Research-Stack/4-Infrastructure/shim/joke_source_literalization_guardrail_probe.py
2026-05-11 22:18:31 -05:00

326 lines
12 KiB
Python

#!/usr/bin/env python3
"""Receipt-backed guardrail for joke sources with unsafe literal payloads.
The project can cite a joke as a real source prompt, but that does not mean the
literal payload becomes an executable instruction. This probe adds the missing
guardrail for encoding systems:
* parody/satire may be retained as provenance metadata;
* unsafe domain advice must not be promoted into actionable logograms;
* compression may preserve the existence and role of the source without
preserving operationally harmful details as callable procedures.
"""
from __future__ import annotations
import hashlib
import json
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
REPO = Path(__file__).resolve().parents[2]
OUT_DIR = REPO / "shared-data" / "data" / "joke_source_literalization_guardrail"
REGISTRY = OUT_DIR / "joke_source_literalization_guardrail_registry.json"
RECEIPT = OUT_DIR / "joke_source_literalization_guardrail_receipt.json"
SUMMARY = OUT_DIR / "joke_source_literalization_guardrail.md"
TIDDLER = REPO / "6-Documentation" / "tiddlywiki-local" / "wiki" / "tiddlers" / "Joke Source Literalization Guardrail.tid"
SOURCE_REFS = [
REPO / "shared-data" / "data" / "bibliographic_event_horizon" / "bibliographic_event_horizon_receipt.json",
REPO / "shared-data" / "data" / "asymptotic_closure_horizon" / "asymptotic_closure_horizon_receipt.json",
REPO / "shared-data" / "data" / "hutter_torsion_clock_adaptation" / "hutter_torsion_clock_adaptation_receipt.json",
REPO / "6-Documentation" / "docs" / "specs" / "FORWARD_FOUNDATION_EQUATION_COMPILER.md",
REPO / "6-Documentation" / "docs" / "specs" / "OMINDIRECTION_LOGOGRAM_DESIGN_AND_COMPILER.md",
]
LITERALIZATION_HORIZON = 1000
def stable_json(obj: Any) -> str:
return json.dumps(obj, sort_keys=True, separators=(",", ":"), ensure_ascii=True)
def sha256_bytes(data: bytes) -> str:
return hashlib.sha256(data).hexdigest()
def hash_obj(obj: Any) -> str:
return sha256_bytes(stable_json(obj).encode("utf-8"))
def rel(path: Path) -> str:
try:
return str(path.relative_to(REPO))
except ValueError:
return str(path)
def file_hash(path: Path) -> str | None:
return sha256_bytes(path.read_bytes()) if path.exists() else None
def source_ref(path: Path) -> dict[str, Any]:
return {"path": rel(path), "exists": path.exists(), "sha256": file_hash(path)}
def route(
*,
route_id: str,
payload_role: str,
domain_risk: int,
satire_marker: bool,
actionable_literalization: bool,
safety_context_present: bool,
preserves_operational_steps: bool,
) -> dict[str, Any]:
literalization_risk = (
domain_risk
+ (350 if satire_marker and actionable_literalization else 0)
+ (350 if preserves_operational_steps else 0)
+ (250 if not safety_context_present else 0)
)
horizon = literalization_risk >= LITERALIZATION_HORIZON
if horizon and actionable_literalization:
decision = "QUARANTINE_UNSAFE_LITERALIZATION"
elif horizon:
decision = "HOLD_SAFETY_CONTEXT_REQUIRED"
else:
decision = "ADMIT_METADATA_ONLY"
item = {
"route_id": route_id,
"payload_role": payload_role,
"domain_risk": domain_risk,
"satire_marker": satire_marker,
"actionable_literalization": actionable_literalization,
"safety_context_present": safety_context_present,
"preserves_operational_steps": preserves_operational_steps,
"literalization_risk": literalization_risk,
"literalization_horizon": horizon,
"decision": decision,
}
item["route_hash"] = hash_obj({k: v for k, v in item.items() if k != "route_hash"})
return item
def build_registry() -> dict[str, Any]:
routes = [
route(
route_id="joke_post_as_source_prompt",
payload_role="provenance_metadata",
domain_risk=200,
satire_marker=True,
actionable_literalization=False,
safety_context_present=True,
preserves_operational_steps=False,
),
route(
route_id="unsafe_chemistry_tips_as_logogram",
payload_role="callable_procedure_candidate",
domain_risk=650,
satire_marker=True,
actionable_literalization=True,
safety_context_present=False,
preserves_operational_steps=True,
),
route(
route_id="unsafe_chemistry_tips_as_training_text",
payload_role="raw_text_corpus_payload",
domain_risk=650,
satire_marker=True,
actionable_literalization=True,
safety_context_present=False,
preserves_operational_steps=True,
),
route(
route_id="guardrail_summary",
payload_role="non_operational_safety_summary",
domain_risk=250,
satire_marker=True,
actionable_literalization=False,
safety_context_present=True,
preserves_operational_steps=False,
),
]
return {
"schema": "joke_source_literalization_guardrail_registry_v1",
"source_prompt": {
"id": "user_supplied_unsafe_chemistry_joke_image",
"role": "source_prompt",
"status": "user_supplied_image_prompt",
"description": (
"A joke post framed as chemistry advice. Its value is as a guardrail "
"example: satire can be cited as provenance, but unsafe procedural "
"content must not be promoted into executable encoding atoms."
),
"verbatim_payload_policy": "do_not_reproduce_operational_steps",
},
"source_refs": [source_ref(path) for path in SOURCE_REFS],
"claim_boundary": (
"Joke-source literalization guardrail only. The source may be retained as "
"metadata and cited as an encoding-risk example. Unsafe literal payloads "
"are not admitted as callable procedures, training targets, or logograms."
),
"encoding_rule": {
"metadata_channel": "ADMIT source identity, satire role, risk label, and receipt hash",
"payload_channel": "QUARANTINE operational unsafe steps",
"summary_channel": "ADMIT non-operational safety summary",
"logogram_channel": "HOLD or QUARANTINE if a glyph would expand into unsafe procedure",
},
"admissibility_equation": (
"A_joke_payload=1[source_role_metadata] * 1[not actionable_literalization] * "
"1[not preserves_operational_steps] * 1[safety_context_present]"
),
"hutter_adaptation": {
"codec_torsion_component": "unsafe_literalization_debt",
"rule": "a corpus atom that compresses into harmful procedural replay adds horizon-level torsion unless quarantined",
"safe_compression": "preserve citation and risk class, not executable unsafe procedure",
},
"routes": routes,
"aggregates": {
"route_count": len(routes),
"metadata_admit_count": sum(1 for item in routes if item["decision"] == "ADMIT_METADATA_ONLY"),
"quarantine_count": sum(1 for item in routes if item["decision"] == "QUARANTINE_UNSAFE_LITERALIZATION"),
"hold_count": sum(1 for item in routes if item["decision"].startswith("HOLD")),
"literalization_horizon": LITERALIZATION_HORIZON,
},
}
def build_receipt(registry: dict[str, Any]) -> dict[str, Any]:
receipt = {
"schema": "joke_source_literalization_guardrail_receipt_v1",
"generated_at_utc": datetime.now(timezone.utc).isoformat(),
"timestamp_role": "metadata_only",
"generated_at_utc_included_in_receipt_hash": False,
"registry": rel(REGISTRY),
"registry_hash": hash_obj(registry),
"aggregates": registry["aggregates"],
"decision": "ADMIT_JOKE_SOURCE_METADATA_QUARANTINE_UNSAFE_LITERALIZATION",
"claim_boundary": registry["claim_boundary"],
}
receipt["receipt_hash"] = sha256_bytes(
stable_json({k: v for k, v in receipt.items() if k not in {"receipt_hash", "generated_at_utc"}}).encode("utf-8")
)
return receipt
def write_summary(registry: dict[str, Any], receipt: dict[str, Any]) -> None:
lines = [
"# Joke Source Literalization Guardrail",
"",
f"Decision: `{receipt['decision']}` ",
f"Receipt hash: `{receipt['receipt_hash']}`",
"",
registry["claim_boundary"],
"",
"## Rule",
"",
f"`{registry['admissibility_equation']}`",
"",
"## Encoding Channels",
"",
]
for key, value in registry["encoding_rule"].items():
lines.append(f"- `{key}`: {value}")
lines.extend(
[
"",
"## Routes",
"",
"| Route | Payload role | Risk | Decision |",
"|---|---|---:|---|",
]
)
for item in registry["routes"]:
lines.append(f"| `{item['route_id']}` | `{item['payload_role']}` | {item['literalization_risk']} | `{item['decision']}` |")
lines.extend(["", "## Hutter Adaptation", ""])
for key, value in registry["hutter_adaptation"].items():
lines.append(f"- `{key}`: {value}")
lines.extend(["", "## Source Refs", ""])
for source in registry["source_refs"]:
lines.append(f"- `{source['path']}` exists: `{source['exists']}`")
SUMMARY.write_text("\n".join(lines) + "\n", encoding="utf-8")
def write_tiddler(receipt: dict[str, Any]) -> None:
text = f"""created: 20260509000000000
modified: 20260509000000000
tags: ResearchStack Safety Encoding Guardrail Receipt
title: Joke Source Literalization Guardrail
type: text/vnd.tiddlywiki
! Joke Source Literalization Guardrail
Durable runner:
```
4-Infrastructure/shim/joke_source_literalization_guardrail_probe.py
```
Receipt:
```
{rel(RECEIPT)}
```
Receipt hash:
```
{receipt['receipt_hash']}
```
!! Doctrine
A joke source can be a real citation, but unsafe literal payloads cannot become executable encoding atoms.
```
metadata channel -> ADMIT
unsafe payload channel -> QUARANTINE
summary channel -> ADMIT only if non-operational
logogram channel -> HOLD/QUARANTINE if expansion becomes unsafe procedure
```
!! Hutter Link
Unsafe literalization is codec torsion. If a compressed atom expands into harmful procedural replay, it stays outside the candidate archive unless quarantined or transformed into non-operational metadata.
!! Links
* [[Bibliographic Event Horizon]]
* [[Asymptotic Closure Horizon]]
* [[Hutter Torsion Clock Adaptation]]
* [[Omindirection Logogram Contract]]
"""
TIDDLER.write_text(text, encoding="utf-8")
def main() -> int:
OUT_DIR.mkdir(parents=True, exist_ok=True)
registry = build_registry()
receipt = build_receipt(registry)
REGISTRY.write_text(json.dumps(registry, indent=2, sort_keys=True) + "\n", encoding="utf-8")
RECEIPT.write_text(json.dumps(receipt, indent=2, sort_keys=True) + "\n", encoding="utf-8")
write_summary(registry, receipt)
write_tiddler(receipt)
print(
json.dumps(
{
"registry": rel(REGISTRY),
"receipt": rel(RECEIPT),
"summary": rel(SUMMARY),
"tiddler": rel(TIDDLER),
"receipt_hash": receipt["receipt_hash"],
"decision": receipt["decision"],
"aggregates": registry["aggregates"],
},
indent=2,
sort_keys=True,
)
)
return 0
if __name__ == "__main__":
raise SystemExit(main())