diff --git a/.claude/hooks/install-toolchains.sh b/.claude/hooks/install-toolchains.sh
index bd127b8..b9d1ec0 100755
--- a/.claude/hooks/install-toolchains.sh
+++ b/.claude/hooks/install-toolchains.sh
@@ -75,13 +75,16 @@ else
 fi
 
 # ---------------------------------------------------------------------------
-# R + jsonlite  (economy organ: M3a statistical sims)
+# R + HiddenMarkov  (economy organ: M3a statistical sims)
 # ---------------------------------------------------------------------------
-if command -v Rscript >/dev/null 2>&1 && Rscript -e 'library(jsonlite)' >/dev/null 2>&1; then
-  log "R + jsonlite already present; skipping."
+if command -v Rscript >/dev/null 2>&1 && Rscript -e 'library(HiddenMarkov)' >/dev/null 2>&1; then
+  log "R + HiddenMarkov already present; skipping."
 else
-  log "Installing r-base-core r-cran-jsonlite via apt-get ..."
-  sudo apt-get install -y r-base-core r-cran-jsonlite || die "apt-get install of r-base-core/r-cran-jsonlite failed."
+  log "Installing r-base-core via apt-get ..."
+  sudo apt-get install -y r-base-core || die "apt-get install of r-base-core failed."
+  log "Installing HiddenMarkov from CRAN ..."
+  Rscript -e 'install.packages("HiddenMarkov", repos="https://cloud.r-project.org", quiet=TRUE)' \
+    || die "CRAN install of HiddenMarkov failed."
 fi
 
 # ---------------------------------------------------------------------------
diff --git a/core/docs/plans/M3a-statistical-sims.md b/core/docs/plans/M3a-statistical-sims.md
index 153caeb..ea4cd82 100644
--- a/core/docs/plans/M3a-statistical-sims.md
+++ b/core/docs/plans/M3a-statistical-sims.md
@@ -17,10 +17,11 @@ for crypto markets C1.
 
 ## 3. Language & location
 **R 4.x** (apt `r-base-core`) · `src/economy/sims/statistical/`. Minimal dependencies:
-`r-base-core` + `jsonlite` (Hub JSON protocol). All statistical models — GARCH, Heston SDE,
-HMM, DCC, copula, jump-diffusion, fBM — are hand-rolled using base R primitives (`optim`,
-`fft`, `arima`, matrix ops). No CRAN statistical packages. Fractional Brownian motion via
-spectral methods (Hosking 1984 / Wood & Chan 1994) uses base R `fft()`.
+`r-base-core` + `HiddenMarkov` (CRAN — Viterbi filter, forward-backward, Baum-Welch). GARCH,
+Heston SDE, DCC, copula, jump-diffusion, and fBM are hand-rolled using base R primitives
+(`optim`, `fft`, `arima`, matrix ops). JSON I/O for the Hub stdin/stdout protocol is
+hand-rolled. Fractional Brownian motion via spectral methods (Hosking 1984 / Wood & Chan 1994)
+uses base R `fft()`.
 
 ## 4. Does / does-not
 - **Does:** run Monte Carlo price simulations (GBM, Merton jump-diffusion, Heston stochastic