yf176 commited on
Commit ·
0887bee
1
Parent(s): 7e4b3af
docs: draft V8.0 routing and artifact ACL contracts
Browse files- docs/polyhedraV8.0.md +461 -0
docs/polyhedraV8.0.md
ADDED
|
@@ -0,0 +1,461 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Polyhedra V8.0 - PRD (Structured Agent Collaboration)
|
| 2 |
+
|
| 3 |
+
> Discovery draft. Paper-trading only. No real capital. Not financial advice.
|
| 4 |
+
> V8.0 inherits the existing quant product behavior while replacing implicit
|
| 5 |
+
> agent sequencing and file conventions with explicit runtime contracts.
|
| 6 |
+
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
## Vision
|
| 10 |
+
|
| 11 |
+
V8.0 moves Polyhedra from a primarily linear chain of quant agents to
|
| 12 |
+
collaboration through the session folder. The collaboration model is governed
|
| 13 |
+
by three structured contracts:
|
| 14 |
+
|
| 15 |
+
1. A Routing Table that declares agent dependencies and controls activation.
|
| 16 |
+
2. A session-artifact ACL that controls which artifacts each agent may read,
|
| 17 |
+
write, edit, and list.
|
| 18 |
+
3. Typed artifact schemas that define agent inputs and outputs.
|
| 19 |
+
|
| 20 |
+
The Routing Table is the workflow authority. The runtime evaluates it as a
|
| 21 |
+
dependency graph, but the PRD and UI present it as a table. An LLM performs
|
| 22 |
+
reasoning inside an activated agent, but it does not select the next workflow
|
| 23 |
+
agent.
|
| 24 |
+
|
| 25 |
+
---
|
| 26 |
+
|
| 27 |
+
## Inherited Product Contract
|
| 28 |
+
|
| 29 |
+
V8.0 preserves the existing quant product and its session-based YOLO behavior:
|
| 30 |
+
|
| 31 |
+
- The first YOLO run in a session creates a complete baseline.
|
| 32 |
+
- Later YOLO runs in the same session reuse prior context and focus on changes.
|
| 33 |
+
- RAG remains available and must obey the same artifact ACL as every agent.
|
| 34 |
+
- Finley remains the final user-facing output layer.
|
| 35 |
+
- The product remains paper-trading only.
|
| 36 |
+
|
| 37 |
+
V8.0 focuses only on quant agents. Academic and research agents are outside the
|
| 38 |
+
V8.0 product scope.
|
| 39 |
+
|
| 40 |
+
---
|
| 41 |
+
|
| 42 |
+
## Current Architecture Baseline
|
| 43 |
+
|
| 44 |
+
V8.0 starts from the current implementation rather than replacing the quant
|
| 45 |
+
product with a separate workflow.
|
| 46 |
+
|
| 47 |
+
- Every agent currently implements the common runtime call shape
|
| 48 |
+
`handle(mention, thread_context, research_state)` and returns an
|
| 49 |
+
`AgentResponse` envelope.
|
| 50 |
+
- Quant output models exist, but typed outputs are not populated and enforced
|
| 51 |
+
consistently across all agents.
|
| 52 |
+
- Most agent-to-agent inputs are markdown artifacts selected by filename
|
| 53 |
+
convention rather than validated typed artifact contracts.
|
| 54 |
+
- The coordinator injects a session-scoped `FileWriter` into each agent.
|
| 55 |
+
Agent code can currently read, write, and list session-relative files without
|
| 56 |
+
a per-agent permission contract.
|
| 57 |
+
- LLMs do not receive direct filesystem access. Agent implementation code
|
| 58 |
+
builds prompts from files and performs writes on behalf of the LLM.
|
| 59 |
+
- Next-agent selection is currently distributed across deterministic code,
|
| 60 |
+
LLM routing output, prompt instructions, and cycle guards.
|
| 61 |
+
- The session folder already acts as the persistent context space for raw data,
|
| 62 |
+
analyses, proposals, traces, RAG material, and final output.
|
| 63 |
+
|
| 64 |
+
V8.0 formalizes these existing mechanisms. It does not create a second product
|
| 65 |
+
pipeline beside them.
|
| 66 |
+
|
| 67 |
+
---
|
| 68 |
+
|
| 69 |
+
## Confirmed V8.0 Decisions
|
| 70 |
+
|
| 71 |
+
| Area | Confirmed decision |
|
| 72 |
+
|---|---|
|
| 73 |
+
| Product scope | Polyhedra V8.0 is quant-only. Academic and research agents are removed from the V8.0 product runtime and workflow. |
|
| 74 |
+
| Collaboration model | Quant agents collaborate through artifacts in the session folder instead of relying on one implicit linear mention chain. |
|
| 75 |
+
| Workflow authority | A structured Routing Table controls activation. LLM routing output no longer selects the next workflow agent. |
|
| 76 |
+
| Existing behavior | The Routing Table adopts the current quant logic unless a dependency is explicitly reviewed and changed in this PRD. |
|
| 77 |
+
| Table contents | Agent entries use only exact registered agent names. Inputs, services, helpers, versions, markets, symbols, and humans are not agent entries. |
|
| 78 |
+
| Parallel execution | Agents whose Routing Table dependencies are satisfied may be activated in parallel. |
|
| 79 |
+
| File access | Each agent receives a structured session-artifact ACL defining what it may read, create, overwrite/edit, and list. |
|
| 80 |
+
| Enforcement | Runtime code controls artifact access through the injected file interface. An agent or LLM is not trusted to enforce ACL rules from prompt text. |
|
| 81 |
+
| Contract violations | Unauthorized file access is prevented by construction and covered by contract tests; it is not designed as a normal agent decision or product branch. |
|
| 82 |
+
| Agent I/O | Every quant agent is migrated to typed input and output artifact schemas in V8.0. |
|
| 83 |
+
| Migration | All quant workflow agents migrate to the V8.0 contracts together; V8.0 does not leave a permanent mixed governed/ungoverned workflow. |
|
| 84 |
+
| Artifact storage | Final storage layout follows from the approved dependency, ACL, and typed-schema contracts. It is not designed independently first. |
|
| 85 |
+
| Activation | Each agent declares a supported scope. At the beginning of a run, the runtime calculates every agent's effective scope and freezes the resulting active routing plan. Required dependency outputs then determine when each active agent can run. |
|
| 86 |
+
| RAG | RAG remains part of the quant system and may expose only artifacts permitted by the requesting agent's ACL. |
|
| 87 |
+
| Human role | No human is an agent entry or participant inside the V8.0 Routing Table. Finley remains the product-facing output layer, and the system remains paper-trading only. |
|
| 88 |
+
| UI | The existing UI remains and is extended to show current agent progress, the Routing Table, and ACL information. |
|
| 89 |
+
| Design process | The Routing Table, ACL, and schemas are reviewed one agent group at a time before implementation. |
|
| 90 |
+
|
| 91 |
+
---
|
| 92 |
+
|
| 93 |
+
## Structural Requirements
|
| 94 |
+
|
| 95 |
+
### Routing Table
|
| 96 |
+
|
| 97 |
+
- The Routing Table contains agent entries only.
|
| 98 |
+
- Every agent entry uses its exact registered name.
|
| 99 |
+
- Dependency columns represent required agent outputs, not version labels,
|
| 100 |
+
ticker branches, helper services, session inputs, or LLM-selected routing.
|
| 101 |
+
- Symbol and market applicability are declared as agent supported scope, not
|
| 102 |
+
workflow branches.
|
| 103 |
+
- Independent agents may run in parallel when their required dependencies are
|
| 104 |
+
satisfied.
|
| 105 |
+
- The Routing Table will be defined and approved one agent group at a time.
|
| 106 |
+
- At the beginning of each YOLO run, the runtime calculates each agent's
|
| 107 |
+
effective scope as the intersection of requested tickers and that agent's
|
| 108 |
+
supported scope.
|
| 109 |
+
- An agent is active when its effective scope is not empty.
|
| 110 |
+
- A dependency applies only to the overlapping effective scope of its upstream
|
| 111 |
+
and downstream agents.
|
| 112 |
+
- The resulting active routing plan is fixed for that run. Agent or LLM output cannot
|
| 113 |
+
add, remove, or reroute workflow nodes while the run is in progress.
|
| 114 |
+
- A later YOLO run initializes a new active routing plan. Changes to ticker
|
| 115 |
+
scope take effect in that later run rather than mutating a plan already in
|
| 116 |
+
progress.
|
| 117 |
+
- Agent prompts may request more evidence or report missing input, but they do
|
| 118 |
+
not activate another workflow agent directly.
|
| 119 |
+
- The runtime evaluates dependency completion and activation conditions.
|
| 120 |
+
- Existing LLM routing JSON may remain during migration only as non-authoritative
|
| 121 |
+
diagnostic output; it must not control V8.0 workflow execution.
|
| 122 |
+
- The Routing Table is the runtime workflow definition. It is distinct from the
|
| 123 |
+
`EventRouter` agent, which assigns detected event evidence to in-scope
|
| 124 |
+
tickers but does not activate or skip workflow agents.
|
| 125 |
+
|
| 126 |
+
### Session-Artifact ACL
|
| 127 |
+
|
| 128 |
+
- Each agent receives explicit `read`, `create`, `edit`, and `list` permissions.
|
| 129 |
+
- Permissions are defined in a structured runtime contract rather than being
|
| 130 |
+
scattered across prompts and agent implementation code.
|
| 131 |
+
- The runtime gives an agent only the artifact operations allowed by its ACL.
|
| 132 |
+
- LLMs do not receive direct filesystem access.
|
| 133 |
+
- Prompt construction and RAG retrieval use the same ACL-filtered artifact view.
|
| 134 |
+
- Agent implementations must not receive an unrestricted session-folder file
|
| 135 |
+
interface after migration.
|
| 136 |
+
- Per-agent artifact permissions will be approved one agent at a time; they are not
|
| 137 |
+
inferred from current unrestricted access.
|
| 138 |
+
|
| 139 |
+
#### ACL Fields
|
| 140 |
+
|
| 141 |
+
The Routing Table determines when an agent runs. The ACL determines which
|
| 142 |
+
artifacts it can read and change, including context retrieved through RAG.
|
| 143 |
+
|
| 144 |
+
| Field | Meaning |
|
| 145 |
+
|---|---|
|
| 146 |
+
| `read` | Artifacts allowed as input. |
|
| 147 |
+
| `create` | Artifacts the agent may create. |
|
| 148 |
+
| `edit` | Existing artifacts the agent may update. |
|
| 149 |
+
| `list` | Session-relative filenames the agent may discover through enumeration. |
|
| 150 |
+
|
| 151 |
+
Every permission field is a list. `[]` grants no permission for that operation.
|
| 152 |
+
Permissions apply within the active session and the agent's effective ticker
|
| 153 |
+
scope. Unlisted artifacts are inaccessible through the runtime file interface.
|
| 154 |
+
All entries are filenames or filename templates relative to the session folder,
|
| 155 |
+
including subdirectories. They are not logical artifact identifiers.
|
| 156 |
+
`{ticker}` resolves only to tickers in the agent's effective scope; `{date}`
|
| 157 |
+
identifies the artifact date within the session, including prior dates needed
|
| 158 |
+
for delta comparisons. Listing returns only filenames allowed by `list`;
|
| 159 |
+
reading their contents separately requires `read` permission.
|
| 160 |
+
|
| 161 |
+
#### ACL Table
|
| 162 |
+
|
| 163 |
+
These permissions are the V8.0 draft based on existing artifact usage. The
|
| 164 |
+
table covers all 19 registered quant agents; it does not claim that runtime ACL
|
| 165 |
+
enforcement is already implemented.
|
| 166 |
+
|
| 167 |
+
| Agent | `read` | `create` | `edit` | `list` |
|
| 168 |
+
|---|---|---|---|---|
|
| 169 |
+
| `DataFetcher` | `[quant/raw/{ticker}_{date}.json, quant/raw/{ticker}_{date}_summary.md]` | `[quant/raw/{ticker}_{date}.json, quant/raw/{ticker}_{date}_summary.md]` | `[quant/raw/{ticker}_{date}.json, quant/raw/{ticker}_{date}_summary.md]` | `[quant/raw/{ticker}_{date}.json, quant/raw/{ticker}_{date}_summary.md]` |
|
| 170 |
+
| `OptionsFlowWatcher` | `[]` | `[quant/options_flow_{date}.md]` | `[quant/options_flow_{date}.md]` | `[]` |
|
| 171 |
+
| `TrendAnalyst` | `[quant/raw/{ticker}_{date}_summary.md]` | `[quant/trend_analysis_{date}.md]` | `[quant/trend_analysis_{date}.md]` | `[quant/raw/{ticker}_{date}_summary.md]` |
|
| 172 |
+
| `SentimentAnalyst` | `[quant/raw/{ticker}_{date}_summary.md, quant/trend_analysis_{date}.md]` | `[quant/sentiment_analysis_{date}.md]` | `[quant/sentiment_analysis_{date}.md]` | `[quant/raw/{ticker}_{date}_summary.md]` |
|
| 173 |
+
| `MarketEventWatcher` | `[quant/raw/{ticker}_{date}_summary.md, quant/trend_analysis_{date}.md]` | `[quant/market_events_{date}.md]` | `[quant/market_events_{date}.md]` | `[quant/raw/{ticker}_{date}_summary.md]` |
|
| 174 |
+
| `MacroContextWatcher` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md]` | `[quant/macro_context_{date}.md]` | `[quant/macro_context_{date}.md]` | `[]` |
|
| 175 |
+
| `UnusualEventDetector` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md, quant/macro_context_{date}.md]` | `[quant/unusual_events_{date}.md]` | `[quant/unusual_events_{date}.md]` | `[]` |
|
| 176 |
+
| `EventRouter` | `[quant/unusual_events_{date}.md, quant/market_events_{date}.md, quant/options_flow_{date}.md]` | `[quant/event_route_{date}.md]` | `[quant/event_route_{date}.md]` | `[]` |
|
| 177 |
+
| `BTCAnalyst` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md, quant/macro_context_{date}.md, quant/unusual_events_{date}.md, quant/event_route_{date}.md, quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md]` | `[quant/btc_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[quant/btc_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[]` |
|
| 178 |
+
| `SPYAnalyst` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md, quant/macro_context_{date}.md, quant/unusual_events_{date}.md, quant/event_route_{date}.md, quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md]` | `[quant/spy_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[quant/spy_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[]` |
|
| 179 |
+
| `QQQAnalyst` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md, quant/macro_context_{date}.md, quant/unusual_events_{date}.md, quant/event_route_{date}.md, quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md]` | `[quant/qqq_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[quant/qqq_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[]` |
|
| 180 |
+
| `GLDAnalyst` | `[quant/market_events_{date}.md, quant/options_flow_{date}.md, quant/macro_context_{date}.md, quant/unusual_events_{date}.md, quant/event_route_{date}.md, quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md]` | `[quant/gld_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[quant/gld_analysis_{date}.md, quant/asset_analysis_{date}.md]` | `[]` |
|
| 181 |
+
| `QuantResearcher` | `[quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md, quant/asset_analysis_{date}.md, quant/macro_context_{date}.md, quant/unusual_events_{date}.md]` | `[quant/signals_{date}.md]` | `[quant/signals_{date}.md]` | `[]` |
|
| 182 |
+
| `Backtester` | `[quant/signals_{date}.md, quant/raw/{ticker}_{date}_summary.md]` | `[quant/backtest_results_{date}.md]` | `[quant/backtest_results_{date}.md]` | `[quant/raw/{ticker}_{date}_summary.md]` |
|
| 183 |
+
| `RiskAnalyst` | `[quant/backtest_results_{date}.md, quant/trend_analysis_{date}.md, quant/asset_analysis_{date}.md, quant/macro_context_{date}.md]` | `[quant/risk_assessment_{date}.md]` | `[quant/risk_assessment_{date}.md]` | `[]` |
|
| 184 |
+
| `QuantStrategist` | `[quant/risk_assessment_{date}.md, quant/backtest_results_{date}.md, quant/sentiment_analysis_{date}.md, quant/signals_{date}.md, quant/asset_analysis_{date}.md, quant/macro_context_{date}.md, quant/strategist_performance.json]` | `[quant/trade_proposals_{date}.md, quant/strategist_performance.json]` | `[quant/trade_proposals_{date}.md, quant/strategist_performance.json]` | `[]` |
|
| 185 |
+
| `Publisher` | `[quant/trade_proposals_{date}.md, quant/trend_analysis_{date}.md, quant/sentiment_analysis_{date}.md, quant/risk_assessment_{date}.md, quant/asset_analysis_{date}.md, quant/macro_context_{date}.md]` | `[quant/post_{date}.md]` | `[quant/post_{date}.md]` | `[]` |
|
| 186 |
+
| `VizAgent` | `[quant/raw/{ticker}_{date}.json, quant/backtest_results_{date}.md, quant/risk_assessment_{date}.md]` | `[quant/viz/candlestick_{ticker}_{date}.png, quant/viz/rsi_macd_{ticker}_{date}.png, quant/viz/backtest_{date}.png, quant/viz/risk_heatmap_{date}.png, quant/viz/portfolio_pnl_{date}.png]` | `[quant/viz/candlestick_{ticker}_{date}.png, quant/viz/rsi_macd_{ticker}_{date}.png, quant/viz/backtest_{date}.png, quant/viz/risk_heatmap_{date}.png, quant/viz/portfolio_pnl_{date}.png]` | `[quant/raw/{ticker}_{date}.json]` |
|
| 187 |
+
| `PortfolioMonitor` | `[]` | `[quant/portfolio_log_{date}.md]` | `[quant/portfolio_log_{date}.md]` | `[]` |
|
| 188 |
+
|
| 189 |
+
#### Baseline File Permissions
|
| 190 |
+
|
| 191 |
+
These lists extend the corresponding agent's `read` and `list` permissions
|
| 192 |
+
above for session-delta runs. `{baseline}` is the runtime-selected existing
|
| 193 |
+
baseline directory inside this session. `{date}` is the saved artifact date.
|
| 194 |
+
No agent receives `create` or `edit` permission for baseline files. The
|
| 195 |
+
runtime preserves those snapshots. Trade-state context uses the selected
|
| 196 |
+
proposal or no-trade/watch file when available.
|
| 197 |
+
|
| 198 |
+
| Agent | Additional `read` | Additional `list` |
|
| 199 |
+
|---|---|---|
|
| 200 |
+
| `DataFetcher` | `[quant/yolo_baselines/{baseline}/quant/raw/{ticker}_{date}_summary.md]` | `[quant/yolo_baselines/{baseline}/quant/raw/{ticker}_{date}_summary.md]` |
|
| 201 |
+
| `OptionsFlowWatcher` | `[quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 202 |
+
| `TrendAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 203 |
+
| `SentimentAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 204 |
+
| `MarketEventWatcher` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 205 |
+
| `MacroContextWatcher` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 206 |
+
| `UnusualEventDetector` | `[quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 207 |
+
| `EventRouter` | `[quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 208 |
+
| `BTCAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 209 |
+
| `SPYAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 210 |
+
| `QQQAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 211 |
+
| `GLDAnalyst` | `[quant/yolo_baselines/{baseline}/quant/trend_analysis_{date}.md, quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 212 |
+
| `QuantResearcher` | `[quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/backtest_results_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 213 |
+
| `Backtester` | `[quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/backtest_results_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md]` | `[]` |
|
| 214 |
+
| `RiskAnalyst` | `[quant/yolo_baselines/{baseline}/quant/backtest_results_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 215 |
+
| `QuantStrategist` | `[quant/yolo_baselines/{baseline}/quant/signals_{date}.md, quant/yolo_baselines/{baseline}/quant/backtest_results_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md, quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md]` | `[]` |
|
| 216 |
+
| `Publisher` | `[quant/yolo_baselines/{baseline}/quant/trade_proposals_{date}.md, quant/yolo_baselines/{baseline}/quant/no_trade_watch_{date}.md, quant/yolo_baselines/{baseline}/quant/post_{date}.md, quant/yolo_baselines/{baseline}/quant/risk_assessment_{date}.md]` | `[]` |
|
| 217 |
+
| `VizAgent` | `[]` | `[]` |
|
| 218 |
+
| `PortfolioMonitor` | `[]` | `[]` |
|
| 219 |
+
|
| 220 |
+
#### Permission Boundaries
|
| 221 |
+
|
| 222 |
+
- For dated files, `create` and `edit` refer to working outputs for the current run's date.
|
| 223 |
+
Historical working files and baseline snapshots are read-only context.
|
| 224 |
+
- Read permission does not add a Routing Table dependency or imply that an
|
| 225 |
+
optional file is available. For example, current trend context is an optional
|
| 226 |
+
input to `SentimentAnalyst` and `MarketEventWatcher` under the agreed routing.
|
| 227 |
+
- `DataFetcher` no longer needs to read trend output solely to choose the next
|
| 228 |
+
agent; that decision belongs to the Routing Table.
|
| 229 |
+
- Helpers acting for an agent, including provider writers and committee
|
| 230 |
+
tracking, use that agent's runtime-controlled permissions.
|
| 231 |
+
- Writes to `quant/strategist_performance.json` are performed by the trusted
|
| 232 |
+
performance tracker for `QuantStrategist`. LLM-emitted weights do not replace
|
| 233 |
+
runtime-owned weights.
|
| 234 |
+
- The shared `quant/asset_analysis_{date}.md` entries record the existing
|
| 235 |
+
append behavior of all four asset analysts. Filename ACL alone cannot limit
|
| 236 |
+
an edit to one analyst's section. Shared-file ownership and parallel writes
|
| 237 |
+
must be settled with typed schemas and artifact storage before this ACL is
|
| 238 |
+
implemented.
|
| 239 |
+
- Shared reports may contain multiple tickers. Their filename permission alone
|
| 240 |
+
does not enforce the effective ticker scope; the typed input contract must
|
| 241 |
+
define the scope-filtered content exposed to each agent.
|
| 242 |
+
- Session metadata, selected portfolio state, prior backtest state, and recent
|
| 243 |
+
conversation context currently arrive through runtime inputs. These tables
|
| 244 |
+
do not grant agents unrestricted access to `research_state.json` or
|
| 245 |
+
`thread.jsonl`. Their allowed input fields remain part of the typed-schema
|
| 246 |
+
discussion. Runtime state patches, traces, snapshots, and evidence-ledger
|
| 247 |
+
writes remain runtime-owned.
|
| 248 |
+
- Listing and RAG retrieval must observe the same session and filename
|
| 249 |
+
permissions. Final storage changes must update these tables accordingly.
|
| 250 |
+
|
| 251 |
+
|
| 252 |
+
#### File Content Reference
|
| 253 |
+
|
| 254 |
+
This table explains the files referenced by the ACL above using their current
|
| 255 |
+
contents and explicitly identified V8 changes. Exact typed schemas remain a
|
| 256 |
+
separate design step. A listed filename describes an artifact contract, not a
|
| 257 |
+
guarantee that a provider or agent successfully produced it in every run.
|
| 258 |
+
|
| 259 |
+
| Filename relative to session folder | Format | Content | Purpose |
|
| 260 |
+
|---|---|---|---|
|
| 261 |
+
| `quant/raw/{ticker}_{date}.json` | JSON | Ticker and fetch timestamp; available daily OHLCV history; computed SMA-20, EMA-20, RSI-14, and MACD series; news records; earnings records where applicable; provider errors, partial-fetch indicators, and OHLCV quality metadata when present. | Structured fetched market dataset used to generate summaries and support downstream analysis. Despite the `raw` directory name, this file contains normalized data and computed indicators rather than an untouched API response. |
|
| 262 |
+
| `quant/raw/{ticker}_{date}_summary.md` | Markdown | Ticker and fetch timestamp; data-quality notes and excluded rows when present; latest 10 available OHLCV rows; latest technical-indicator values; up to 15 news headlines with available sentiment fields; up to four quarterly earnings records where applicable; a prior-baseline summary excerpt during session-delta runs when available. | Readable market-data context for downstream agents, including context for comparing the current run with the previous baseline. |
|
| 263 |
+
| `quant/trend_analysis_{date}.md` | Markdown | Technical interpretation of price structure, indicators, trend direction, and uncertainty; delta observations where available. | Technical context produced by TrendAnalyst. |
|
| 264 |
+
| `quant/sentiment_analysis_{date}.md` | Markdown | News and earnings sentiment assessment, cited themes, sentiment/technical alignment, and insufficient-data notes. | Sentiment context produced by SentimentAnalyst. |
|
| 265 |
+
| `quant/market_events_{date}.md` | Markdown | In-scope raw summary excerpts, optional existing trend context, check timestamp, and prior-session context. | Market context assembled by MarketEventWatcher; the current implementation does not fetch a separate event feed. |
|
| 266 |
+
| `quant/options_flow_{date}.md` | Markdown | Options provider status and timestamp, sampled contracts and expirations, call/put volume, open interest, IV summaries, preliminary FLOW_CANDIDATE observations, and explicit data gaps. | Options evidence from OptionsFlowWatcher, distinguishing options-chain context from unsupported unusual-flow claims. |
|
| 267 |
+
| `quant/macro_context_{date}.md` | Markdown | LLM macro main-line assessment, candidate drivers, relevance weights, confidence, uncertainty, and affected markets, or an explicit analysis-unavailable result. | Macro interpretation produced by MacroContextWatcher. |
|
| 268 |
+
| `quant/unusual_events_{date}.md` | Markdown | LLM event candidates with event type, affected markets, significance, rationale, and missing evidence; a no-meaningful-event assessment where appropriate. | Event detection output from UnusualEventDetector. |
|
| 269 |
+
| `quant/event_route_{date}.md` | Markdown | Currently selected symbols, selected analyst names, rationale, and delta context. V8 uses this output for event-to-ticker evidence assignments under the fixed Routing Table. | Evidence assignment output from EventRouter; it cannot activate agents in V8. |
|
| 270 |
+
| `quant/btc_analysis_{date}.md` | Markdown | BTC market interpretation, weighted signals, confidence, setup hypothesis, confirmation/invalidation conditions, and recommended outcome or explicit analysis failure. | Individual output of BTCAnalyst. |
|
| 271 |
+
| `quant/spy_analysis_{date}.md` | Markdown | SPY market interpretation, weighted signals, confidence, setup hypothesis, confirmation/invalidation conditions, and recommended outcome or explicit analysis failure. | Individual output of SPYAnalyst. |
|
| 272 |
+
| `quant/qqq_analysis_{date}.md` | Markdown | QQQ market interpretation, weighted signals, confidence, setup hypothesis, confirmation/invalidation conditions, and recommended outcome or explicit analysis failure. | Individual output of QQQAnalyst. |
|
| 273 |
+
| `quant/gld_analysis_{date}.md` | Markdown | GLD market interpretation, weighted signals, confidence, setup hypothesis, confirmation/invalidation conditions, and recommended outcome or explicit analysis failure. | Individual output of GLDAnalyst. |
|
| 274 |
+
| `quant/asset_analysis_{date}.md` | Markdown | Combined ticker-labelled sections appended by the specialized analysts. | Shared specialized-analysis input for QuantResearcher, RiskAnalyst, QuantStrategist, and Publisher; write ownership remains a storage decision. |
|
| 275 |
+
| `quant/signals_{date}.md` | Markdown | Named signal hypotheses, entry/exit conditions, confidence, falsification criteria, and run/delta updates. | QuantResearcher output for Backtester and strategy review. |
|
| 276 |
+
| `quant/backtest_results_{date}.md` | Markdown | LLM-produced backtest assessment from supplied signals and market summaries, approximate metrics, assumptions, and limitations. | Backtester output for risk and strategy review; its presence alone does not establish an executed historical experiment. |
|
| 277 |
+
| `quant/risk_assessment_{date}.md` | Markdown | Risk ratings, concentration/correlation and drawdown assessment, scenarios, and risk verdicts using available analysis and portfolio context. | RiskAnalyst output consumed before strategy decisions. |
|
| 278 |
+
| `quant/trade_proposals_{date}.md` | Markdown | Strategy-review narrative, committee summary, prior proposal lifecycle updates, and paper-trading conclusions. Structured proposals also travel in the runtime agent output/state patch. | QuantStrategist's readable strategy result and prior trade-state context. |
|
| 279 |
+
| `quant/no_trade_watch_{date}.md` | Markdown | Explicit no-trade or watch state when such a prior artifact exists. | Supported alternative baseline trade-state input; the ACL does not introduce a new writer for this file. |
|
| 280 |
+
| `quant/strategist_performance.json` | JSON | Trusted member weights, vote history, weight-change history, and latest weighted aggregation. | Session-owned committee record maintained through the existing performance tracker. |
|
| 281 |
+
| `quant/post_{date}.md` | Markdown | Finley's readable final market output, baseline/delta interpretation, and paper-trading disclaimer. | Publisher's user-facing result. |
|
| 282 |
+
| `quant/portfolio_log_{date}.md` | Markdown | Timestamped open-position checks, available current prices, threshold breaches, missing-price notes, and alerts. | PortfolioMonitor's daily monitoring record. |
|
| 283 |
+
| `quant/viz/candlestick_{ticker}_{date}.png` | PNG chart | Price candles, volume, and a moving-average overlay when usable source data exists. | Market-price visualization produced by VizAgent. |
|
| 284 |
+
| `quant/viz/rsi_macd_{ticker}_{date}.png` | PNG chart | RSI and MACD panels based on available indicator data. | Technical-indicator visualization produced by VizAgent. |
|
| 285 |
+
| `quant/viz/backtest_{date}.png` | PNG chart | Signal win-rate chart from parseable backtest metrics. | Backtest visualization produced by VizAgent. |
|
| 286 |
+
| `quant/viz/risk_heatmap_{date}.png` | PNG chart | Risk verdicts rendered as a labelled risk-rating chart. | Risk visualization produced by VizAgent. |
|
| 287 |
+
| `quant/viz/portfolio_pnl_{date}.png` | PNG chart | Currently position-size bars and entry-price markers from portfolio state. | Portfolio visualization produced by VizAgent; despite its filename, the current code does not calculate realized/unrealized P&L here. |
|
| 288 |
+
| `quant/yolo_baselines/{baseline}/quant/...` | Same as saved file | Saved copies of the exact quant filenames enumerated in the Baseline File Permissions table, retaining each original file's content and relative path. | Read-only prior-run context. The ellipsis here describes that directory layout and is not an ACL wildcard. |
|
| 289 |
+
| `research_state.json` | JSON | Runtime session state, including quant watchlist, portfolio, proposals, prior backtest state, and YOLO metadata. | Runtime-owned state source for selected typed agent inputs; not a blanket file permission. |
|
| 290 |
+
| `thread.jsonl` | JSON Lines | Recorded conversation and agent messages for the session. | Runtime-owned conversation history; agent context must be filtered according to its allowed inputs. |
|
| 291 |
+
|
| 292 |
+
Missing sections do not imply zero values or neutral evidence. Availability and
|
| 293 |
+
data-quality metadata must be respected when consuming these files.
|
| 294 |
+
|
| 295 |
+
### Typed Artifact Schemas
|
| 296 |
+
|
| 297 |
+
- Every migrated quant agent has typed input and output artifact contracts.
|
| 298 |
+
- Agent activation is based on declared dependencies and valid required input
|
| 299 |
+
artifacts.
|
| 300 |
+
- Required inputs, optional inputs, produced outputs, provenance, freshness,
|
| 301 |
+
and explicit data-gap states are represented structurally where applicable.
|
| 302 |
+
- A downstream agent receives only schema-valid dependency outputs exposed by
|
| 303 |
+
its ACL.
|
| 304 |
+
- Human-readable markdown may remain as a rendered artifact, but markdown
|
| 305 |
+
conventions alone are not the V8.0 machine contract.
|
| 306 |
+
- Artifact storage details will be finalized after the dependency, ACL, and
|
| 307 |
+
schema contracts are agreed.
|
| 308 |
+
|
| 309 |
+
### Runtime And Product Boundaries
|
| 310 |
+
|
| 311 |
+
- All quant agents migrate to the V8.0 contract together.
|
| 312 |
+
- No human is an agent entry inside the V8.0 Routing Table.
|
| 313 |
+
- Current UI behavior is retained and extended to show agent progress, the
|
| 314 |
+
Routing Table, and ACL information.
|
| 315 |
+
- Existing strategy logic is preserved unless this PRD explicitly changes it.
|
| 316 |
+
|
| 317 |
+
### Session Collaboration
|
| 318 |
+
|
| 319 |
+
- The session folder is the shared persistent context space for one analysis
|
| 320 |
+
session.
|
| 321 |
+
- Agents collaborate by producing and consuming governed session artifacts.
|
| 322 |
+
- Baseline and session-delta YOLO behavior continue to use the same agents and
|
| 323 |
+
the same Routing Table.
|
| 324 |
+
- A later run may reuse prior session artifacts only when its dependency
|
| 325 |
+
contract, ACL, and freshness requirements allow it.
|
| 326 |
+
- The Routing Table controls readiness; the ACL controls visibility and
|
| 327 |
+
mutation; typed schemas control the meaning of exchanged artifacts.
|
| 328 |
+
|
| 329 |
+
### UI Visibility
|
| 330 |
+
|
| 331 |
+
The V8.0 UI retains the current YOLO run experience and adds:
|
| 332 |
+
|
| 333 |
+
1. Current agent activation and completion state from the Routing Table.
|
| 334 |
+
2. The Routing Table using exact registered agent names.
|
| 335 |
+
3. ACL information showing which session artifacts each agent may read and
|
| 336 |
+
produce.
|
| 337 |
+
|
| 338 |
+
The UI must not present services, schemas, ACL entries, or dependency-engine
|
| 339 |
+
internals as additional agents.
|
| 340 |
+
|
| 341 |
+
---
|
| 342 |
+
|
| 343 |
+
## Agreed Routing Table
|
| 344 |
+
|
| 345 |
+
This table contains every registered quant agent. Scope is one of four enum
|
| 346 |
+
values. A dependency applies only where the two agents' effective scopes
|
| 347 |
+
overlap. `Required agent dependencies` is always a list; `[]` means the agent
|
| 348 |
+
has no required upstream agent.
|
| 349 |
+
|
| 350 |
+
```text
|
| 351 |
+
QUANT = any valid quant ticker
|
| 352 |
+
GENERAL = any valid quant ticker except BTC, SPY, QQQ, GLD
|
| 353 |
+
SPECIALIZED = BTC, SPY, QQQ, GLD
|
| 354 |
+
OPTIONS = SPY, QQQ, GLD
|
| 355 |
+
```
|
| 356 |
+
|
| 357 |
+
| Agent | Required agent dependencies | Scope |
|
| 358 |
+
|---|---|---|
|
| 359 |
+
| `DataFetcher` | `[]` | `QUANT` |
|
| 360 |
+
| `OptionsFlowWatcher` | `[]` | `OPTIONS` |
|
| 361 |
+
| `TrendAnalyst` | `[DataFetcher]` | `QUANT` |
|
| 362 |
+
| `SentimentAnalyst` | `[DataFetcher]` | `GENERAL` |
|
| 363 |
+
| `MarketEventWatcher` | `[DataFetcher]` | `SPECIALIZED` |
|
| 364 |
+
| `MacroContextWatcher` | `[MarketEventWatcher, OptionsFlowWatcher]` | `SPECIALIZED` |
|
| 365 |
+
| `UnusualEventDetector` | `[MarketEventWatcher, MacroContextWatcher, OptionsFlowWatcher]` | `SPECIALIZED` |
|
| 366 |
+
| `EventRouter` | `[UnusualEventDetector]` | `SPECIALIZED` |
|
| 367 |
+
| `BTCAnalyst` | `[TrendAnalyst, EventRouter]` | `SPECIALIZED` |
|
| 368 |
+
| `SPYAnalyst` | `[TrendAnalyst, EventRouter]` | `SPECIALIZED` |
|
| 369 |
+
| `QQQAnalyst` | `[TrendAnalyst, EventRouter]` | `SPECIALIZED` |
|
| 370 |
+
| `GLDAnalyst` | `[TrendAnalyst, EventRouter]` | `SPECIALIZED` |
|
| 371 |
+
| `QuantResearcher` | `[TrendAnalyst, SentimentAnalyst, BTCAnalyst, SPYAnalyst, QQQAnalyst, GLDAnalyst]` | `QUANT` |
|
| 372 |
+
| `Backtester` | `[QuantResearcher]` | `QUANT` |
|
| 373 |
+
| `RiskAnalyst` | `[Backtester]` | `QUANT` |
|
| 374 |
+
| `QuantStrategist` | `[RiskAnalyst]` | `QUANT` |
|
| 375 |
+
| `Publisher` | `[QuantStrategist]` | `QUANT` |
|
| 376 |
+
| `VizAgent` | `[]` | `QUANT` |
|
| 377 |
+
| `PortfolioMonitor` | `[]` | `QUANT` |
|
| 378 |
+
|
| 379 |
+
`BTCAnalyst`, `SPYAnalyst`, `QQQAnalyst`, and `GLDAnalyst` use the
|
| 380 |
+
`SPECIALIZED` scope enum. Their existing fixed `symbol` property further
|
| 381 |
+
restricts each effective scope to its own ticker; the ticker does not create a
|
| 382 |
+
new scope enum.
|
| 383 |
+
|
| 384 |
+
When their effective scopes are not empty, `TrendAnalyst`, `SentimentAnalyst`,
|
| 385 |
+
and `MarketEventWatcher` become eligible in parallel after `DataFetcher`
|
| 386 |
+
completes successfully for the corresponding ticker scope.
|
| 387 |
+
|
| 388 |
+
### Root Agents
|
| 389 |
+
|
| 390 |
+
`DataFetcher`, `OptionsFlowWatcher`, and `VizAgent` have no required upstream
|
| 391 |
+
agents. Their ticker, date, session, provider, run-mode, or visualization inputs
|
| 392 |
+
are input-contract fields and must not be represented as agent entries.
|
| 393 |
+
|
| 394 |
+
`PortfolioMonitor` is also a root agent, but only in a separately initialized
|
| 395 |
+
portfolio-monitor run. It is not added to the active YOLO routing plan.
|
| 396 |
+
|
| 397 |
+
### Conditional Joins
|
| 398 |
+
|
| 399 |
+
The Routing Table uses effective-scope overlap to resolve joins:
|
| 400 |
+
|
| 401 |
+
- `MacroContextWatcher` waits for `OptionsFlowWatcher` only for `SPY`, `QQQ`,
|
| 402 |
+
and `GLD`; BTC has no options dependency.
|
| 403 |
+
- `QuantResearcher` waits for `SentimentAnalyst` only for general-market scope.
|
| 404 |
+
- `QuantResearcher` waits for each specialized analyst whose single-symbol
|
| 405 |
+
effective scope is active.
|
| 406 |
+
- Mixed-ticker runs may therefore activate both the general analysis branch and
|
| 407 |
+
one or more specialized analysis branches before joining at
|
| 408 |
+
`QuantResearcher`.
|
| 409 |
+
|
| 410 |
+
### Run Routing Initialization
|
| 411 |
+
|
| 412 |
+
At the beginning of each YOLO run, the runtime:
|
| 413 |
+
|
| 414 |
+
1. Normalizes the complete ticker scope.
|
| 415 |
+
2. Selects the execution context: YOLO, visualization request, or
|
| 416 |
+
portfolio-monitor run.
|
| 417 |
+
3. Intersects the requested tickers with each applicable agent's supported
|
| 418 |
+
scope.
|
| 419 |
+
4. Stores the resulting effective scope for each agent.
|
| 420 |
+
5. Removes agents with an empty effective scope, agents outside the selected
|
| 421 |
+
execution context, and dependencies with no
|
| 422 |
+
overlapping effective scope.
|
| 423 |
+
6. Creates the fixed active routing plan for that run.
|
| 424 |
+
7. Activates each included agent when its applicable required dependencies have
|
| 425 |
+
produced valid typed outputs.
|
| 426 |
+
|
| 427 |
+
Inactive agents are absent from that run's active routing plan. A data-gap
|
| 428 |
+
output does not change the plan; it is a typed result consumed by already
|
| 429 |
+
activated downstream agents. Mixed ticker scopes may activate both general and
|
| 430 |
+
specialized agents in the same run, with each agent receiving only its
|
| 431 |
+
applicable ticker scope.
|
| 432 |
+
|
| 433 |
+
---
|
| 434 |
+
|
| 435 |
+
## Remaining Contract Decisions
|
| 436 |
+
|
| 437 |
+
The Routing Table and draft ACL rows now cover all registered quant agents.
|
| 438 |
+
The remaining decisions are typed input/output schemas, shared-file ownership
|
| 439 |
+
and ticker filtering, final artifact storage, and UI presentation. The tables
|
| 440 |
+
are design specifications; runtime migration and enforcement are not yet
|
| 441 |
+
implemented.
|
| 442 |
+
|
| 443 |
+
---
|
| 444 |
+
|
| 445 |
+
## V8.0 Definition Checklist
|
| 446 |
+
|
| 447 |
+
- [x] Limit V8.0 product scope to quant agents.
|
| 448 |
+
- [x] Define the Routing Table as the workflow authority.
|
| 449 |
+
- [x] Require agent-only Routing Table entries with exact registered names.
|
| 450 |
+
- [x] Define the first data-and-context dependency slice.
|
| 451 |
+
- [x] Define the `UnusualEventDetector` dependency slice.
|
| 452 |
+
- [x] Add all registered quant agents to the Routing Table.
|
| 453 |
+
- [x] Define the complete Routing Table at the agent-dependency and scope level.
|
| 454 |
+
- [x] Draft filename-based ACL rows and file-content references for all quant agents.
|
| 455 |
+
- [x] Record baseline snapshot read permissions for session-delta context.
|
| 456 |
+
- [ ] Finalize shared-file ownership and scope-filtered access with typed schemas.
|
| 457 |
+
- [ ] Finalize the ACL contract for each quant agent after those decisions.
|
| 458 |
+
- [ ] Define typed input and output artifact schemas for each quant agent.
|
| 459 |
+
- [ ] Finalize artifact storage from the approved contracts.
|
| 460 |
+
- [ ] Define UI presentation for dependency and ACL state.
|
| 461 |
+
- [ ] Write the V8.0 implementation and verification plan.
|