Add ARX forecasting module with telemetry staleness and metrics

CMakeLists.txt

@@ -35,6 +35,7 @@ set(THERMAL_SIMD_CORE_SOURCES
     src/thermal_signals.c
     src/policy/policy_config.c
     src/policy/dispatcher_policy.cpp
+    src/policy/arx_model.cpp
     src/policy/mpc_controller.cpp
 )
 
@@ -47,6 +48,10 @@ target_include_directories(thermal_simd_core
     PRIVATE
         ${CMAKE_CURRENT_SOURCE_DIR}/src
 )
+target_compile_definitions(thermal_simd_core
+    PRIVATE
+        TSD_DEFAULT_COEFF_PATH="${CMAKE_CURRENT_SOURCE_DIR}/config/controller_coeffs.json"
+)
 
 add_executable(thermal_simd src/main.cpp src/thermal_simd.c)
 target_compile_options(thermal_simd PRIVATE -O2 -pthread -fPIC -mno-avx ${THERMAL_SIMD_DISPATCHER_CPU_FLAGS})
@@ -76,6 +81,10 @@ if(BUILD_TESTING)
             ${CMAKE_CURRENT_SOURCE_DIR}/src
     )
     target_compile_definitions(thermal_simd_core_tests PRIVATE TSD_ENABLE_TESTS)
+    target_compile_definitions(thermal_simd_core_tests
+        PRIVATE
+            TSD_DEFAULT_COEFF_PATH="${CMAKE_CURRENT_SOURCE_DIR}/config/controller_coeffs.json"
+    )
 
     add_executable(test_config_parser tests/test_config_parser.c)
     target_link_libraries(test_config_parser PRIVATE thermal_simd_core_tests)
@@ -106,6 +115,13 @@ if(BUILD_TESTING)
     target_compile_options(test_policy_controller PRIVATE -Wall -Wextra)
     add_test(NAME policy_controller COMMAND test_policy_controller)
 
+    add_executable(test_arx_model tests/policy/test_arx_model.cpp)
+    target_link_libraries(test_arx_model PRIVATE thermal_simd_core_tests)
+    target_compile_options(test_arx_model PRIVATE -Wall -Wextra)
+    target_include_directories(test_arx_model PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/src)
+    target_compile_definitions(test_arx_model PRIVATE TSD_ENABLE_TESTS)
+    add_test(NAME policy_arx_model COMMAND test_arx_model)
+
     add_executable(test_healthcheck_runtime_flags tests/healthcheck/runtime_flags.cpp)
     target_link_libraries(test_healthcheck_runtime_flags PRIVATE thermal_simd_core_tests)
     target_compile_options(test_healthcheck_runtime_flags PRIVATE -Wall -Wextra)

README.md

@@ -87,6 +87,7 @@ Sensor dropouts automatically trigger exponential back-off retries and emit logs
 
 See dedicated docs for subsystem details:
 - [Predictive Controller](docs/predictive-controller.md)
+- [Controller Coefficient Format](docs/controller_coeffs.md)
 - [Telemetry Fusion](docs/telemetry-fusion.md)
 - [Metrics Endpoints](docs/metrics-endpoints.md)
 - [Sandbox Workflow](docs/sandbox-workflow.md)

config/controller_coeffs.json

@@ -0,0 +1,9 @@
+{
+  "bias": 1200.0,
+  "ar_temperature": [0.85],
+  "ratio": [-0.35],
+  "severity": [0.05],
+  "trimmed_ratio": [0.0],
+  "ma": 0.25,
+  "staleness_window_ms": 750
+}

docs/controller_coeffs.md

@@ -0,0 +1,40 @@
+# Controller Coefficient File
+
+The predictive controller ingests coefficients from `config/controller_coeffs.json` (or the path provided via `--coeff-path`). The file is a JSON object with the following fields:
+
+| Field | Type | Required | Description |
+| --- | --- | --- | --- |
+| `bias` | number | Yes | Constant term applied to the forecast (millicelsius). |
+| `ar_temperature` | array<number> | Yes | Auto-regressive coefficients applied to historical package temperatures. The array length determines the minimum history window. |
+| `ratio` | array<number> | No | Coefficients applied to historical SIMD ratio measurements (milli-units). |
+| `trimmed_ratio` | array<number> | No | Coefficients applied to the trimmed ratio (if available). |
+| `severity` | array<number> | No | Coefficients applied to the severity metric reported in telemetry (milli-units). |
+| `ma` | number | No | Moving-average gain applied to the most recent residual (`actual - forecast`). |
+| `staleness_window_ms` | number | No | Maximum age (in milliseconds) of telemetry used for prediction. Defaults to 500 ms. |
+
+Example:
+
+```json
+{
+  "bias": 1200.0,
+  "ar_temperature": [0.85, 0.05],
+  "ratio": [-0.30],
+  "severity": [0.04],
+  "ma": 0.25,
+  "staleness_window_ms": 750
+}
+```
+
+## Hot Reload Workflow
+
+1. Update the JSON file on disk (e.g., write a new revision into the ConfigMap or local path).
+2. Send `SIGHUP` to the dispatcher process. The controller marks a reload for the next control tick.
+3. On the following recommendation cycle, the controller attempts to parse the file:
+   - Success increments `predictive_coeff_reload_total` and logs an INFO entry with the new history window and staleness guard.
+   - Failure increments `predictive_coeff_reload_errors_total`, logs an ERROR entry, and falls back to the previous coefficients or averaging forecast.
+
+## Validation Tips
+
+- Use `tests/policy/test_arx_model.cpp` as a reference for crafting deterministic coefficients during development.
+- Monitor `predictive_abs_error_millic_total` to evaluate how well the updated coefficients track observed temperatures.
+- Pair coefficient adjustments with updates to [Predictive Controller](predictive-controller.md) documentation to keep operational guidance in sync.

docs/metrics-endpoints.md

@@ -21,7 +21,11 @@ The dispatcher exports metrics and health data via a multi-channel strategy tail
 | Metric | Type | Description |
 | --- | --- | --- |
 | `predictive_forecasts_total` | Counter | Forecast cycles executed by the predictive controller. |
-| `predictive_downgrades_total` | Counter | Controller-driven SIMD downgrades. |
+| `predictive_decisions_total` | Counter | Control loop iterations that issued a predictive decision. |
+| `predictive_abs_error_millic_total` | Counter | Accumulated absolute error between forecast and observed temperature. |
+| `predictive_stale_samples_total` | Counter | Telemetry snapshots rejected because they exceeded the staleness window. |
+| `predictive_coeff_reload_total` | Counter | Successful coefficient reloads (startup and SIGHUP). |
+| `predictive_coeff_reload_errors_total` | Counter | Failed attempts to reload the coefficient file. |
 | `telemetry_snapshots_total` | Counter | Telemetry fusion snapshots published. |
 | `telemetry_degraded_total` | Counter | Snapshots flagged as degraded due to missing signals. |
 | `patch_transitions_total` | Counter | Successful SIMD trampoline swaps. |

docs/predictive-controller.md

@@ -18,17 +18,24 @@ The predictive controller combines reactive thermal throttling with a short-hori
 Each signal is tagged with a monotonic timestamp. Stale signals (>2 intervals) are discarded and treated as unavailable.
 
 ## Forecast Model
-The controller uses a single-step ARX model:
+The controller uses a single-step ARX/ARMAX model implemented in `src/policy/arx_model.cpp` and driven by coefficients stored in `config/controller_coeffs.json` (see [Controller Coefficients](controller_coeffs.md)). The model consumes a sliding window of recent telemetry samples and projects the next package temperature in millicelsius:
 
 ```
-T[t+1] = a0 + a1 * T[t] + a2 * CPI[t] + a3 * Freq[t] + a4 * Power[t]
+T[t+1] = bias
+         + Σ φᵢ · T[t-i]
+         + Σ θᵢ · Ratio[t-i]
+         + Σ γᵢ · Severity[t-i]
+         + ψ · ε[t]
 ```
 
-- Coefficients `a1..a4` are calibrated offline using lab traces and stored in `config/controller_coeffs.json`.
-- The bias `a0` compensates for ambient temperature.
-- Missing inputs zero out their coefficients and raise the `predictive_input_gaps` metric.
+- `φᵢ`, `θᵢ`, and `γᵢ` are configurable auto-regressive and exogenous coefficients.
+- `ψ` is an optional moving-average gain applied to the most recent residual `ε[t] = T[t] - T̂[t]`.
+- Missing temperature samples disable the prediction path and fall back to a simple moving average.
+- Coefficient files support hot-reload: the controller listens for `SIGHUP` and re-reads `config/controller_coeffs.json` on the next control tick. Successful reloads and failures are logged and exported via metrics.
 
-The forecast produces a projected temperature and CPI value under the current SIMD width. The controller evaluates transitions (`SSE4.1`, `AVX2`, `AVX-512`) and selects the highest width whose projected temperature remains below `temp_ceiling_c - safety_margin_c` and whose CPI ratio is under `up_ratio`.
+Telemetry freshness is enforced prior to forecasting. If the latest sample exceeds the configured `staleness_window_ms`, the controller skips predictive evaluation, logs a warning, and records `predictive_stale_samples_total`.
+
+The forecast produces a projected temperature under the current SIMD width. The controller evaluates transitions (`SSE4.1`, `AVX2`, `AVX-512`) and selects the highest width whose projected temperature remains below `temp_ceiling_c - safety_margin_c` and whose CPI ratio is under `up_ratio`.
 
 ## Decision Pipeline
 1. **Acquire Inputs:** Pull the latest telemetry fusion snapshot (all `TelemetrySnapshot` values share a generation number).
@@ -52,11 +59,12 @@ The forecast produces a projected temperature and CPI value under the current SI
 | `--predictive-alpha` | EWMA alpha applied to CPI history. | 0.25 |
 
 ## Telemetry & Metrics
-- `predictive_forecasts_total`: incremented each control tick.
-- `predictive_downgrades_total`: decision to reduce SIMD width due to forecast.
-- `predictive_input_gaps_total`: missing telemetry inputs for a tick.
-- `predictive_emergency_transitions_total`: emergency scalar fallbacks.
-- `predictive_coeff_reload_errors_total`: failure to read coefficients on reload.
+- `predictive_forecasts_total`: ARX/ARMAX forecasts executed with valid telemetry.
+- `predictive_decisions_total`: control decisions driven by the predictive controller.
+- `predictive_abs_error_millic_total`: accumulated absolute prediction error in millicelsius.
+- `predictive_stale_samples_total`: telemetry snapshots rejected due to staleness.
+- `predictive_coeff_reload_total`: successful coefficient reloads (including on startup).
+- `predictive_coeff_reload_errors_total`: failures to read or parse the coefficient file.
 
 Metrics are exposed through the metrics subsystem documented in [Metrics Endpoints](metrics-endpoints.md).
 

docs/sandbox-workflow.md

@@ -52,7 +52,7 @@ This workflow describes how to exercise the dispatcher in a non-production sandb
 
 ## Exit Criteria
 - Dispatcher exits 0.
-- `artifacts/*/metrics.ndjson` contains expected counters (`predictive_downgrades_total > 0` during spike scenario).
+- `artifacts/*/metrics.ndjson` contains expected counters (`predictive_decisions_total > 0` during spike scenario).
 - No `state=emergency` logs during nominal runs.
 
 ## Automation

include/thermal/simd/metrics.h

@@ -22,6 +22,12 @@ typedef enum {
     TSD_METRIC_PATCH_FAILURES,
     TSD_METRIC_HEALTH_CHECK_FAILURES,
     TSD_METRIC_SOFTWARE_TIMEOUT_ESCALATIONS,
+    TSD_METRIC_PREDICTIVE_FORECASTS,
+    TSD_METRIC_PREDICTIVE_STALE_SAMPLES,
+    TSD_METRIC_PREDICTIVE_RELOADS,
+    TSD_METRIC_PREDICTIVE_RELOAD_ERRORS,
+    TSD_METRIC_PREDICTIVE_ABS_ERROR_MILLIC,
+    TSD_METRIC_PREDICTIVE_DECISIONS,
     TSD_METRIC_COUNT
 } tsd_metric_counter_t;