implement Addendum 8: 6-arm token-space recurrence suite (arms A, B, C, D1, D2, D3), tests, and jobs/e8.csv

1 files changed, 17 insertions, 4 deletions

diff --git a/src/config.py b/src/config.py
index 945c259..915cdae 100644
--- a/src/config.py
+++ b/src/config.py
@@ -25,6 +25,9 @@ class Config:
     min_steps: int = 2               # ACT: halt prob forced to 0 for the first min_steps-1 steps
     n_layers: int = 2
     n_heads: int = 4
+    # scratchpad / token-space recurrence (Addendum 8, E8)
+    scratch_mode: str = "none"       # "none" | "structured" | "filler" | "random_learned" | "random_frozen" | "random_noise"
+    scratch_len: int = 16            # length of filler / random token sequence
     # training
     lr: float = 1e-3
     lr_decay: bool = False           # cosine 1e-3 -> 1e-4 over the run (E3)
@@ -45,7 +48,11 @@ class Config:
 
     @property
     def vocab(self) -> int:
-        """Token count. Integers mode: value tokens 0..next_prime(range_end), plus EOS (+1) and pad (+1)."""
+        """Token count.
+        - Integers mode: value tokens 0..next_prime(range_end) + EOS + pad.
+        - Digits mode (none): 10 digits + EOS = 11.
+        - Digits mode (scratchpad): 10 digits + EOS + SEP + PAUSE + 16 random (a-p) + 4 structured symbols (c,=,d,:) + 1 noise slot = 34.
+        """
         if self.vocab_mode == "integers":
             # inline mini-sieve: next prime above range_end bounds the target domain
             limit = self.range_end + 100
@@ -57,15 +64,21 @@ class Config:
                         is_prime[m] = False
             np_ = next(i for i in range(self.range_end + 1, limit + 1) if is_prime[i])
             return np_ + 2                    # values 0..np_, EOS=np_+1, pad=np_+2
-        return 11                              # digits 0-9 + EOS
+        if self.scratch_mode != "none":
+            return 34                         # expanded vocabulary for scratchpad/filler/random tokens
+        return 11                             # digits 0-9 + EOS
 
     @property
     def eos_id(self) -> int:
-        return self.vocab - 1
+        if self.vocab_mode == "integers":
+            return self.vocab - 1
+        if self.scratch_mode != "none":
+            return 10                         # fixed ID 10 for EOS in digits mode
+        return self.vocab - 1                 # 10 for digits mode none
 
     @property
     def pad_id(self) -> int:
-        return self.vocab              # one extra embedding row reserved for pad
+        return self.vocab                     # one extra embedding row reserved for pad
 
     def to_json(self) -> dict:
         d = {f.name: getattr(self, f.name) for f in fields(self)}