> ## Documentation Index > Fetch the complete documentation index at: https://defendable.science/llms.txt > Use this file to discover all available pages before exploring further. # defendable-science — Decision Log (ADRs) > MADR-style Architecture Decision Records for the defendable-science plugin. Each records the context, the options considered, the rejected alternatives and why… MADR-style Architecture Decision Records for the `defendable-science` plugin. Each records the context, the options considered, the rejected alternatives **and why**, and links to the spec section + grounding digest. Captured 2026-07-17 during the design session; the reasoning here is not reconstructable from the specs alone. Migrates to the plugin's `decisions/` alongside `resources/references/`. > Purpose beyond the record: this log is the raw material for a planned blog > post / paper explaining the skills and the reasoning behind them — the intent > is eventually to write that *using* `defendable-science` (`paper-synthesis` + `defend`). | ADR | Decision | Status | | ---------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | -------- | | [0001](/decisions/0001-separate-plugin-repo) | Package the workflow as a standalone plugin, not in-mononet | accepted | | [0002](/decisions/0002-scientific-scope-only) | Scientific scope only; engineering delegated via a contract | accepted | | [0003](/decisions/0003-agency-principle) | Agency principle — assistants, not researchers | accepted | | [0004](/decisions/0004-understanding-principle-and-defend) | Understanding principle + the `defend` skill | accepted | | [0005](/decisions/0005-three-level-mirror) | Three-level mirror; thesis as a partial mirror | accepted | | [0006](/decisions/0006-two-skills-per-level) | Two pipeline skills per level (generate/resolve) | accepted | | [0007](/decisions/0007-literature-one-skill-two-modes) | `literature` = one skill, scout/position modes | accepted | | [0008](/decisions/0008-literature-bib-plus-triage) | Literature registry = standard bib + triage sidecar | accepted | | [0009](/decisions/0009-dataset-thin-manifest) | `dataset` one skill + thin `datasets.yml` (reject DVC/DataLad/lakeFS) | accepted | | [0010](/decisions/0010-storage-tiers) | Tiers A/B/C by (license × access); mirror ≠ redistribution grant | accepted | | [0011](/decisions/0011-rclone-mirror-sha256) | rclone private mirror; SHA-256 authoritative fixity | accepted | | [0012](/decisions/0012-shared-substrate) | Shared asset substrate — share the mechanism, not the file | accepted | | [0013](/decisions/0013-experiment-backend-contract) | Experiment-backend *contract* + pluggable backend | accepted | | [0014](/decisions/0014-progress-cross-cutting) | `progress` cross-cutting; semantic roll-up; anti-Goodhart | accepted | | [0015](/decisions/0015-defend-cross-cutting) | `defend` cross-cutting; guardrail stop-and-confirm; teaching; non-grading | accepted | | [0016](/decisions/0016-mentor-personas) | Mentor personas author-selectable; no personality inference | accepted | | [0017](/decisions/0017-research-init-one-skill) | `research-init` one skill, init/adopt modes | accepted | | [0018](/decisions/0018-git-native-source-of-truth) | Git-native plain text as source of truth; cross-repo out of scope | accepted | | [0019](/decisions/0019-public-plugin-visibility) | Public plugin, named `defendable-science` | accepted | | [0020](/decisions/0020-bib-format-csl-json) | CSL-JSON source of truth; BibTeX exported on demand | accepted | | [0021](/decisions/0021-thesis-gate-per-gap-confirmation) | Thesis gate — per-gap acknowledged confirmation | accepted | | [0022](/decisions/0022-license-apache-2.0) | License the plugin under Apache-2.0 | accepted | | [0023](/decisions/0023-engineering-delegation-contract) | Delegate engineering via a contract, not a named tool | accepted | | [0024](/decisions/0024-tooling-package-and-bootstrap) | Supporting tooling = one package (Typer CLI, optional MCP) + markdown `ensure-tooling` bootstrap | accepted | | [0025](/decisions/0025-disclosure-and-citation-growth) | Honest AI-use disclosure + citation as the growth mechanism (auto-proposed by `paper-synthesis`) | accepted | | [0026](/decisions/0026-independent-versioning-compat-pin) | Independent plugin/package versioning with a compatibility pin | accepted | | [0027](/decisions/0027-release-driven-oidc-publish) | Publish via GitHub Releases + PyPI Trusted Publishing (OIDC, no tokens) | accepted | | [0028](/decisions/0028-100-percent-coverage-gate) | 100% statement + branch coverage gate on the package | accepted | | [0029](/decisions/0029-api-key-handling) | Unified API-key handling — a CLI-managed JSON store (not `.env`), stdin writes, scoped in-memory env for children | accepted | | [0030](/decisions/0030-docs-site-mintlify) | Docs site on Mintlify, published from a generated docs repo to `defendable.science` | accepted | | [0031](/decisions/0031-config-driven-cache-dir) | Source the cache root from `config.yml` (`cache_dir:`), not a hardcoded literal | accepted | | [0032](/decisions/0032-keys-store-outside-repo-by-default) | Key store defaults outside the repo (XDG); in-repo store is an explicit, gitignored opt-in with a committable-store runtime guardrail | accepted | | [0033](/decisions/0033-evidentiary-point-records) | Evidentiary per-point accountability records (`points`, not a bare pass/fail), shared by `defend` and `digest` | accepted | | [0034](/decisions/0034-digest-skill) | Dedicated `digest` skill for inbound paper comprehension, not a `defend` target | accepted | | [0035](/decisions/0035-rename-to-defendable-science) | Rename the project to Defendable Science — repo, plugin, distribution, CLI, config dir, env vars, and docs domain; old PyPI packages dropped | accepted | Format: MADR (Markdown Any Decision Records). Deciders: Davor Runje (with Claude).