id`, clearing state via try/finally.
CT002 coalescing tests `src/astrameter/ct002/ct002_test.py`	Adds `_RecordingTransport`, `_poll()`, `_gated_before_send()` helpers and three async tests verifying coalescing, per-consumer isolation, and post-burst recovery.
Shelly request coalescing `src/astrameter/shelly/shelly.py`	Adds `_inflight_batteries` set and reorganizes `_handle_request` to drop duplicate polls per battery IP with try/finally cleanup.
Shelly UDP coalescing tests `src/astrameter/shelly/shelly_udp_test.py`	Adds `GatedPowermeter` and tests for UDP burst coalescing, per-battery isolation, post-burst recovery, and cleanup after meter-read failure; removes prior throttling test.
Changelog entry `CHANGELOG.md`	Documents the fix ensuring one battery reply per meter reading.

Estimated code review effort: 4 (Complex) | ~55 minutes

Sequence Diagram(s)

sequenceDiagram
  participant ClientA
  participant ClientB
  participant Handler
  participant Meter
  ClientA->>Handler: poll request
  Handler->>Handler: mark key as in-flight
  Handler->>Meter: read meter value (slow)
  ClientB->>Handler: duplicate poll (same key)
  Handler-->>ClientB: dropped (already in-flight)
  Meter-->>Handler: reading returned
  Handler-->>ClientA: single response
  Handler->>Handler: clear in-flight key (finally)

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: coalescing concurrent polls to prevent burst responses.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch claude/wait-for-next-message-deltas-4valul

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands.}

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

github-actions · 2026-07-02T20:09:39Z

Steering evaluation (base vs head)

Overall: 0 improved, 0 regressed, 15 unchanged across 15 metrics — mean 0% (unchanged).

Priority: priority-weighted 0% (unchanged) — ✅ no do-no-harm guardrail regressions.

Lower is better for every metric. See src/astrameter/simulator/evaluation.py for definitions.

Metrics are the per-scenario mean of 5 seeds.

Aggregate — mean across 32 scenarios

Metric	Base	Head	Δ
settle_mean_s	35.2	35.2	=
settle_p95_s	61.6	61.6	=
unsettled_events	0.5	0.5	=
overshoot_mean_w	44.0	44.0	=
overshoot_max_w	94.6	94.6	=
band_crossings_per_h	297.2	297.2	=
grid_p2p_w	257.8	257.8	=
grid_rms_w	239.5	239.5	=
steady_rms_w	111.6	111.6	=
mean_abs_grid_w	95.2	95.2	=
share_imbalance_w	60.8	60.8	=
avoidable_import_wh	35.0	35.0	=
avoidable_export_wh	25.5	25.5	=
cost_regret_ct	0.82	0.82	=
battery_travel_w_per_h	23700.4	23700.4	=

📊 Interactive grid-power charts (zoom / hover / toggle series) are in the self-contained steering-eval-report.html report — see the link below (it opens directly in the browser).

What do these metrics mean?

Metric	Meaning
`settle_mean_s`	Mean seconds after a load/PV step for grid power to return inside the ±25 W settle band and hold for 10 s (reaction speed).
`settle_p95_s`	95th-percentile settle time — the slow tail of reactions.
`unsettled_events`	Number of disturbance events that never settled within the 10-minute measurement window.
`overshoot_mean_w`	Mean overshoot (W): how far grid power swings past zero to the opposite sign after an event.
`overshoot_max_w`	Worst-case overshoot (W) across all events.
`band_crossings_per_h`	Sign flips per hour across the ±20 W hysteresis band — oscillation / hunting frequency.
`grid_p2p_w`	Sustained peak-to-peak grid swing (95th - 5th percentile) over the whole run — oscillation amplitude. Non-zero whenever the loop keeps hunting, including continuous oscillation the step-response metrics (settle/overshoot) miss.
`grid_rms_w`	RMS grid power (W) over the whole run, transients included — the L2 tracking error: how cleanly the loop held zero, penalising big excursions (overshoot, swings) far harder than a small steady offset. Pairs with battery_travel_w_per_h as the control-effort term.
`steady_rms_w`	RMS grid power (W) during steady state (excluding the 120 s after each event) — residual jitter when nothing is changing.
`mean_abs_grid_w`	Mean absolute grid power (W) over the whole run — overall tracking accuracy.
`share_imbalance_w`	Time-weighted watts misallocated between batteries sharing a phase (sum of each battery's deviation from the even fair share) — 0 when the pool splits load evenly, higher when one battery is left lopsided (issue #523). 0 for scenarios with at most one battery per phase.
`avoidable_import_wh`	Energy imported from the grid (Wh) the battery could have supplied (it had charge and discharge headroom) — missed self-consumption.
`avoidable_export_wh`	Energy exported to the grid (Wh) an AC-chargeable battery could have absorbed (it had room and charge headroom) — missed charging.
`cost_regret_ct`	Money north-star: electricity bill (eurocents, import @ 30 ct/kWh, export @ 8 ct/kWh) over what a perfect-foresight optimal battery would have paid on the same load. Ungameable (both grid directions cost); 0 = matched the optimum. The single number that says how much the controller left on the table.
`battery_travel_w_per_h`	Total absolute change in battery setpoints per hour (W/h) — control effort / actuator wear; lower is smoother.

Per-scenario tables (32 scenarios)

full_battery_low_pace — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 29.1→29.1W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	0.0	0.0	=
grid_p2p_w	70.0	70.0	=
grid_rms_w	29.1	29.1	=
steady_rms_w	29.1	29.1	=
mean_abs_grid_w	14.5	14.5	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	0.0	0.0	=
avoidable_export_wh	3.6	3.6	=
cost_regret_ct	0.0	0.0	=
battery_travel_w_per_h	31900.0	31900.0	=

mixed_cadence/eff — settle 50.8→50.8s, overshoot 164.5→164.5W, RMS 21.9→21.9W

Metric	Base	Head	Δ
settle_mean_s	50.8	50.8	=
settle_p95_s	72.4	72.4	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	64.0	64.0	=
overshoot_max_w	164.5	164.5	=
band_crossings_per_h	24.4	24.4	=
grid_p2p_w	160.6	160.6	=
grid_rms_w	255.3	255.3	=
steady_rms_w	21.9	21.9	=
mean_abs_grid_w	64.1	64.1	=
share_imbalance_w	223.0	223.0	=
avoidable_import_wh	38.5	38.5	=
avoidable_export_wh	25.6	25.6	=
cost_regret_ct	0.95	0.95	=
battery_travel_w_per_h	31306.0	31306.0	=

mixed_cadence/fair — settle 47.7→47.7s, overshoot 70.4→70.4W, RMS 12.9→12.9W

Metric	Base	Head	Δ
settle_mean_s	47.7	47.7	=
settle_p95_s	67.4	67.4	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	30.7	30.7	=
overshoot_max_w	70.4	70.4	=
band_crossings_per_h	21.6	21.6	=
grid_p2p_w	131.9	131.9	=
grid_rms_w	246.0	246.0	=
steady_rms_w	12.9	12.9	=
mean_abs_grid_w	58.6	58.6	=
share_imbalance_w	101.3	101.3	=
avoidable_import_wh	33.7	33.7	=
avoidable_export_wh	24.9	24.9	=
cost_regret_ct	0.81	0.81	=
battery_travel_w_per_h	28457.4	28457.4	=

mixed_cadence_solar/eff — settle 52.5→52.5s, overshoot 384.7→384.7W, RMS 28.6→28.6W

Metric	Base	Head	Δ
settle_mean_s	52.5	52.5	=
settle_p95_s	95.6	95.6	=
unsettled_events	2.0	2.0	=
overshoot_mean_w	62.4	62.4	=
overshoot_max_w	384.7	384.7	=
band_crossings_per_h	37.3	37.3	=
grid_p2p_w	192.3	192.3	=
grid_rms_w	268.4	268.4	=
steady_rms_w	28.6	28.6	=
mean_abs_grid_w	76.4	76.4	=
share_imbalance_w	148.1	148.1	=
avoidable_import_wh	60.5	60.5	=
avoidable_export_wh	54.1	54.1	=
cost_regret_ct	1.38	1.38	=
battery_travel_w_per_h	37319.8	37319.8	=

mixed_cadence_solar/fair — settle 56.0→56.0s, overshoot 75.4→75.4W, RMS 23.7→23.7W

Metric	Base	Head	Δ
settle_mean_s	56.0	56.0	=
settle_p95_s	119.5	119.5	=
unsettled_events	1.8	1.8	=
overshoot_mean_w	28.7	28.7	=
overshoot_max_w	75.4	75.4	=
band_crossings_per_h	28.3	28.3	=
grid_p2p_w	147.8	147.8	=
grid_rms_w	261.6	261.6	=
steady_rms_w	23.7	23.7	=
mean_abs_grid_w	72.2	72.2	=
share_imbalance_w	119.0	119.0	=
avoidable_import_wh	56.0	56.0	=
avoidable_export_wh	52.4	52.4	=
cost_regret_ct	1.26	1.26	=
battery_travel_w_per_h	32435.8	32435.8	=

mixed_venus_b2500/eff — settle 81.2→81.2s, overshoot 221.9→221.9W, RMS 18.6→18.6W

Metric	Base	Head	Δ
settle_mean_s	81.2	81.2	=
settle_p95_s	217.1	217.1	=
unsettled_events	2.4	2.4	=
overshoot_mean_w	100.2	100.2	=
overshoot_max_w	221.9	221.9	=
band_crossings_per_h	49.2	49.2	=
grid_p2p_w	67.3	67.3	=
grid_rms_w	197.5	197.5	=
steady_rms_w	18.6	18.6	=
mean_abs_grid_w	43.2	43.2	=
share_imbalance_w	263.0	263.0	=
avoidable_import_wh	44.2	44.2	=
avoidable_export_wh	20.5	20.5	=
cost_regret_ct	1.16	1.16	=
battery_travel_w_per_h	39704.6	39704.6	=

mixed_venus_b2500/fair — settle 75.4→75.4s, overshoot 231.1→231.1W, RMS 22.3→22.3W

Metric	Base	Head	Δ
settle_mean_s	75.4	75.4	=
settle_p95_s	191.0	191.0	=
unsettled_events	1.4	1.4	=
overshoot_mean_w	110.5	110.5	=
overshoot_max_w	231.1	231.1	=
band_crossings_per_h	53.3	53.3	=
grid_p2p_w	67.1	67.1	=
grid_rms_w	189.9	189.9	=
steady_rms_w	22.3	22.3	=
mean_abs_grid_w	44.5	44.5	=
share_imbalance_w	100.7	100.7	=
avoidable_import_wh	47.4	47.4	=
avoidable_export_wh	19.4	19.4	=
cost_regret_ct	1.27	1.27	=
battery_travel_w_per_h	43475.4	43475.4	=

phase_imbalance — settle 53.4→53.4s, overshoot 145.2→145.2W, RMS 30.3→30.3W

Metric	Base	Head	Δ
settle_mean_s	53.4	53.4	=
settle_p95_s	123.6	123.6	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	64.9	64.9	=
overshoot_max_w	145.2	145.2	=
band_crossings_per_h	87.6	87.6	=
grid_p2p_w	42.4	42.4	=
grid_rms_w	199.6	199.6	=
steady_rms_w	30.3	30.3	=
mean_abs_grid_w	40.4	40.4	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	25.6	25.6	=
avoidable_export_wh	14.8	14.8	=
cost_regret_ct	0.65	0.65	=
battery_travel_w_per_h	18274.2	18274.2	=

single_venus_d_solar — settle 24.2→24.2s, overshoot 94.4→94.4W, RMS 15.9→15.9W

Metric	Base	Head	Δ
settle_mean_s	24.2	24.2	=
settle_p95_s	27.1	27.1	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	76.8	76.8	=
overshoot_max_w	94.4	94.4	=
band_crossings_per_h	10.7	10.7	=
grid_p2p_w	29.9	29.9	=
grid_rms_w	104.3	104.3	=
steady_rms_w	15.9	15.9	=
mean_abs_grid_w	18.5	18.5	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	17.9	17.9	=
avoidable_export_wh	9.9	9.9	=
cost_regret_ct	0.46	0.46	=
battery_travel_w_per_h	9425.4	9425.4	=

single_venus_d_steps — settle 26.3→26.3s, overshoot 90.3→90.3W, RMS 15.5→15.5W

Metric	Base	Head	Δ
settle_mean_s	26.3	26.3	=
settle_p95_s	33.8	33.8	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	73.3	73.3	=
overshoot_max_w	90.3	90.3	=
band_crossings_per_h	25.8	25.8	=
grid_p2p_w	30.9	30.9	=
grid_rms_w	256.4	256.4	=
steady_rms_w	15.5	15.5	=
mean_abs_grid_w	58.6	58.6	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	32.3	32.3	=
avoidable_export_wh	26.2	26.2	=
cost_regret_ct	0.76	0.76	=
battery_travel_w_per_h	21425.2	21425.2	=

single_venus_d_washer — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 61.0→61.0W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	322.4	322.4	=
grid_p2p_w	192.7	192.7	=
grid_rms_w	61.0	61.0	=
steady_rms_w	61.0	61.0	=
mean_abs_grid_w	42.8	42.8	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	13.3	13.3	=
avoidable_export_wh	8.1	8.1	=
cost_regret_ct	0.33	0.33	=
battery_travel_w_per_h	24204.4	24204.4	=

single_venus_drain — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 907.3→907.3W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	73.2	73.2	=
grid_p2p_w	1598.9	1598.9	=
grid_rms_w	907.3	907.3	=
steady_rms_w	907.3	907.3	=
mean_abs_grid_w	645.3	645.3	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	14.4	14.4	=
avoidable_export_wh	10.2	10.2	=
cost_regret_ct	0.21	0.21	=
battery_travel_w_per_h	4062.0	4062.0	=

single_venus_fill — settle 360.0→360.0s, overshoot 0.0→0.0W, RMS 953.6→953.6W

Metric	Base	Head	Δ
settle_mean_s	360.0	360.0	=
settle_p95_s	600.0	600.0	=
unsettled_events	4.0	4.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	10.1	10.1	=
grid_p2p_w	1713.1	1713.1	=
grid_rms_w	978.6	978.6	=
steady_rms_w	953.6	953.6	=
mean_abs_grid_w	662.8	662.8	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	4.8	4.8	=
avoidable_export_wh	6.7	6.7	=
cost_regret_ct	0.14	0.14	=
battery_travel_w_per_h	3173.0	3173.0	=

single_venus_noisy — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 94.3→94.3W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	1452.6	1452.6	=
grid_p2p_w	296.7	296.7	=
grid_rms_w	94.3	94.3	=
steady_rms_w	94.3	94.3	=
mean_abs_grid_w	79.1	79.1	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	42.0	42.0	=
avoidable_export_wh	37.0	37.0	=
cost_regret_ct	0.96	0.96	=
battery_travel_w_per_h	22758.8	22758.8	=

single_venus_pv — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 60.8→60.8W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	38.5	38.5	=
grid_p2p_w	42.9	42.9	=
grid_rms_w	60.8	60.8	=
steady_rms_w	60.8	60.8	=
mean_abs_grid_w	17.3	17.3	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	9.9	9.9	=
avoidable_export_wh	16.1	16.1	=
cost_regret_ct	0.17	0.17	=
battery_travel_w_per_h	6782.0	6782.0	=

single_venus_solar — settle 26.8→26.8s, overshoot 80.3→80.3W, RMS 17.8→17.8W

Metric	Base	Head	Δ
settle_mean_s	26.8	26.8	=
settle_p95_s	32.6	32.6	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	67.1	67.1	=
overshoot_max_w	80.3	80.3	=
band_crossings_per_h	26.3	26.3	=
grid_p2p_w	41.9	41.9	=
grid_rms_w	108.7	108.7	=
steady_rms_w	17.8	17.8	=
mean_abs_grid_w	21.2	21.2	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	15.5	15.5	=
avoidable_export_wh	16.3	16.3	=
cost_regret_ct	0.34	0.34	=
battery_travel_w_per_h	7331.6	7331.6	=

single_venus_solar_slow — settle 33.9→33.9s, overshoot 68.3→68.3W, RMS 22.8→22.8W

Metric	Base	Head	Δ
settle_mean_s	33.9	33.9	=
settle_p95_s	39.9	39.9	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	32.1	32.1	=
overshoot_max_w	68.3	68.3	=
band_crossings_per_h	5.6	5.6	=
grid_p2p_w	60.6	60.6	=
grid_rms_w	131.7	131.7	=
steady_rms_w	22.8	22.8	=
mean_abs_grid_w	31.6	31.6	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	21.3	21.3	=
avoidable_export_wh	26.1	26.1	=
cost_regret_ct	0.43	0.43	=
battery_travel_w_per_h	6486.0	6486.0	=

single_venus_steps — settle 26.0→26.0s, overshoot 88.0→88.0W, RMS 14.7→14.7W

Metric	Base	Head	Δ
settle_mean_s	26.0	26.0	=
settle_p95_s	32.7	32.7	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	69.7	69.7	=
overshoot_max_w	88.0	88.0	=
band_crossings_per_h	23.8	23.8	=
grid_p2p_w	27.1	27.1	=
grid_rms_w	266.9	266.9	=
steady_rms_w	14.7	14.7	=
mean_abs_grid_w	61.1	61.1	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	34.1	34.1	=
avoidable_export_wh	27.0	27.0	=
cost_regret_ct	0.8	0.8	=
battery_travel_w_per_h	19543.6	19543.6	=

single_venus_steps_slow — settle 40.5→40.5s, overshoot 98.5→98.5W, RMS 14.8→14.8W

Metric	Base	Head	Δ
settle_mean_s	40.5	40.5	=
settle_p95_s	56.2	56.2	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	37.1	37.1	=
overshoot_max_w	98.5	98.5	=
band_crossings_per_h	9.4	9.4	=
grid_p2p_w	78.2	78.2	=
grid_rms_w	331.7	331.7	=
steady_rms_w	14.8	14.8	=
mean_abs_grid_w	88.2	88.2	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	47.5	47.5	=
avoidable_export_wh	40.7	40.7	=
cost_regret_ct	1.1	1.1	=
battery_travel_w_per_h	17699.8	17699.8	=

single_venus_trace — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 278.9→278.9W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	284.8	284.8	=
grid_p2p_w	689.3	689.3	=
grid_rms_w	278.5	278.5	=
steady_rms_w	278.9	278.9	=
mean_abs_grid_w	117.0	117.0	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	81.4	81.4	=
avoidable_export_wh	35.5	35.5	=
cost_regret_ct	1.36	1.36	=
battery_travel_w_per_h	33179.4	33179.4	=

single_venus_washer — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 61.0→61.0W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	320.4	320.4	=
grid_p2p_w	196.1	196.1	=
grid_rms_w	61.1	61.1	=
steady_rms_w	61.0	61.0	=
mean_abs_grid_w	40.6	40.6	=
share_imbalance_w	0.0	0.0	=
avoidable_import_wh	11.3	11.3	=
avoidable_export_wh	9.0	9.0	=
cost_regret_ct	0.27	0.27	=
battery_travel_w_per_h	24290.4	24290.4	=

two_venus/eff — settle 18.1→18.1s, overshoot 126.1→126.1W, RMS 14.0→14.0W

Metric	Base	Head	Δ
settle_mean_s	18.1	18.1	=
settle_p95_s	23.0	23.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	70.7	70.7	=
overshoot_max_w	126.1	126.1	=
band_crossings_per_h	33.6	33.6	=
grid_p2p_w	26.2	26.2	=
grid_rms_w	221.6	221.6	=
steady_rms_w	14.0	14.0	=
mean_abs_grid_w	43.1	43.1	=
share_imbalance_w	166.4	166.4	=
avoidable_import_wh	25.7	25.7	=
avoidable_export_wh	17.3	17.3	=
cost_regret_ct	0.64	0.64	=
battery_travel_w_per_h	22029.2	22029.2	=

two_venus/fair — settle 18.4→18.4s, overshoot 116.7→116.7W, RMS 13.8→13.8W

Metric	Base	Head	Δ
settle_mean_s	18.4	18.4	=
settle_p95_s	24.4	24.4	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	94.4	94.4	=
overshoot_max_w	116.7	116.7	=
band_crossings_per_h	21.6	21.6	=
grid_p2p_w	23.1	23.1	=
grid_rms_w	217.6	217.6	=
steady_rms_w	13.8	13.8	=
mean_abs_grid_w	41.4	41.4	=
share_imbalance_w	18.8	18.8	=
avoidable_import_wh	24.2	24.2	=
avoidable_export_wh	17.1	17.1	=
cost_regret_ct	0.59	0.59	=
battery_travel_w_per_h	19385.4	19385.4	=

two_venus_noisy/eff — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 94.3→94.3W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	2898.0	2898.0	=
grid_p2p_w	294.5	294.5	=
grid_rms_w	94.3	94.3	=
steady_rms_w	94.3	94.3	=
mean_abs_grid_w	79.4	79.4	=
share_imbalance_w	26.9	26.9	=
avoidable_import_wh	43.6	43.6	=
avoidable_export_wh	35.8	35.8	=
cost_regret_ct	1.02	1.02	=
battery_travel_w_per_h	23402.6	23402.6	=

two_venus_noisy/fair — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 94.2→94.2W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	2893.6	2893.6	=
grid_p2p_w	294.1	294.1	=
grid_rms_w	94.2	94.2	=
steady_rms_w	94.2	94.2	=
mean_abs_grid_w	79.4	79.4	=
share_imbalance_w	26.4	26.4	=
avoidable_import_wh	43.6	43.6	=
avoidable_export_wh	35.8	35.8	=
cost_regret_ct	1.02	1.02	=
battery_travel_w_per_h	22778.8	22778.8	=

two_venus_slow/fair — settle 41.8→41.8s, overshoot 174.5→174.5W, RMS 14.0→14.0W

Metric	Base	Head	Δ
settle_mean_s	41.8	41.8	=
settle_p95_s	52.8	52.8	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	56.6	56.6	=
overshoot_max_w	174.5	174.5	=
band_crossings_per_h	10.4	10.4	=
grid_p2p_w	81.8	81.8	=
grid_rms_w	303.7	303.7	=
steady_rms_w	14.0	14.0	=
mean_abs_grid_w	76.9	76.9	=
share_imbalance_w	24.4	24.4	=
avoidable_import_wh	41.5	41.5	=
avoidable_export_wh	35.3	35.3	=
cost_regret_ct	0.96	0.96	=
battery_travel_w_per_h	18563.6	18563.6	=

two_venus_solar/eff — settle 26.0→26.0s, overshoot 396.6→396.6W, RMS 20.4→20.4W

Metric	Base	Head	Δ
settle_mean_s	26.0	26.0	=
settle_p95_s	44.3	44.3	=
unsettled_events	1.6	1.6	=
overshoot_mean_w	100.5	100.5	=
overshoot_max_w	396.6	396.6	=
band_crossings_per_h	40.4	40.4	=
grid_p2p_w	60.2	60.2	=
grid_rms_w	228.7	228.7	=
steady_rms_w	20.4	20.4	=
mean_abs_grid_w	51.9	51.9	=
share_imbalance_w	69.3	69.3	=
avoidable_import_wh	42.4	42.4	=
avoidable_export_wh	35.4	35.4	=
cost_regret_ct	0.99	0.99	=
battery_travel_w_per_h	26336.0	26336.0	=

two_venus_solar/fair — settle 25.9→25.9s, overshoot 151.4→151.4W, RMS 20.4→20.4W

Metric	Base	Head	Δ
settle_mean_s	25.9	25.9	=
settle_p95_s	52.1	52.1	=
unsettled_events	1.4	1.4	=
overshoot_mean_w	98.1	98.1	=
overshoot_max_w	151.4	151.4	=
band_crossings_per_h	30.7	30.7	=
grid_p2p_w	59.8	59.8	=
grid_rms_w	225.6	225.6	=
steady_rms_w	20.4	20.4	=
mean_abs_grid_w	51.1	51.1	=
share_imbalance_w	31.3	31.3	=
avoidable_import_wh	40.1	40.1	=
avoidable_export_wh	36.5	36.5	=
cost_regret_ct	0.91	0.91	=
battery_travel_w_per_h	24206.8	24206.8	=

two_venus_trace/eff — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 283.1→283.1W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	307.6	307.6	=
grid_p2p_w	744.9	744.9	=
grid_rms_w	282.1	282.1	=
steady_rms_w	283.1	283.1	=
mean_abs_grid_w	122.0	122.0	=
share_imbalance_w	398.3	398.3	=
avoidable_import_wh	80.0	80.0	=
avoidable_export_wh	42.0	42.0	=
cost_regret_ct	2.06	2.06	=
battery_travel_w_per_h	47999.0	47999.0	=

two_venus_trace/fair — settle 0.0→0.0s, overshoot 0.0→0.0W, RMS 282.1→282.1W

Metric	Base	Head	Δ
settle_mean_s	0.0	0.0	=
settle_p95_s	0.0	0.0	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	0.0	0.0	=
overshoot_max_w	0.0	0.0	=
band_crossings_per_h	310.0	310.0	=
grid_p2p_w	732.2	732.2	=
grid_rms_w	281.1	281.1	=
steady_rms_w	282.1	282.1	=
mean_abs_grid_w	121.1	121.1	=
share_imbalance_w	30.3	30.3	=
avoidable_import_wh	79.0	79.0	=
avoidable_export_wh	42.0	42.0	=
cost_regret_ct	2.04	2.04	=
battery_travel_w_per_h	46054.4	46054.4	=

venus_d_plus_c/eff — settle 20.1→20.1s, overshoot 128.9→128.9W, RMS 14.7→14.7W

Metric	Base	Head	Δ
settle_mean_s	20.1	20.1	=
settle_p95_s	31.9	31.9	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	78.9	78.9	=
overshoot_max_w	128.9	128.9	=
band_crossings_per_h	35.2	35.2	=
grid_p2p_w	28.0	28.0	=
grid_rms_w	214.7	214.7	=
steady_rms_w	14.7	14.7	=
mean_abs_grid_w	41.6	41.6	=
share_imbalance_w	167.3	167.3	=
avoidable_import_wh	24.9	24.9	=
avoidable_export_wh	16.7	16.7	=
cost_regret_ct	0.61	0.61	=
battery_travel_w_per_h	23208.0	23208.0	=

venus_d_plus_c/fair — settle 21.6→21.6s, overshoot 121.0→121.0W, RMS 14.6→14.6W

Metric	Base	Head	Δ
settle_mean_s	21.6	21.6	=
settle_p95_s	32.5	32.5	=
unsettled_events	0.0	0.0	=
overshoot_mean_w	90.6	90.6	=
overshoot_max_w	121.0	121.0	=
band_crossings_per_h	24.8	24.8	=
grid_p2p_w	27.4	27.4	=
grid_rms_w	211.5	211.5	=
steady_rms_w	14.6	14.6	=
mean_abs_grid_w	40.8	40.8	=
share_imbalance_w	30.2	30.2	=
avoidable_import_wh	24.1	24.1	=
avoidable_export_wh	16.7	16.7	=
cost_regret_ct	0.59	0.59	=
battery_travel_w_per_h	21212.8	21212.8	=

📊 Open the interactive report — steering-eval-report.html, a single self-contained file (opens in-browser; download it if your browser blocks inline scripts).

The end-to-end test's two coalesced-away polls are expected to time out, and send_req swallowed that with `except TimeoutError`. But asyncio.wait_for raises asyncio.TimeoutError, which on Python 3.10 is a distinct class from the builtin TimeoutError (they were unified in 3.11). So on 3.10 the expected timeout escaped and failed the test, while 3.11-3.13 passed. Catch both classes so the expected coalescing timeouts are swallowed on every supported Python version. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

tomquist and others added 2 commits July 2, 2026 19:42

Link changelog bullet to PR #542

e48c311

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RaCn4moMmua81wuKo8nKEK

tomquist marked this pull request as ready for review July 2, 2026 20:29

tomquist merged commit 66604bd into develop Jul 2, 2026
60 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Coalesce concurrent polls to prevent burst responses#542

Coalesce concurrent polls to prevent burst responses#542
tomquist merged 4 commits into
developfrom
claude/wait-for-next-message-deltas-4valul

tomquist commented Jul 2, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Jul 2, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 2, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

tomquist commented Jul 2, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Problem

Solution

Changes

Notes

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Sequence Diagram(s)

Uh oh!

github-actions Bot commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Steering evaluation (base vs head)

Aggregate — mean across 32 scenarios

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

tomquist commented Jul 2, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Jul 2, 2026 •

edited

Loading

github-actions Bot commented Jul 2, 2026 •

edited

Loading