spec: fix 4 critical issues from review — OFI_DIM cascade, deploy atomicity

1. Enumerate all 6+ files with hardcoded OFI_DIM=8 literal
2. Document atomic deploy requirement (code + fxcache together)
3. Rename Depth Renewal Rate → Order Count Flux (MBP-10 approximation)
4. Fix Intra-Bar Momentum to Welford online covariance (O(1), no Vec)
5. Clarify snapshot() is pure read, non-mutating, idempotent
6. Fix precompute time: 8s CPU + 5min I/O = ~5-10min total

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

docs/superpowers/specs/2026-04-16-tick-microstructure-intelligence-design.md

@@ -44,7 +44,7 @@ All features are **incremental** — O(1) per tick via running statistics (EMAs,
 | 12 | **Spread Dynamics** | Running min/max/mean of spread | (max_spread - min_spread) / mean_spread. Widening = market makers pulling back = volatility incoming. |
 | 13 | **Aggression Ratio** | Two counters (ask_trades, total) | trades_at_ask / total_trades (Lee-Ready). > 0.6 = strong buy pressure. |
 | 14 | **Queue Depletion Asymmetry** | Running depletion rates bid/ask | (bid_depletion - ask_depletion) / max. Reveals hidden institutional flow direction. |
-| 15 | **Depth Renewal Rate** | Counters (new_orders, cancels) at top 3 | new_placed / canceled. Market makers refreshing = confidence. Not refreshing = about to pull. |
+| 15 | **Order Count Flux** | Running delta of bid_ct/ask_ct at top 3 | Δbid_ct>0 → new order proxy, Δbid_ct<0 → cancel/fill proxy. Approximation: MBP-10 snapshots can't distinguish cancel from fill. Treat as market maker activity volatility — high flux = active refresh, low flux = about to pull. |
 
 ### NOVELS (new combinations)
 
@@ -83,8 +83,10 @@ pub struct IncrementalMicrostructureCalculator {
     prev_ask_sz_top3: u64,
     new_orders_top3: u64,                    // depth renewal counters
     cancels_top3: u64,
-    returns_first_half: Vec<f64>,            // for intra-bar momentum
-    returns_second_half: Vec<f64>,
+    // Intra-bar momentum: Welford online covariance (O(1) per tick, no allocation)
+    half1_sum_x: f64, half1_sum_x2: f64, half1_n: u64,
+    half2_sum_x: f64, half2_sum_x2: f64, half2_n: u64,
+    cross_sum_xy: f64,  // for correlation between halves
     ofi_prev: f64,                           // for acceleration
     ofi_delta_ema: f64,
     vpin_prev: f64,                          // for toxicity gradient
@@ -100,8 +102,8 @@ pub struct IncrementalMicrostructureCalculator {
 impl IncrementalMicrostructureCalculator {
     pub fn new(bar_duration_ns: u64) -> Self;
     pub fn update(&mut self, snapshot: &Mbp10Snapshot, trade: Option<&Trade>) -> ();
-    pub fn snapshot(&self) -> MicrostructureFeatures;  // [20] f64
-    pub fn reset(&mut self);  // call at bar boundary
+    pub fn snapshot(&self) -> MicrostructureFeatures;  // [20] f64 — pure read, non-mutating, idempotent
+    pub fn reset(&mut self);  // call at bar boundary — clears all running state
 }
 ```
 
@@ -136,6 +138,21 @@ const OFI_DIM: usize = 20;              // was 8
 const RECORD_F64_COUNT: usize = 66;      // was 54 (42+4+20)
 ```
 
+**CRITICAL: OFI_DIM=8 is load-bearing in 6+ files.** All must be updated atomically:
+
+| File | Location | What to change |
+|------|----------|----------------|
+| `fxcache.rs:124` | `validate()` | Reads `ofi_dim` from header — will reject old files. Code and data deploy atomically. |
+| `fxcache.rs:195,377` | `Vec<[f64; OFI_DIM]>` | Compile-time array generic — all callers cascade. |
+| `precompute_features.rs:329,342,434` | `[f64; 8]` literals | Change to `[f64; OFI_DIM]` or use `fxcache::OFI_DIM`. |
+| `gpu_experience_collector.rs:747` | `let ofi_dim = if ... { 8 }` | Hardcoded detection. Must use `OFI_DIM` constant. |
+| `cuda_pipeline/mod.rs:402,477,516` | `if ofi { 8 } else { 0 }` | Same hardcoded literal. |
+| `constructor.rs:52-53,255` | `raw_state_dim = 74` (42+8+16+8) | Must become `42+8+16+OFI_DIM`. Aligned: `(86+7)&!7 = 88`. |
+| `hyperopt/adapters/dqn.rs:1437` | Literal `8` | Use `OFI_DIM`. |
+| Test files | `[0.0; 8]`, `vec![[0.0; 8]; n]` | All OFI fixtures must use `OFI_DIM`. |
+
+**Deploy rule:** Code and fxcache data must deploy atomically. New code with OFI_DIM=20 will reject old fxcache (ofi_dim=8 in header). Regenerate fxcache BEFORE deploying new code, or deploy both in the same Argo workflow step.
+
 ---
 
 ## Precompute Pipeline
@@ -153,7 +170,7 @@ For each minute bar [bar_start, bar_end):
 
 The existing OFI calculation path in `precompute_features.rs` already loads MBP-10 snapshots per bar and calls `OFICalculator`. The change: replace `OFICalculator` with `IncrementalMicrostructureCalculator` which computes all 20 features (including the existing 8).
 
-**Precompute time:** 4M bars × ~2000 ticks/bar = 8B ticks × <1μs = ~8 seconds. Negligible.
+**Precompute time:** Calculator CPU time is ~8 seconds (8B ticks × <1μs). But disk I/O dominates — 148GB MBP-10 at ~500MB/s = ~5 minutes read time. Total precompute: **~5-10 minutes** depending on I/O. Set Argo workflow timeout accordingly.
 
 ---
 
@@ -283,7 +300,8 @@ impl TickProcessor {
 
 | Risk | Mitigation |
 |------|-----------|
-| Existing fxcache files incompatible | Regenerate all. Precompute is ~8 seconds for 4M bars. |
+| Existing fxcache files incompatible | Regenerate all (~5-10 min). Code and data deploy atomically — new code rejects old fxcache. |
+| OFI_DIM=8 hardcoded in 6+ files | All sites enumerated in spec. Must update atomically in single commit. Use shared constant, not literals. |
 | 12 new features → overfitting on training data | ISV feature gating suppresses irrelevant features. 12 features / 4M bars = negligible risk. |
 | IncrementalMicrostructureCalculator state drift | Reset at every bar boundary. Running statistics are bounded by definition. |
 | Databento live feed drops/gaps | Calculator handles missing ticks gracefully (EMA continues with last value). At bar close, features reflect whatever ticks arrived. |