Audit items #7 and #5.
#7 — BrentPricePredictor::generatePrediction previously wrote both an
EWMA row and an LLM row to price_predictions on every run. The
downstream OilSignal already prefers llm_with_context > llm > ewma, so
the EWMA row was dead weight 95% of the time. Now we try LLM first; if
it returns null (no API key, parse failure, etc.) we compute and persist
EWMA as a real fallback. This also avoids redundant work on the success
path.
Updated the "stores both" test to "stores only LLM" — asserts no EWMA
row is written when the provider succeeds.
#5 — BrentPricePredictor and AnthropicPredictionProvider both had
byte-identical computeEwma() methods with identical EWMA_ALPHA = 0.3
constants. Extracted to App\Services\Ewma::compute() and dropped both
private methods + their alpha constants.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Anthropic, Gemini, and OpenAi providers each repeated: API-key gate,
chronological price-list building, response validation
(direction/confidence/reasoning), TrendDirection::tryFrom, confidence
cap at 85, and the top-level try/catch + Log::error.
Now in AbstractLlmPredictionProvider:
- LLM_MAX_CONFIDENCE constant
- buildPriceList(Collection) helper
- buildPrediction(input, ?source) — handles direction validation,
confidence cap, model construction
- defaultPrompt(priceList) — shared by Gemini and OpenAi
- Default predict() flow (apiKey + callProvider + buildPrediction +
try/catch). Gemini and OpenAi only implement apiKey() and
callProvider(). Anthropic overrides predict() because of its
multi-phase web-search + forced-tool flow but reuses the helpers.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
