Grid Energy HarvestBekenstein-gated budget (F19/TH6) · SAMPLE energy · heuristic window
A live, honest grid/harvest view of the Proven Energy Engine. The cumulative series is the Bekenstein-gated energy+information budget — every task receipt's joules_est over time, each passing the gate shannon_bits ≤ output_bytes·8 (F19/TH6, a proven inequality, locked-8 — not an assertion). Energy figures are SAMPLE/ESTIMATE: there is no on-box power meter behind the budget layer, so joules are labeled, never presented as measured. The off-peak / normal / peak window is a client-side heuristic from the server clock — NOT a live tariff feed (no real-time grid price is wired). No free-energy / perpetual-motion claims: the engine harvests WASTED energy and proves bounded information work; any harvest framing is MODELED/SAMPLE.
Why it matters: harvest yield and energy efficiency are measured-or-labeled, never assumed — the only unacceptable outcome is over-claiming (fabricated joules, a fake tariff feed, or a perpetual-motion story). This surface refuses to over-claim: when the budget is empty it shows the structure with an honest "no tasks yet", never fake bars.
Bound endpoints (poll ~5s):
/v1/energy/harvest
/v1/energy/budget
/v1/energy/ledger
Tasks gated
—
Bekenstein receipts
All within bound
—
F19/TH6 gate
Total joules-est sample
—
estimate · not metered
Power window
—
client-side heuristic
(1) Cumulative joules-est over time
sample/v1/energy/harvest · budget.series
No tasks yet. The in-memory Bekenstein budget ledger is empty (it resets on process restart). Track a task via /api/a11oy/v1/energy/budget?bytes=… to populate the series — until then this is the honest empty-state, not a fabricated curve.
joules_est is SAMPLE/ESTIMATE — no on-box meter is wired behind the budget layer (doctrine v11). The curve is the running sum of per-task joules_est; the chip reflects the payload's joules_label. The plot tint mirrors the current heuristic window (green=off-peak, amber=normal, red=peak) — a clock heuristic, NOT a tariff feed.
(2) Off-peak windowserver-clock heuristic
Client-side heuristic, NOT a live tariff feed. The window is derived from the server clock (hour-of-day bands) — it mirrors the engine's off-peak provider but no real-time grid price is wired. No price or carbon figure is claimed as metered.
(3) Bekenstein budget gauge
provenΣ shannon_bits ÷ Σ n·8
No tasks yet. Nothing to gate — the F19/TH6 inequality holds vacuously over an empty ledger.
Proven inequality (F19/TH6, locked-8): empirical byte-entropy is always ≤ n·8, so the needle stays at or below 100% of the ceiling unless an upstream over-claim occurs (then it reads OVER, in red). Kernel-checked in EnergyBudgetWitness.lean (0-sorry). all_within_bound is the real gate result.
(4) Per-task receipt ledger/v1/energy/harvest · per receipt
No per-task receipts yet. The in-memory ledger is empty and resets on process restart. The table structure stays visible so the shape is honest — no placeholder rows.
Harvest composition + signed ledger
modeled/v1/energy/ledger
Signed jobs—
Joules MEASURED-billable— measured
kWh total—
Receipt chain—
Signed ledger unavailable. /v1/energy/ledger did not return a chain — showing no measured total (honest empty-state, never a fabricated number).
Two honesty tiers, never mixed: the SAMPLE/ESTIMATE joules-est above come from the budget layer (no meter); the MEASURED-billable joules here come only from real NVML deltas in the signed receipt chain. The donut splits MEASURED-billable vs SAMPLE-est headroom — a MODELED composition, not a grid source mix.