Project documentation

The repository root is reserved for files with a real discovery or runtime contract: GitHub/Python metadata, OpenClaw bootstrap context, and the live workspace ledgers. Static website source lives in site/; Pages combines it with public runtime outputs through ops/pages/stage_site.py. General documentation belongs here.

Architecture

• harness.md — package/profile/runtime boundaries, CLI lifecycle, context injection contract, and generation-pinned artifacts.
• decision-architecture.md — who owns observations, computation, rule outputs, the hypothesis and the accounting; the observations and compute tools, immutable method proposals and prospective three-arm trials.
• session-reviews.md — same-session plan progress and read-only risk follow-ups on market cards.
• intraday-agent.md — the intraday agent contract: context layering, card blocks, workflow ownership, objective and which rules are gates.
• data-plane.md — why the live JSON snapshot is separate from Pages, what GitHub officially supports, and the replacement bar.
• openclaw-adapter.md — OpenClaw as an external runtime and the parity contract the adapter must keep.
• runtime-protocol.md — how any external agent runtime invokes clawock.
• task-queue.md — the agent-dispatch queue behind the dsh task chip: who owns which layer, the versioned ops entry, queue order, and the frozen task/token summary shown by the chip and /dispatch-stats.
• strategy-plugins.md — strategies, bars and the optimiser as plugins composed by config rows; what a parameter search may conclude.

Research notes

• dsh-superagent-adoption.md — what the external dsh-superagent prototype’s Mandate, stage and notification ideas mean for dispatch, what was adopted where, and what explicitly does not transfer.

Product surfaces

• how-the-desk-works.md — the detail behind the README: information layers, what each run reads, decision gates, grading rules, what was tested and failed, the twelve code-enforced rules (中文).
• decision-map.md — the Decision Map board in Reflect: how to read coverage and snapshot age, the payload, and where it runs.
• decision-mind-ledger.md — the decision-mind ledger schema written by clawock record.
• influencer-radar.md — the eight watched sources, how hits are linked to holdings, and a worked hit/miss example (moved out of the READMEs).
• glossary.md — source of truth for cross-document terminology.
• data-coverage.md — data surfaces, independent sources, failure behavior and remaining gaps.
• data-health.md — the Data Health card’s structure, visual rules and the tests that enforce them.
• design-system.md — the dashboard’s colour layers (semantic vs. everything else), motion scale, component recipes and the list of what a new component may not do, each with the gate that checks it.
• dispatch-stats.md — the chip’s stats tile and chart layer, and the /dispatch-stats command: the windows, what the frozen numbers mean and do not mean, and the effect captures.
• task-chip-patrol.md — the patrol block as one entry with a 4px outcome strip, and a dashboard page (KPI bento, outcome strip, rounds, coverage tables); the motion and the captures.
• task-chip-stats.md — the stats card at the bottom of the task chip and its 7-day stacked provider token chart: brand hues muted into one category band, the state-colour collision rules, zero/unknown days, and no total line or visible unknown caption.

Operations

• release.md — publishing to PyPI/npm, and running the latest code on the host without a release.
• openclaw-patch-kit.md — versioned dist adaptations, isolated rehearsal, transaction receipts and guarded rollback.
• dreaming-review.md — private candidates, explicit review receipts, permanent promotion and rollback.
• models.md — config/models.toml: every model, its parameters and which scenario uses it; OpenClaw, cron, dsh and the MiniMax client follow it.
• cron-schedules.md — generated human view of the tracked cron contract.
• research-cadence.md — which research question runs daily, which runs on an event, and why.

• skills-store-policy.md — registry discovery and installation policy.

• agent-discovery.md — public machine documentation, documented engine discovery paths and a bounded citation measurement protocol.

Reference

• commands.md — the generated command inventory plus the hand-written harness detail (scripts.md is a moved-page stub pointing here).
• tool-operations.md — per-task tool detail; routing lives in the root TOOLS.md.
• product-profile-operations.md — what belongs to the product, profiles, operations and runtime state.

Legal

• third-party-data.md — data-provider terms, attribution, and redistribution boundaries.

Everything under docs/ is published with the Pages build (ops/pages/stage_site.py). Keep this index current when adding or removing docs.

A document here describes how the project works now. The record of one task — measurements, before/after captures, command output, proposed patches, what was and was not verified — belongs in that task’s pull request, where it stays attached to the change it explains. Write the conclusion into the document that owns the subject; do not add a dated folder.

Why OpenClaw files stay at the root

AGENTS.md, TOOLS.md, SOUL.md, IDENTITY.md, USER.md, MEMORY.md, HEARTBEAT.md, BOOTSTRAP.md, skills/ and memory/ participate in the live OpenClaw workspace contract. They remain at their required runtime paths until an adjacent context-parity canary proves a supported alternative. Website files have no such constraint and therefore live under site/.

• 持仓外候选发现：免费全市场筛选、研究交接与参考实现。

Intraday research

• Open-source strategy research: sources, failure conditions, verified fees and market selection.

• Intraday research contract: mounted replay service, LLM boundaries and presentation design.

• Intraday research / 盘中研究验证与卡片： 三种 research 插件、模型回执、净 walk-forward、数据许可、历史研究展示及独立人工条件推荐 / independent manual conditional recommendations。