Tesla Physics
Sidebar group Tesla Physics has one workbench entry; its grouped navigation opens all 15 investigations (or search /explore).
The Tesla Physics workbench at /tesla-physics groups 15 individually lazy-loaded investigation sections into Time & coverage, Motion & states, Charging & range, and Integrity & firmware. Selecting a section retains the same shell and opens its existing /tesla-physics/{option} deep link. Each group has direct links on desktop; the compact mobile selector groups its options by the same four categories so evidence appears sooner on screen. Each section keeps its measured summaries, interpretation, drilldowns, filters, and missing-slice handling. The shell explains requested versus recorded windows, row counts, availability, and caps. Large timestamp tables are closed by default and paginated on demand. Legacy /tesla-only/* bookmarks still redirect to the matching deep link; Physics Ledger and Science Lab remain independent pages with their own evidence windows.
| Screen | Path | What it does | When empty |
|---|---|---|---|
| Physics workbench | /tesla-physics | VIN-scoped source boundaries, trust warnings, and grouped navigation to all 15 investigations. | Renders an empty state when no report is available. |
| Physics Ledger | /tesla-physics/ledger | VIN-scoped energy/force solver: predicted vs measured, unexplained residual, unknown budget. | Shows unknown honestly when signals are missing; never zero-fills. |
| Science Lab | /science | Seven- or 30-day evidence overview linked to rest-voltage/current-step observations, thermal fit diagnostics, matched weather drives, tire model sensitivity, and generated notebook methods. | Each of the five reports loads independently. Unsupported fits stay unknown; CIs and holdouts are shown only when computed, not implied. |
| Three Clocks | /tesla-physics/clocks | Second-precision latest readings, paired ingest lag and gap distribution, six latest samples, a long-gap filter and paginated raw timestamps. | Missing ingest stays unknown. |
| Life Tape | /tesla-physics/life-tape | State-duration and interval-count breakdown, classified/window share with overlap warning, longest interval, chronology and filterable raw intervals — not GPS. | Missing evidence stays unknown. |
| Contradiction Court | /tesla-physics/contradictions | Backend episodes, per-kind episode/reading counts, observed bounds, kind/uncertainty filters and nearby charge-port context. | An episode is not proof of a distinct fault; individual source readings are not in this report. |
| Trip-Meter Genealogy | /tesla-physics/meters | Current counters, observed drop sequence and firmware counter bounds, mode context and before/after drilldown. Null is not zero. | With no returned drops, still shows current counters and firmware evidence, not an invented continuous timeline. |
| Unknown OS | /tesla-physics/unknown | Sampled share, uncovered share, per-signal budget breakdown and links to clock, broker and signal investigations. | Missing signals stay unknown. |
| Car Kept Living | /tesla-physics/car-kept-living | MQTT, queue, replay, paired ingest lag, bounded-source availability and telemetry-gap context. | Unreported broker state stays unknown. |
| Tesla-Language Logbook | /tesla-physics/logbook | First/latest historical observations, session boundaries, positive-ID links and kind-filtered narrative. | Synthetic live state is only a fallback. |
| Firmware Epochs | /tesla-physics/firmware-epochs | Each version's observed bounds, counter changes, dwell availability and version filter, not fleet proof or FSD engagement. | Unknown bounds stay unknown. |
| Charge-Port Court | /tesla-physics/charge-port | Latest state/current with gear, version, observed state changes, schedule context and filterable per-sample evidence. | Missing readings stay unknown. |
| Black Box 90s | /tesla-physics/black-box | Latest selected trigger, 90-second window, frame changes, observed gears/firmware, source caps and optional frames. | Empty frames do not exclude an event. |
| Owner Dictionary | /tesla-physics/dictionary | This car's Complete-to-unplug, Park dwell, source-dwell distribution and guarded quartile comparison. | Missing samples do not imply zero dwell. |
| Physics Vault | /tesla-physics/vault | Hash/HMAC status, capped session boundaries and optional drive/charge links. | Hashing does not prove complete coverage. |
| Mode Laws | /tesla-physics/modes | Valet, Service, Transport inference boundaries with separately returned counter and firmware context. | Unknown mode stays unknown; current modes do not prove past modes. |
| Nervous System | /tesla-physics/nervous-system | Signal-status breakdown, non-alive details, status filter and exact-name Unknown OS budget cross-reference. | Silence is not zero or proof of a component fault. |
| Range Disagreement | /tesla-physics/range | Rated, typical, ideal and energy remaining with spread and pairwise differences; not a true-range forecast. | Missing estimates remain unknown. |
Following the evidence
The hub prioritizes sampled coverage, unknown hours, backend contradiction episodes and meter drops. Contradiction Court uses the backend's episode window and observation count; an episode is not a confirmed independent physical failure. The individual readings grouped by the backend are not included in the exclusive report. Three Clocks leads with second-precision timestamps, stored-ingest lag, gap distribution and a long-gap filter instead of a giant repeated minute-only table. Life Tape and Contradiction Court keep state/kind filters for their evidence. Meter readings have no continuous counter time series; reset pairs and firmware epoch counter bounds are the only historical meter comparisons, and a counter ratio cannot measure FSD engagement. Logbook and Vault session boundaries link to drive and charging details. Charge-port and Black Box views show current, schedule, and latch alongside state; recorded gear and firmware readings appear in both evidence tables (including their mobile views), and absent readings stay unknown. Firmware epochs expose their observed boundaries, while the Vault distinguishes a bare hash from an HMAC-backed certificate and shows its coverage window. Range Disagreement shows the spread of available estimates, not a prediction of actual range.
The exclusive report reads up to 14 days of bounded telemetry and session history; it is not a lifetime archive. The sampled-window percentage measures time with accepted telemetry, not completeness of every signal. A zero finding count only applies to the returned evidence. Per-signal unknown budgets overlap and must not be added. The Black Box captures the latest selected trigger and may have no frames within its 90-second interval. Logbook narrates session boundaries alongside observed historical gear and charge-state changes. The first recorded value establishes observed state, not an exact transition time. Synthetic live state is excluded when recorded entries exist; it appears only as a fallback when none do. Historical signal entries have no session ID and do not link to a drive or charge. Gaps and unobserved changes remain unknown. The evidence boundaries panel lists requested and recorded timestamps, row counts, availability, and row/session cap warnings; it does not equate an empty row set with complete coverage. The hub calls out partial sources beside its summary; Black Box separately warns when its source or trigger-selection history is unavailable or capped. Vault separately warns when history, drive sessions, or charge sessions are capped: its hash attests to the returned boundaries, not completeness of the underlying history. Follow the separate Physics Ledger and Science Lab for their independent model and measurement evidence.
Analysis boundaries
New HTTP routes use internal/handler/v1/analysis and the physicssvc / sciencesvc application services. The ledger uses recorded intervals, exposes gaps and row caps, and does not reconcile partial telemetry against a complete session. Explicit Disconnected state, not latch release, denotes unplugging. Energy/motion integrations reject seed-only values and observations older than two minutes instead of treating forward-filled power as a fresh measurement. Mass-dependent drive terms require TESLASYNC_VEHICLE_MASS_KG; this is a single installation-wide value and must not be set to one vehicle's mass when vehicles differ. Tesla Fleet Telemetry's HvacPower is an on/off state, not watts; grade has no recorded elevation in this ledger. Accessory energy, grade, friction braking, and the complete unexplained residual stay unknown without their respective measured inputs. The predicted total is the sum of known terms only, never a complete estimate when a term is missing.
Science fits use the latest contiguous firmware epoch. Voltage/current steps must be timestamp-aligned; seed-only freshness is unknown. Rest-end voltage is not proven equilibrium OCV, and pack resistance is not cell resistance. Thermal fits do not bridge separate Park episodes, active heating, long gaps, or drifting ambient conditions, and require observed zero HVAC power. Tire uncertainty is a model sensitivity band, not a calibrated 95% confidence interval. Use the Science Lab overview to inspect available evidence by domain. Rest-voltage tables retain dwell, temperature and SOC context; resistance steps show current and pack IR (not cell IR). Thermal cooldowns include their sample counts, confidence bounds, fitted ambient and residual RMSE. Weather rows link back to their drives and expose both session energy per distance and model residual; Pearson correlations require at least five complete matches and are not causal. The notebook exposes generated-row methods, input and missing signals, residual diagnostics and firmware epoch, not persisted experiments. The default window remains seven days; the optional 30-day window is the science endpoint's maximum.
The broader science loop is not fully implemented: throughput/rest exposure does not identify separate cycle/calendar aging; generated notebook rows are not persisted reproducible experiments; not every model has independent-day holdouts or correlation-robust uncertainty. The UI must not imply otherwise. Grafana exposes observations and model inputs, not persisted solver fits. Charging throughput describes returned, completed sessions contained in the requested window, not lifetime exposure. It remains unknown when session energy is missing, unbounded, invalid, or capped; equivalent cycles use the explicitly labeled assumed reference capacity. Rest exposure is limited to the fitted firmware epoch and must not be treated as the same coverage as session throughput.
Weather joins are disabled by default. Set TESLASYNC_SCIENCE_WEATHER_ENABLED=true (Helm: physics.weatherEnabled: true) only after accepting that drive-start coordinates will be sent to Open-Meteo. Archive density uses surface pressure. Without this opt-in, weather remains explicitly unavailable.