You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This repository demonstrates a structural phenomenon:
8
-
9
-
**Information pre-positioning causes cascading execution failures in naive systems.**
10
-
11
-
When structured action bundles contain implicit authority, hidden defaults, ambiguous targets, or overloaded metadata, downstream execution engines can produce deterministic but unintended side effects.
12
-
13
-
This lab shows:
14
-
15
-
1. How syntactically valid inputs can corrupt state under naive execution.
16
-
2. Why runtime mitigation is insufficient.
17
-
3. Why a pre-execution admissibility gate is structurally necessary.
7
+
Deterministic failure harness: syntactically valid inputs corrupt state when there is no upstream admissibility gate.
18
8
19
9
---
20
10
21
-
## Why This Matters
22
-
23
-
Modern AI-integrated systems increasingly execute structured actions proposed by probabilistic models.
24
-
25
-
Most governance efforts focus on runtime mitigation — detecting and correcting undesirable states after partial execution.
26
-
27
-
This lab demonstrates a different architectural claim:
11
+
## Why This Exists
28
12
29
-
> If a system permits certain categories of state to form, mitigation becomes an ongoing operational burden.
This is not an alignment argument. It is a systems design argument.
13
+
Most AI system governance efforts apply runtime mitigation — detecting and correcting bad state after partial execution. This lab demonstrates why that is structurally insufficient. When structured action bundles carry implicit authority, hidden defaults, or ambiguous targets, a naive executor produces deterministic but unintended side effects. The failure is not randomness; it is an admissibility problem. This repo is the negative case that motivates the upstream [interpretation-boundary-lab](https://github.com/LalaSkye/interpretation-boundary-lab).
34
14
35
15
---
36
16
37
-
## What This Repository Is
38
-
39
-
- A deterministic simulation environment.
40
-
- A failure harness.
41
-
- A conformance test surface.
42
-
- A behavioural contract for admissibility gating.
17
+
## Architecture
43
18
44
-
---
45
-
46
-
## What This Repository Is Not
47
-
48
-
- It does **NOT** include the Trinity gate implementation.
49
-
- It does **NOT** expose gating logic.
50
-
- It does **NOT** implement scoring heuristics.
51
-
- It does **NOT** provide production security tooling.
19
+
```
20
+
Input Bundle
21
+
|
22
+
v
23
+
┌─────────────────────┐
24
+
│ naive_executor.py │ <-- no gate, no admissibility check
25
+
└─────────────────────┘
26
+
|
27
+
[state corruption]
28
+
|
29
+
┌───────┴────────┐
30
+
v v
31
+
contaminated clean
32
+
case_1..6.json case_1.json
33
+
|
34
+
v
35
+
reports/example_trace_without_gate.txt
36
+
```
52
37
53
-
This is a boundary demonstration only.
38
+
```
39
+
Input Bundle
40
+
|
41
+
v
42
+
┌──────────────────────┐
43
+
│ gate_api/interface.py│ <-- implement Gate here
44
+
└──────────────────────┘
45
+
|
46
+
[admitted or rejected]
47
+
|
48
+
v
49
+
┌─────────────────────┐
50
+
│ executor pipeline │
51
+
└─────────────────────┘
52
+
|
53
+
v
54
+
reports/example_trace_with_gate.txt
55
+
```
54
56
55
57
---
56
58
57
-
## Core Concept
58
-
59
-
### Information Pre-Positioning
59
+
## Quick Demo
60
60
61
-
Information pre-positioning occurs when structured data embeds executable consequences that are not explicit at the point of execution.
62
-
63
-
Examples include:
64
-
65
-
- Implicit privilege escalation fields.
66
-
- Hidden default behaviours.
67
-
- Ambiguous resource targets.
68
-
- Metadata that alters execution ordering.
69
-
- Conflation of descriptive and authoritative fields.
70
-
71
-
A naive executor processes these structures deterministically, yet produces state corruption.
72
-
73
-
The failure is not randomness. The failure is **admissibility**.
naive_executor.py # executor with no admissibility gate
114
117
/tests
115
-
__init__.py
116
-
conftest.py
117
118
test_prepositioning_failures.py
118
119
test_gate_contract.py
119
120
/reports
@@ -123,32 +124,75 @@ This repository illustrates why the latter reduces systemic workload and failure
123
124
124
125
---
125
126
126
-
## Running Tests
127
+
## Core Concept: Information Pre-Positioning
127
128
128
-
```bash
129
-
pip install pytest
130
-
python -m pytest
131
-
```
129
+
Information pre-positioning occurs when structured data embeds executable consequences that are not explicit at the point of execution. Examples:
130
+
131
+
- Implicit privilege escalation fields
132
+
- Hidden default behaviours
133
+
- Ambiguous resource targets
134
+
- Metadata that alters execution ordering
135
+
- Conflation of descriptive and authoritative fields
136
+
137
+
A naive executor processes these structures deterministically, yet produces state corruption. The failure is not randomness. The failure is **admissibility**.
138
+
139
+
---
140
+
141
+
## Architectural Distinction
142
+
143
+
| Approach | Mechanism | Consequence |
144
+
|---|---|---|
145
+
| Runtime mitigation | Detects and handles problematic states *after* they form | Ongoing operational burden; failure surface grows with system complexity |
146
+
| Pre-execution admissibility gating | Prevents problematic state categories from *entering* the system | Reduces mitigation workload downstream; failure surface is bounded upstream |
147
+
148
+
> If a system requires mitigation logic at runtime, it has already admitted instability. Admissibility is upstream of enforcement.
149
+
150
+
---
151
+
152
+
## What This Repository Is
153
+
154
+
- A deterministic simulation environment
155
+
- A failure harness for six contamination patterns
156
+
- A conformance test surface for gate implementations
157
+
- A behavioural contract for admissibility gating
158
+
159
+
## What This Repository Is Not
160
+
161
+
- It does **not** include the Trinity gate implementation
162
+
- It does **not** expose gating logic
163
+
- It does **not** implement scoring heuristics
164
+
- It does **not** provide production security tooling
165
+
166
+
This is a boundary demonstration only.
132
167
133
168
---
134
169
135
-
## Extension
170
+
## Extending with a Real Gate
136
171
137
-
To test a real admissibility gate:
172
+
To test an admissibility gate against the conformance suite:
138
173
139
-
1. Implement the `Gate` interface defined in `gate_api/interface.py`.
140
-
2. Inject it into the executor pipeline.
141
-
3. Ensure all conformance tests pass.
174
+
1. Implement the `Gate` interface defined in `gate_api/interface.py`
175
+
2. Inject it into the executor pipeline
176
+
3. Ensure all conformance tests pass
142
177
143
178
---
144
179
145
-
## Position
180
+
## Part of the Execution Boundary Series
146
181
147
-
> If a system requires mitigation logic at runtime, it has already admitted instability.
148
-
> Admissibility is upstream of enforcement.
182
+
| Repo | Layer | What It Does |
183
+
|---|---|---|
184
+
|[interpretation-boundary-lab](https://github.com/LalaSkye/interpretation-boundary-lab)| Upstream boundary | 10-rule admissibility gate for interpretations |
185
+
|[dual-boundary-admissibility-lab](https://github.com/LalaSkye/dual-boundary-admissibility-lab)| Full corridor | Dual-boundary model with pressure monitoring and C-sector rotation |
0 commit comments