104 lines
3.2 KiB
Markdown
104 lines
3.2 KiB
Markdown
# Intellectual Archaeology Depth Gate
|
|
|
|
Date: 2026-06-17
|
|
|
|
Status: draft rule hardening for `model_library_mvp`
|
|
|
|
## 1. Purpose
|
|
|
|
Intellectual Archaeology is a deep modeling model. It is not a default analysis mode, article summary mode, or ornament for ordinary questions.
|
|
|
|
The depth gate exists to prevent over-analysis and to force explicit stopping reasons.
|
|
|
|
## 2. When `should_call=false`
|
|
|
|
Set `should_call=false` when:
|
|
|
|
- the user only needs a fact lookup;
|
|
- the user asks for rewrite, translation, typo correction, or direct execution;
|
|
- QPI has not completed problem definition for an ambiguous problem;
|
|
- the input has deep words such as "底层", "模型", or "哲学" but the user intent is light;
|
|
- there is not enough material to separate real assumptions from generic abstraction;
|
|
- the analysis would not change judgment, route, verification, or action.
|
|
|
|
## 3. Depth Levels
|
|
|
|
Allowed `recommended_max_depth` values:
|
|
|
|
- `application`
|
|
- `domain`
|
|
- `process`
|
|
- `purpose`
|
|
- `core_mechanism`
|
|
- `human_capability`
|
|
- `philosophical_bedrock`
|
|
|
|
## 4. Depth Derivation
|
|
|
|
Use the shallowest level that can change the next action.
|
|
|
|
Stop at `application` when the request is about usage, examples, or immediate application.
|
|
|
|
Stop at `domain` when domain assumptions or contextual constraints explain the behavior.
|
|
|
|
Stop at `process` when the missing value is workflow, sequence, role boundary, or operating mechanism.
|
|
|
|
Stop at `purpose` when the problem is about what the system is trying to optimize or preserve.
|
|
|
|
Stop at `core_mechanism` when the request is to extract a reusable model or explain repeated failure.
|
|
|
|
Use `human_capability` only when the question depends on human cognition, judgment, agency, fatigue, attention, or embodied limitation.
|
|
|
|
Use `philosophical_bedrock` only when the problem explicitly involves value premises, epistemic assumptions, ontology of the problem, or the model's deepest ground.
|
|
|
|
## 5. Philosophical Bedrock Is Not Default
|
|
|
|
`philosophical_bedrock` must not be selected only because the input contains:
|
|
|
|
- "底层";
|
|
- "模型";
|
|
- "哲学";
|
|
- "第一性原理";
|
|
- "深层".
|
|
|
|
It is allowed only when deeper analysis changes the framing, validity, or action boundary.
|
|
|
|
## 6. Required Runtime Fields
|
|
|
|
The model-specific runtime contract requires:
|
|
|
|
- `should_call`
|
|
- `entry_reason`
|
|
- `recommended_max_depth`
|
|
- `layers_to_analyze`
|
|
- `stop_reason`
|
|
- `no_deeper_reason`
|
|
- `assumptions_by_layer`
|
|
- `validation_needed`
|
|
- `action_implication`
|
|
|
|
`value_of_deeper_analysis` may be added later as an optional field, but `no_deeper_reason` is required because it prevents depth inflation.
|
|
|
|
## 7. Stop Reason Templates
|
|
|
|
Use concrete stopping reasons:
|
|
|
|
- "Stop at purpose because the decision depends on what the system optimizes; deeper ontology would not change the next action."
|
|
- "Stop at core_mechanism because the reusable pattern is now explicit and validation can begin."
|
|
- "Stop before philosophical_bedrock because the input lacks evidence for value-premise or epistemic claims."
|
|
- "No call because QPI has not completed the problem frame."
|
|
- "No call because the user requested direct execution."
|
|
|
|
## 8. Status Boundary
|
|
|
|
This gate does not upgrade the model. Intellectual Archaeology remains:
|
|
|
|
```json
|
|
{
|
|
"status": "draft",
|
|
"stability_level": "B",
|
|
"regression_status": "pending"
|
|
}
|
|
```
|
|
|