Add the clock-schedule baseline: stage 10 by epoch 50, still at chance

README.md

@@ -45,18 +45,41 @@ Each is a full training state: `model`, `optimizer`, `epoch`, `stage`, `metrics`
 embeds the wandb run id (project `lyot-chain10`). All four were at curriculum stage 10 when
 saved. The best is `shuffle_nodes=true`, the shortcut-free setting.
 
+### The clock baseline (`c10clk-*`), at chance
+
+The four `c10clk-*` files are the control, and they are **chance-level by design, not
+broken**. Everything about them matches the arms above except that the curriculum advances
+on upstream's fixed schedule (`epochs_per_stage: 5`) rather than on measured accuracy:
+
+| file | uniform_prob | shuffle_nodes | epoch | eval/acc |
+|---|---|---|---|---|
+| `c10clk-...fpncimyu-ep162-s10-eval0.510.pt` | 0.5 | false | 162 | 0.510 |
+| `c10clk-...3yqogp0e-ep161-s10-eval0.509.pt` | 0.1 | false | 161 | 0.509 |
+| `c10clk-...zkz6ansr-ep158-s10-eval0.496.pt` | 0.1 | true | 158 | 0.496 |
+| `c10clk-...1vleof96-ep160-s10-eval0.488.pt` | 0.5 | true | 160 | 0.488 |
+
+Chance is 0.500. All four reached stage 10 by epoch 50 — three times sooner than the adaptive
+arms — and none exceeded 0.537 at any point in ~160 epochs. They are published because the
+depth-10 claim rests on this comparison: same task, same model, same `reset_optimizer=true`,
+differing only in how stages advance.
+
 ## What made depth 10 reachable
 
-Two changes over the standard recipe, from a 2x2x2x2 factorial:
+Two changes over the standard recipe, from a 2x2x2x2 factorial plus this baseline:
 
 1. **`reset_optimizer=true`** — rebuild AdamW every epoch, discarding both moment estimates.
-   This decides the outcome: all four arms with it reached stage 10 and >0.96 eval; **no arm
+   Necessary: all four adaptive arms with it reached stage 10 and >0.96 eval; **no arm
    without it passed stage 7 or 0.07 eval**, given ~2x the epochs. The curriculum keeps
    shifting the input distribution, and stale second moments appear to misscale the new
    stage's gradients.
 2. **Accuracy-driven staging** (`staging: revisit`, threshold 0.9, patience 5) — advance only
    after a stage passes 5 consecutive epochs at its own budget, and fall back to any earlier
-   stage that regresses. A clock schedule advances whether or not the stage was learned.
+   stage that regresses.
+
+Neither alone is enough. The `c10clk-*` arms have the optimizer reset and still sit at chance,
+so the reset is necessary but not sufficient; and no arm without the reset got near stage 10
+under either schedule. Arriving at the final stage early is worth nothing if the stages were
+never learned on the way.
 
 `uniform_prob` and `shuffle_nodes` only modulate how fast the schedule progresses; neither
 rescues a run without `reset_optimizer`.

c10clk-L2-u0.1-rtrue-sfalse-lyot-chain10_3yqogp0e-ep161-s10-eval0.509.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88d053353caf83ac61d4e67d72ad17a913ffee0463e6b67522cd3a2225ec618d
+size 179887739

c10clk-L2-u0.1-rtrue-strue-lyot-chain10_zkz6ansr-ep158-s10-eval0.496.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aef1123d81b5f79c9631b192c5a3790e9e7f3a8462740f6418079b8367950fb
+size 179887739

c10clk-L2-u0.5-rtrue-sfalse-lyot-chain10_fpncimyu-ep162-s10-eval0.510.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d26a7431cf7e46b96d3df52f5f25494871eb441b833f695142b21bfe0d4f2a69
+size 179887739

c10clk-L2-u0.5-rtrue-strue-lyot-chain10_1vleof96-ep160-s10-eval0.488.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:95d3568cd94e5e0afdb31588c70fc7152c5498a1fe9cfb4668147c9cc214a198
+size 179887739