docs: adopt anti-rot debt ledger and ADR directory

docs/DEBT.md

@@ -0,0 +1,26 @@
+---
+type: canonical
+source: none
+sla: on-change
+last_updated: "2026-06-06"
+audience: [ai-agents, contributors]
+---
+
+# Technical Debt Ledger
+
+The accumulated cost of deliberate shortcuts. The goal is not zero debt, it is
+zero untracked debt. Anything recorded here was a conscious choice with a known
+fix. Add entries with `/debt-log`. Remove an entry when the debt is paid (note it
+in the PR).
+
+<!-- New entries are appended below, newest first. Format:
+
+### <short title>
+- **Date:** YYYY-MM-DD
+- **Where:** <file/module/path>
+- **What:** the shortcut and why the proper fix was not done
+- **Risk if left:** what degrades over time
+- **Suggested fix:** the path to doing it right
+- **Owner:** <from CODEOWNERS or repo-framework ownership>
+
+-->

docs/adr/0000-template.md

@@ -0,0 +1,33 @@
+---
+type: canonical
+source: none
+sla: on-change
+last_updated: "2026-06-06"
+audience: [ai-agents, contributors]
+---
+
+# ADR-0000: <Title of the decision>
+
+> Template. Copy to `docs/adr/<NNNN>-<kebab-title>.md` (use `/new-adr`).
+> One decision per record. Records are append-only history: supersede, do not rewrite.
+
+- **Status:** Proposed | Accepted | Superseded by ADR-XXXX
+- **Date:** YYYY-MM-DD
+- **Deciders:** <names / from repo-framework ownership>
+
+## Context
+
+What is the situation? What forces, constraints, and requirements make a decision
+necessary? Keep it factual.
+
+## Decision
+
+What we are doing, stated in active voice: "We will ...". Be specific about scope
+and boundaries (module layout, dependency direction, technology choice).
+
+## Consequences
+
+- **Positive:** what gets easier or safer.
+- **Negative / tradeoffs:** what gets harder, and any debt this introduces
+  (cross-reference `docs/DEBT.md` if applicable).
+- **Follow-ups:** anything that must happen as a result.

docs/deployment.md

@@ -6,13 +6,58 @@ last-reviewed: 2026-03-31
 
 # Deployment and Release · fallax
 
-> TODO: Document deployment process, release strategy, and rollback procedures.
+Fallax is a research and benchmarking tool, not a deployed service. There is
+no server, no container, and no production environment to operate. All
+evaluation runs execute locally against the provider APIs you configure.
 
-## Deployment Process
+## Running Evaluations Locally
 
-## Release Strategy
+Follow the Quick Start in the root [README](../README.md):
 
-## Rollback Procedures
+```bash
+# Install deps (core + all extras)
+uv sync --all-extras
 
-## Environment Configuration
+# Run a model evaluation
+uv run python -m fallax run \
+  --models claude-sonnet-4-6 \
+  --judge claude-haiku-4-5-20251001 \
+  --output results.jsonl
 
+# Capture a baseline against the v1 benchmark
+uv run python -m fallax baseline capture \
+  --version v1 \
+  --model claude-sonnet-4-6 \
+  --judge claude-haiku-4-5-20251001
+
+# Compare against a captured baseline
+uv run python -m fallax baseline compare \
+  --version v1 \
+  --model claude-sonnet-4-6 \
+  --judge claude-haiku-4-5-20251001
+```
+
+Required environment variables:
+
+| Provider | Variable |
+|---|---|
+| Anthropic (default) | `ANTHROPIC_API_KEY` |
+| OpenAI / OpenRouter | `OPENAI_API_KEY` or `OPENROUTER_API_KEY` |
+| Google Gemini | `GOOGLE_API_KEY` |
+
+## Optional Dashboard
+
+The FastAPI results explorer is available locally only:
+
+```bash
+uv sync --extra dashboard
+uv run python -m fallax dashboard
+```
+
+It does not require any deployment infrastructure.
+
+## Versioning
+
+Fallax follows [Semantic Versioning](https://semver.org/). The current package
+version is in `pyproject.toml`. Releases are tagged in git; there is no
+release pipeline or package registry publication at this time.

docs/troubleshooting.md

@@ -6,8 +6,6 @@ last-reviewed: 2026-03-31
 
 # Troubleshooting · fallax
 
-> TODO: Document known failure modes, diagnostic steps, and common fixes.
-
 ## Common Issues
 
 ### `git status` shows many files modified right after a clone or pull