diff --git a/.well-known/README.adoc b/.well-known/README.adoc
new file mode 100644
index 0000000..166b4ac
--- /dev/null
+++ b/.well-known/README.adoc
@@ -0,0 +1,17 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= .well-known/ — RFC 8615 web metadata
+:status: populated web metadata (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+RFC 8615 well-known URIs, published at the site root when this repo is served
+via GitHub Pages. This is web-standard metadata, *not* a pipeline data sink —
+it is listed here only because issue #5 grouped it with the data dirs.
+
+Files::
+* `ai.txt` — AI interaction policy (training/summarisation/generation
+  permissions; PMPL Emotional Lineage requirement). Points agents at
+  `0-AI-MANIFEST.a2ml`.
+* `humans.txt` — humanstxt.org credits/colophon.
+* `security.txt` — RFC 9116 security disclosure contact.
+
+Writer: maintainers (hand-edited). Reader: web clients, crawlers, AI
+agents, security researchers. Edited in place; not deleted.
diff --git a/dispatch/README.adoc b/dispatch/README.adoc
new file mode 100644
index 0000000..4f93a19
--- /dev/null
+++ b/dispatch/README.adoc
@@ -0,0 +1,21 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= dispatch/ — scan→triage work queue
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+Append-only JSONL work queue emitted by the panic-attacker scan → triage
+stage. One JSON object per line; one finding per object.
+
+Files::
+* `dispatch-YYYY-MM-DD.jsonl` — the day's dispatched findings (immutable once
+  written).
+* `pending.jsonl` — findings accepted but not yet actioned by a bot.
+* `held.jsonl` — findings deferred or quarantined (e.g. low confidence,
+  manual-review required).
+
+Record shape::
+`action`, `category`, `confidence`, `auto_fixable`, `description`,
+`pattern_id`, `recipe_id`, `replacement`, `repo`, `program_path`, `severity`,
+`strategy`, `tier`, `timestamp`.
+
+Writer: the triage stage of the ingest pipeline. Reader: the autofix /
+review bots. Append-only — never rewritten in place.
diff --git a/docs/decisions/ADR-0002-data-subdir-population.adoc b/docs/decisions/ADR-0002-data-subdir-population.adoc
new file mode 100644
index 0000000..0ea8bf0
--- /dev/null
+++ b/docs/decisions/ADR-0002-data-subdir-population.adoc
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+// Copyright (c) 2026 Jonathan D.A. Jewell (hyperpolymath) <j.d.a.jewell@open.ac.uk>
+= ADR-0002: data subdirectories are populated and self-documented, not removed
+:revdate: 2026-05-17
+:status: Accepted
+
+== Status
+
+Accepted — 2026-05-17.
+
+Resolves: https://github.com/hyperpolymath/verisimdb-data/issues/5[V-L3-S1].
+
+Builds on: xref:ADR-0001-repo-purpose.adoc[ADR-0001] (the two-purpose framing;
+these directories are purpose 1, the flat-file data store).
+
+== Context
+
+Issue #5 observed that `dispatch/`, `patterns/`, `recipes/`, `outcomes/`,
+`policy/`, `health/` and `.well-known/` looked like "mostly empty placeholder
+directories" and asked for a binary decision: *populate* them (one issue per
+subdir to make the contents concrete) or *delete* them and rebuild as content
+lands.
+
+That premise is now stale. ADR-0001 already declared all of these subtrees as
+the repository's flat-file data store, and the panic-attacker → triage →
+autofix pipeline has since filled every one of them with production data:
+
+* `dispatch/` — dated work-queue JSONL plus `pending.jsonl` / `held.jsonl`
+* `patterns/` — `registry.json` (the cross-repo pattern registry)
+* `recipes/` — ~46 `recipe-*.json` remediation recipes
+* `outcomes/` — monthly applied-fix ledgers plus a fleet import
+* `policy/` — `policy.ncl` (the Nickel baseline policy contract)
+* `health/` — `sitrep.txt` and `hypatia.json` operational telemetry
+* `.well-known/` — RFC 8615 web metadata (`ai.txt`, `humans.txt`,
+  `security.txt`)
+
+There are no empty placeholder directories left in the tree. The only real
+gap against the issue's acceptance criteria was the absence of a per-directory
+README explaining what each one holds.
+
+== Decision
+
+**Populate, not remove.** The data subdirectories are a permanent, declared
+part of the repository (per ADR-0001) and are now backed by real, actively
+written content. They are *not* placeholders and must not be deleted.
+
+To make the contents concrete and self-documenting — the spirit of the issue's
+"one issue per subdir" alternative, achieved in one PR — each data directory
+carries a `README.adoc` that states:
+
+. what artefact(s) it holds and their on-disk format,
+. who writes it (which pipeline stage / bot) and who reads it,
+. whether it is append-only or rewritten in place,
+. its retention / `.gitkeep` behaviour.
+
+Splitting this into seven tracking issues was rejected: the directories are
+already populated, so seven issues would each open and immediately close as
+"add a README" with no design content. One ADR plus seven READMEs records the
+decision and discharges the acceptance criteria together.
+
+== Consequences
+
+. The `dispatch/ patterns/ recipes/ outcomes/ policy/ health/ .well-known/`
+  directories each gain a `README.adoc`; no data files are moved or removed.
+. `outcomes/.gitkeep` is retained: the ledger is monthly, so the directory
+  can legitimately contain only `.gitkeep` at the start of a month, and that
+  is now documented rather than mistaken for an empty placeholder.
+. Future data directories under purpose 1 must ship with a `README.adoc`
+  from the first commit (close the "is this a placeholder?" question at
+  creation time).
+. ADR-0001's directory-layout list remains authoritative for *which* subtree
+  serves *which* purpose; this ADR documents the *contents* of the data
+  subtree.
+
+== Alternatives considered
+
+Delete the directories and rebuild as content lands::
+Rejected. The content has already landed and is being written continuously by
+the ingest/triage/autofix pipeline; deleting live data sinks would break that
+pipeline and lose history. This alternative only made sense under the (now
+false) assumption that the directories were empty.
+
+One tracking issue per subdirectory::
+Rejected as busywork. The directories are populated; each issue would reduce
+to "add a README." Folded into this single ADR + the seven READMEs instead.
+
+Move telemetry / web metadata out of the data store::
+Out of scope. `health/` and `.well-known/` are small and co-located by
+design; relocating them is a separate decision if either grows.
diff --git a/health/README.adoc b/health/README.adoc
new file mode 100644
index 0000000..fe38e54
--- /dev/null
+++ b/health/README.adoc
@@ -0,0 +1,17 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= health/ — operational telemetry
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+Latest-state operational telemetry for the data store and its pipeline.
+These are *snapshots* (rewritten in place each run), not append-only logs —
+history lives in `outcomes/` and `dispatch/`.
+
+Files::
+* `sitrep.txt` — human-readable situation report: last-run timestamp,
+  counts (scans / patterns / actions / outcomes / recipes), the
+  auto/review/report split, process up/down, contingency status, run time.
+* `hypatia.json` — machine-readable health snapshot consumed by the
+  Hypatia scanner / Scorecard surface.
+
+Writer: the pipeline's reporting stage. Reader: humans (sitrep) and the
+Hypatia / health-monitoring tooling (json).
diff --git a/outcomes/README.adoc b/outcomes/README.adoc
new file mode 100644
index 0000000..6c7675c
--- /dev/null
+++ b/outcomes/README.adoc
@@ -0,0 +1,19 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= outcomes/ — applied-fix outcome ledger
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+Append-only JSONL ledger of what happened when a recipe was applied to a
+repo. Feeds the recipe success/fail counters and health reporting.
+
+Files::
+* `YYYY-MM.jsonl` — one month of outcomes (immutable once the month closes).
+* `*-fleet-import.jsonl` — bulk historical imports.
+* `.gitkeep` — retained on purpose: the ledger is monthly, so at the start
+  of a month this directory can legitimately hold only `.gitkeep`. That is
+  the expected steady state, *not* an empty placeholder (see ADR-0002).
+
+Record shape::
+`pattern`, `repo`, `bot`, `outcome` (`success` / failure), `fixed_at`.
+
+Writer: the autofix/review bots after each attempt. Reader: recipe
+counter rollup and `health/`. Append-only.
diff --git a/patterns/README.adoc b/patterns/README.adoc
new file mode 100644
index 0000000..ba4896d
--- /dev/null
+++ b/patterns/README.adoc
@@ -0,0 +1,16 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= patterns/ — canonical cross-repo pattern registry
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+`registry.json` — the single canonical pattern registry. It deduplicates
+raw findings across every scanned repo into trackable, long-lived patterns.
+
+Shape::
+Top-level `description`, `last_updated`, and a `patterns` map keyed by
+pattern id. Each entry: `id`, `category`, `description`, `pa_rule`,
+`occurrences`, `first_seen`, `last_seen`, `recipe_id` (the remediation
+recipe, or `null` if none yet), `repo_paths`.
+
+Writer: the pattern-aggregation stage of the ingest pipeline (rewritten in
+place on each run — it is a registry, not a log). Reader: triage (to attach
+`recipe_id`) and reporting/health.
diff --git a/policy/README.adoc b/policy/README.adoc
new file mode 100644
index 0000000..ba1f098
--- /dev/null
+++ b/policy/README.adoc
@@ -0,0 +1,18 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= policy/ — baseline policy contract
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+`policy.ncl` — the Nickel baseline policy contract for this data store.
+
+Contents::
+* `enforcement` — repo hygiene flags (`require_spdx_headers`,
+  `require_ci_security_checks`, `block_committed_secrets`,
+  `require_pinned_actions`).
+* `triage` — confidence thresholds: `auto_fix_min_confidence`,
+  `review_min_confidence`.
+* `version` — policy schema version.
+
+Writer: maintainers (hand-edited; version-bumped on change). Reader:
+`contractiles/must/Mustfile` consumes it to gate enforcement, and triage
+reads the confidence thresholds to choose auto-fix vs. review vs. report.
+Single file, edited in place.
diff --git a/recipes/README.adoc b/recipes/README.adoc
new file mode 100644
index 0000000..f64f9ae
--- /dev/null
+++ b/recipes/README.adoc
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: PMPL-1.0-or-later
+= recipes/ — remediation recipe knowledge base
+:status: populated data store (see docs/decisions/ADR-0002-data-subdir-population.adoc)
+
+One `recipe-*.json` file per remediation recipe — the knowledge base the
+autofix bot consults to turn a matched pattern into a fix.
+
+Shape (per file)::
+`id`, `action`, `description`, `confidence`, `auto_fixable`, `fix_script`
+(or `null`), `languages`, `match`, `replacement`, `pattern_ids` (the
+patterns this recipe remediates), `proven_module`, `triangle_tier`
+(eliminate / substitute / …), and running counters
+`total_attempts` / `successful_fixes` / `failed_fixes`.
+
+Naming::
+`recipe-<slug>.json`; the `recipe-scorecard-*` family maps to OpenSSF
+Scorecard checks.
+
+Writer: curated, plus counter updates from outcome ingest. Reader: the
+autofix bot and triage (recipe lookup by `pattern_id`).