feat(atlas): UX sprint — progressive disclosure (Plain default + transitions)

Makes the atlas self-explanatory. Every stage/node gains a plain-language layer
(plain_title + plain "what happens" + jargon-free node text); the previous
technical copy demotes to a subtitle + on-demand detail. A Plain⇄Technical
toggle (default Plain, persisted) flips the whole atlas. Biggest win: every
transition arrow is now LABELLED with "what must be true to advance" (the gated-
flow story that was invisible), gate hops (human @04, CI @06) styled distinctly.
Spine repositioned into a uniform header band so labels never collide with copy.

Copy grounded in a fresh-eyes UX review (docs/UX-REVIEW.md, reviewer≠implementer).
Data model: plain_title/plain/trans_label/trans on Stage, plain on Node — guarded
by a test (every stage has plain_title + a transition). Live overlays unchanged.

Verified: build/vet/lint(0)/test green; Plain render screenshot-checked.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.gitea/workflows/cd.yml

@@ -62,18 +62,23 @@ jobs:
       image-tag: ${{ steps.meta.outputs.sha-tag }}
     steps:
       - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0   # full history + tags so `git describe` sees the SemVer tag
 
       - name: Derive image tags
         id: meta
         run: |
           SHA=$(git rev-parse --short HEAD)
+          VERSION=$(git describe --tags --always --dirty)
           echo "sha-tag=${SHA}" >> "$GITHUB_OUTPUT"
+          echo "version=${VERSION}" >> "$GITHUB_OUTPUT"
 
       - name: Build and push to local registry
         run: |
           REGISTRY="localhost:5000"
           REF="${REGISTRY}/${{ env.IMAGE }}:${{ steps.meta.outputs.sha-tag }}"
           buildah build \
+            --build-arg VERSION="${{ steps.meta.outputs.version }}" \
             --label "org.opencontainers.image.revision=${{ github.sha }}" \
             -t ${REF} \
             -t ${REGISTRY}/${{ env.IMAGE }}:latest \

Dockerfile

@@ -5,7 +5,10 @@ RUN go install github.com/a-h/templ/cmd/templ@latest
 COPY go.mod ./
 RUN go mod download
 COPY . .
-RUN templ generate && CGO_ENABLED=0 go build -trimpath -ldflags='-s -w' -o /out/app ./cmd/cad-atlas
+ARG VERSION=dev
+RUN templ generate && CGO_ENABLED=0 go build -trimpath \
+    -ldflags="-s -w -X git.d-ma.be/mathias/cad-atlas/internal/version.Value=${VERSION}" \
+    -o /out/app ./cmd/cad-atlas
 
 FROM gcr.io/distroless/static-debian12:nonroot
 COPY --from=build /out/app /app

docs/INCEPTION-OATH.md

@@ -28,9 +28,10 @@ Oath holds. Methodology: brain `wiki/homelab/decisions/inception-sprint-and-oath
 
 ## Deployment
 
-Live on koala k3s: namespace `cad-atlas`, 1 replica, served via ClusterIP `cad-atlas:80 → :8080`
+Live at **https://atlas.d-ma.be** (Authentik forward-auth — authed users only) on koala k3s:
-(manifests in `mathias/infra` `k3s/apps/cad-atlas/`, reconciled by Flux). First deploy was
+namespace `cad-atlas`, 1 replica, `cad-atlas:80 → :8080` (manifests in `mathias/infra`
-GitOps-direct; the CI `deploy` job auto-deploys future pushes once `INFRA_DEPLOY_KEY` is set (#2).
+`k3s/apps/cad-atlas/`, reconciled by Flux). CI `deploy` job **auto-deploys** on every main push —
+`INFRA_DEPLOY_KEY` (write deploy-key `cad-atlas-ci` on infra) is wired (#2 done).
 
 ## The honesty rule
 

docs/UX-REVIEW.md

@@ -0,0 +1,215 @@
+# CAD Atlas — Fresh-Eyes UX Review
+
+Reviewer role: fresh-eyes UX (not implementer). This document critiques the copy and
+information architecture of the "From Signal to Pod" atlas and specifies the
+progressive-disclosure layer for the coming sprint. It does not change code.
+
+Sources reviewed:
+- `internal/atlas/atlas.json` (authored stages + nodes)
+- `internal/web/handler.go` (which stages get live data overlaid)
+- `.context/PROJECT.md` (ground-truth meaning of each stage/gate)
+
+---
+
+## 1. Diagnosis — why the current atlas is hard for a naive viewer
+
+The atlas is written by the person who built the pipeline, for the person who built the
+pipeline. Almost every node names a **mechanism** (`Ed25519 admission controller`,
+`var-go Oath`, `dma-cli`, `assessor-loop ledger`, `agentsquad`, `ISC`, `TELOS`) rather
+than the **thing that happens to a piece of work**. A compliance officer or a new
+engineer cannot answer the two questions they actually have: *"what is happening to the
+work at this step?"* and *"why does it move to the next step?"*
+
+The second question is completely unanswered. The atlas renders nine stages connected by
+arrows, but the arrows carry **zero copy**. There is no statement of what has to be true
+for work to advance — which is exactly where the interesting governance lives (a human
+sign-off at 04, a green CI gate at 06, an integrity check at 03). The pipeline's whole
+selling point is "auditable, gated flow," yet the gates between stages are invisible.
+A viewer sees a row of jargon boxes and an implied left-to-right drift, with no sense of
+what earns each hop.
+
+Two smaller aggravators: (a) the reel is dense — 20+ nodes, colour-coded pills whose
+meaning is never keyed, and Swedish-homelab proper nouns (koala, iguana, flamingo) that a
+stakeholder can't decode; (b) Stage 06 is authored-empty (it's generated live from CI),
+so on a cold/offline load that column can read as "nothing happens here," which is the
+opposite of the truth — CI is a governance gate.
+
+The fix is not to dumb it down. It is to make **plain-language the default layer** — a
+one-line "what happens + why" per stage, and a labelled "what must be true to advance"
+per arrow — and demote today's precise, correct technical copy to an **on-demand layer**.
+
+---
+
+## 2. Per-stage plain-language map (00–08)
+
+| stage | plain_title | plain_what (one jargon-free sentence) | keep_technical (on-demand) |
+|---|---|---|---|
+| **00 Signals** | Notice what's happening | New ideas and developments worth reacting to are collected — mostly an automated daily/weekly scan of AI news, plus things saved by hand. | "Signals → mathias/signals"; Applied AI Radar (Tier-1 daily / Tier-2 weekly), verified-primary bar; brain capture; aspirational inbox surfaces (not built). |
+| **01 TELOS** | Why we're here | The mission, goals, and problems we're actually trying to solve live here — every piece of work downstream has to trace back to one of these goals. | "TELOS — intention substrate", `wiki/telos/`, `brain_query wing=telos`. |
+| **02 Strategic session** | Think it through | A human and AI models work out *what* to do and *why*, debating hard calls and writing down the decision and what "done" will mean. | "Strategic session" — claude.ai frontier + brain MCP; ADRs/specs; ISC acceptance criteria; LLM Council (fan-out → anonymous cross-review → chairman synth); Autoresearch Council. |
+| **03 Spec → Gitea issue** | Write the work order | The decision is turned into a precise, self-contained work order that an AI agent can execute unsupervised — with a pass/fail definition of done, a risk rating, and a tamper-proof seal. | "Spec → Gitea issue" — binary ISC, risk tier LOW/MED/HIGH, reg-risk assessment, no open human deps; Ed25519 admission controller (#36); var-go Oath (single fenced block, fail-closed). |
+| **04 Human dispatch gate** | Human says go | A person reviews the work order and its risk and decides whether to release it — this is the one and only checkpoint where work does not move on its own. | "Human dispatch gate — the only checkpoint"; ratify plan + risk tier; Session-Dispatch bridge (claude.ai MCP → `workflow_run_trigger` → `cad-dispatch.yml` → agentsquad); dispatch-allow eligibility. |
+| **05 Execute · agentsquad** | Agents do the work | AI agents actually build the thing — one writes, a second independent one reviews it to avoid marking its own homework — and every step is logged for the audit trail. | "Execute · agentsquad" on koala; Task API (`POST /tasks`); executor+reviewer loop (ADK Go + LiteLLM, reviewer on distinct tier); dma-cli routing + 3-layer scope guardrail; assessor-loop attestation ledger + brain session_log. |
+| **06 PR → CI** | Automatic quality checks | The proposed change is run through automated tests and safety checks — including a check that it actually satisfies the work order's definition of done — and only a clean pass lets it continue. | "PR → CI" — Gitea Actions `cd.yml`; go test/vet/lint/govulncheck; **var-go/oath gate** (correctness floor over the reviewer, anti-rubber-stamp #55). *Nodes generated live from the latest CI run.* |
+| **07 CD → pod** | Ship it | Once everything is green, the change is deployed automatically to the live server — with the rule that merging code alone doesn't ship it; the release has to be pointed at the new version. | "CD → pod" — Flux GitOps → k3s on koala; "push ≠ deploy: bump tag in mathias/infra"; ntfy on deploy. *Deploy + Flux state overlaid live.* |
+| **08 Loop back** | Did it work? | The result is scored against the goal that started it and fed back into the mission board, so the next round of planning learns from what shipped. | "Loop back → TELOS (feedback bus)"; session_log + attestation → brain; outcome scored vs originating goal; arc partly manual (improvement target). |
+
+---
+
+## 3. Per-node plain restatements
+
+The existing `d` text stays as the **technical detail layer**. Each `plain` line below is
+the jargon-free default. Kept accurate to PROJECT.md.
+
+**Stage 00 — Signals**
+- *Applied AI Radar* → **"An automated scan reads AI news every day (and deeper every week) and keeps only claims backed by a real paper, benchmark, code, or named lab."**
+- *Manual capture* → **"Anything interesting spotted by hand gets saved into the same inbox."**
+- *Aspirational surfaces* → **"Planned-but-not-built: sending ideas in by Telegram, voice, or a URL."** (mark clearly as a gap / not yet real.)
+
+**Stage 01 — TELOS**
+- *Intention substrate* → **"The master list of mission, goals, problems, and current status — the yardstick everything downstream is measured against."**
+
+**Stage 02 — Strategic session**
+- *Design · ADRs · specs* → **"A human and a top-tier AI model figure out the approach and write down the decision plus what a finished result must prove."**
+- *LLM Council* → **"For hard calls, several AI models answer independently, anonymously critique each other, and a 'chair' model synthesises one verdict — reduces any single model's bias."**
+- *Autoresearch Council* → **"A parallel version of the same review that vets research findings before they're allowed through."**
+
+**Stage 03 — Spec → Gitea issue**
+- *Contract enforced* → **"The work order must have a clear pass/fail test, a risk rating, a regulatory-risk note, and no unfinished human dependencies before it counts as agent-ready."**
+- *Admission controller* → **"The work order is cryptographically signed when created, so any later tampering is detectable and the eventual change can be checked against it."**
+- *var-go Oath* → **"A machine-checkable 'definition of done' is embedded in the work order — exactly one, or the order is rejected — later used to prove the result actually meets the spec."**
+
+**Stage 04 — Human dispatch gate**
+- *Human triggers execution* → **"A person confirms the plan and its risk level, then releases the work — nothing runs until they do."**
+- *Session-Dispatch bridge* → **"The approval flips a switch that hands the signed work order over to the agents to start execution."**
+
+**Stage 05 — Execute · agentsquad**
+- *Task API* → **"A request kicks off a job and hands back an id you can poll for progress."**
+- *Executor + reviewer loop* → **"One agent does the work; a second, independent agent on a different model reviews it — so nothing marks its own homework."**
+- *dma-cli · routing + scope* → **"A router sends each agent to the right AI backend and enforces what it is and isn't allowed to touch, with a confirmation gate as a guardrail."**
+- *assessor-loop ledger* → **"Every step is recorded in a tamper-evident log so the whole run can be audited afterwards."**
+
+**Stage 06 — PR → CI** *(nodes generated live from the latest CI run — no authored nodes)*
+- Live jobs render here; the plain framing for the column is: **"Automated tests and safety checks run on the proposed change, including a check that it truly satisfies the work order — only a clean pass moves on."**
+
+**Stage 07 — CD → pod**
+- *Deploy on green* → **"When all checks pass, the release system rolls the new version onto the live server automatically — but only once the release is pointed at that version (merging code alone doesn't ship it)."** *(live deploy + Flux status also shown.)*
+
+**Stage 08 — Loop back**
+- *Close the loop* → **"The outcome is scored against the goal that started it and written back to the mission board, so future planning learns from what actually shipped."**
+
+---
+
+## 4. Transitions — the key deliverable
+
+For each arrow: *what moves the work forward, and what must be true for it to advance.*
+These should be rendered **on the arrows themselves** (see §5). Today they are blank.
+
+- **00 → 01 — "Does it matter to us?"**
+  A raw signal only advances if it connects to something we actually care about. Most
+  captured signals stop here; the few that touch the mission get pulled up against a goal.
+
+- **01 → 02 — "Worth a session?"**
+  A goal or problem on the board becomes the seed for a design session when it's decided
+  it's worth working on now. The goal is the input the session must trace back to.
+
+- **02 → 03 — "Decision reached."**
+  Once the debate converges on a decision (and what "done" will mean), it advances only
+  when that thinking is written down as a concrete, testable specification — not while
+  the answer is still open.
+
+- **03 → 04 — "Order written, sealed, agent-ready."**
+  Work advances to the gate only when the spec is a complete contract: a pass/fail test, a
+  risk tier, a regulatory note, no open human dependencies, one embedded Oath, and a valid
+  cryptographic signature. A malformed or unsigned order fails closed and does not reach
+  the gate.
+
+- **04 → 05 — "A human said go."**
+  This is the hard stop. Nothing crosses automatically. A person must review the plan and
+  risk and explicitly release it, and the repo must be on the allow-list, before any agent
+  starts. This is the single human checkpoint in the whole pipeline.
+
+- **05 → 06 — "Agents produced a change."**
+  Work advances when the agents finish and open a proposed change (a PR) with its audit
+  log attached. Until there's a concrete change to test, nothing moves.
+
+- **06 → 07 — "All checks green."**
+  The change advances only if every automated check passes — tests, linters, security
+  scan, **and** the Oath check proving it meets the original work order. Any red gate stops
+  it here; a passing reviewer is not enough to override a failed Oath.
+
+- **07 → 08 — "It's live."**
+  Once the new version is actually running on the server, the deployed outcome becomes the
+  input to scoring. Advancing means "shipped and observable," not just "merged."
+
+- **08 → TELOS (feedback bus, dashed) — "What did we learn?"**
+  The scored outcome flows back into the mission board so goals, problems, and priorities
+  update. This is the loop that makes the pipeline a cycle rather than a line. Note per
+  PROJECT.md this arc is **partly manual today** and is an explicit improvement target —
+  the dashed styling should read as "aspirational / not fully automated," not just decorative.
+
+---
+
+## 5. Progressive-disclosure recommendations
+
+**Default (Plain) layer — what everyone sees on load:**
+- Each stage column shows: the **plain_title** as the headline, the technical title as a
+  smaller subtitle, and the one-line **plain_what** directly under it.
+- Each node shows its **plain** one-liner as the primary text. The current `d` string is
+  hidden by default.
+- Each arrow shows a short **transition label** (the bolded phrase from §4, e.g. "A human
+  said go", "All checks green") — this is the single biggest comprehension win and must
+  ship in the default layer, not behind a toggle.
+
+**On hover / expand (per node):**
+- Reveal the technical `d` text, the `tags`, and the pill's meaning.
+- On the arrow, hovering expands the short label into the full "what must be true to
+  advance" sentence from §4.
+
+**Plain ⇄ Technical toggle (global):**
+- A single top-level switch, defaulting to **Plain**. Persist the choice (localStorage).
+- Plain: plain_title headline, plain_what, plain node lines, short arrow labels. Proper
+  nouns (koala/iguana/agentsquad/TELOS) suppressed or shown only as a footnote.
+- Technical: today's exact copy — titles, `d` strings, tags, substrate host specs — i.e.
+  the atlas as it exists now. Nothing is lost; the current view becomes "Technical."
+- The toggle should crossfade in place, not reflow the whole layout, so a viewer can flip
+  back and forth and map plain↔technical on the same node.
+
+**Visual cues for the currently-bare transitions:**
+- Give every arrow a **label chip** sitting on the spine. Gate arrows (04→05 human, 06→07
+  CI) get a distinct treatment — a lock/shield glyph and a stronger colour — because those
+  are the governance moments the whole atlas exists to show.
+- Make the **08 → TELOS feedback bus** visibly different (dashed + "partly manual" tag) so
+  its aspirational status is honest, matching PROJECT.md's dogfooding-honesty discipline.
+- Add a small, always-visible **legend** keying the pill colours and the three gate types
+  (integrity / eligibility / correctness), since colour is currently unexplained.
+- For **Stage 06** (authored-empty, generated live): when no live CI data is present, show
+  the plain_what placeholder ("Automated tests and safety checks run…") rather than an
+  empty column, so it never reads as "nothing happens here."
+
+**Three-gate overlay (stretch, high value for the compliance audience):**
+- A "show governance gates" toggle that highlights the three orthogonal gates on top of
+  the pipeline: integrity (03, Ed25519), eligibility (04/05, dispatch-allow), correctness
+  (06, Oath). This directly serves the "audit chain is the viz data" thesis for a
+  compliance/exec viewer.
+
+---
+
+## 6. Prioritized punch list (top 8 by comprehension impact)
+
+1. **Label every arrow with a plain "what must be true to advance" phrase** (§4). Biggest
+   miss, biggest win — turns a row of boxes into a story of gated flow. Default layer.
+2. **Add a plain_what one-liner per stage** as the default column copy (§2), with the
+   technical title demoted to subtitle.
+3. **Ship the Plain ⇄ Technical global toggle, defaulting to Plain**, persisting choice;
+   current copy becomes the Technical view (nothing thrown away).
+4. **Rewrite node primary text to the plain lines** (§3); move existing `d` to hover/expand.
+5. **Visually distinguish the two real gates (04 human, 06 CI)** with lock/shield glyphs and
+   stronger colour so the checkpoints read as checkpoints.
+6. **Add a legend** keying pill colours and the three gate types — colour currently carries
+   meaning nobody can decode.
+7. **Fix the Stage-06 empty-column problem**: show a plain placeholder when live CI data is
+   absent, so the CI gate never looks like a no-op.
+8. **Make the 08→TELOS feedback bus honestly aspirational** (dashed + "partly manual" tag),
+   and suppress homelab proper nouns (koala/iguana/flamingo/agentsquad/TELOS) in Plain mode,
+   surfacing them only in Technical or a footnote.

embed.go

@@ -0,0 +1,12 @@
+// Package cadatlas embeds module-root build artifacts that live outside any
+// subpackage directory. go:embed cannot reach parent directories, so the CI
+// workflow — which the atlas parses to generate its CI/CD stage from the real
+// pipeline — is embedded here at the module root.
+package cadatlas
+
+import _ "embed"
+
+// CDWorkflow is the repo's own Gitea Actions workflow, embedded at build time.
+//
+//go:embed .gitea/workflows/cd.yml
+var CDWorkflow []byte

go.mod

@@ -3,3 +3,5 @@ module git.d-ma.be/mathias/cad-atlas
 go 1.26
 
 require github.com/a-h/templ v0.3.1020
+
+require gopkg.in/yaml.v3 v3.0.1

go.sum

@@ -2,3 +2,7 @@ github.com/a-h/templ v0.3.1020 h1:ypAT/L5ySWEnZ6Zft/5yfoWXYYkhFNvEFOeeqecg4tw=
 github.com/a-h/templ v0.3.1020/go.mod h1:A2DlK61v+K+NRoGnhmYbNYVmtYHcFO5/AisMvBdDxTM=
 github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
 github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

internal/atlas/atlas.json

@@ -0,0 +1,55 @@
+{
+  "substrate": [
+    {"n":"koala","k":"RTX 5070 · k3s control-plane · Gitea · LiteLLM :30401 · llama-swap :31234 · searxng"},
+    {"n":"iguana","k":"M2 Ultra · Ollama / mlx"},
+    {"n":"flamingo","k":"daily driver · ~/dev"},
+    {"n":"piguard","k":"NGINX reverse-proxy · ntfy"}
+  ],
+  "ns": "Tailscale mesh · ns: ai-stack · supervisor(→brain) · gitea-mcp · infra-mcp · council",
+  "stages": [
+    {"no":"STAGE 00","title":"Signals","plain_title":"Notice what's happening","plain":"New ideas and developments worth reacting to are collected — mostly an automated daily/weekly scan of AI news, plus things saved by hand.","path":"→ mathias/signals",
+     "trans_label":"Does it matter to us?","trans":"A raw signal only advances if it connects to something we actually care about. Most captured signals stop here; the few that touch the mission get pulled up against a goal.","nodes":[
+      {"t":"Applied AI Radar","plain":"An automated scan reads AI news every day (and deeper every week) and keeps only claims backed by a real paper, benchmark, code, or named lab.","d":"Daily Tier-1 + weekly Tier-2 deep pass. Verified-primary bar (paper/benchmark/code/named-lab).","tags":["cron · daily/weekly","→ signals #1–26+"]},
+      {"t":"Manual capture","plain":"Anything interesting spotted by hand gets saved into the same inbox.","d":"claude.ai strategic drop · brain capture tool.","tags":["ad-hoc"]},
+      {"t":"Aspirational surfaces","plain":"Planned but not built yet: sending ideas in by Telegram, voice, or a URL.","pill":"var(--dim)","d":"Telegram / voice / URL → inbox. NOT built.","tags":["gap"]}
+    ]},
+    {"no":"STAGE 01","cls":"telos","title":"TELOS","plain_title":"Why we're here","plain":"The mission, goals, and problems we're actually trying to solve live here — every piece of work downstream has to trace back to one of these goals.","path":"wiki/telos/",
+     "trans_label":"Worth a session?","trans":"A goal or problem on the board becomes the seed for a design session when it's decided worth working on now. The goal is the input the session must trace back to.","nodes":[
+      {"t":"Intention substrate","plain":"The master list of mission, goals, problems, and current status — the yardstick everything downstream is measured against.","pill":"var(--violet)","d":"Mission · goals · problems · strategies · status. Every downstream item traces to a goal.","tags":["brain_query wing=telos"]}
+    ]},
+    {"no":"STAGE 02","title":"Strategic session","plain_title":"Think it through","plain":"A human and AI models work out what to do and why, debating hard calls and writing down the decision and what \"done\" will mean.","path":"claude.ai frontier + brain MCP",
+     "trans_label":"Decision reached","trans":"It advances only when the thinking converges on a decision and is written down as a concrete, testable specification — not while the answer is still open.","nodes":[
+      {"t":"Design · ADRs · specs","plain":"A human and a top-tier AI model figure out the approach and write down the decision plus what a finished result must prove.","d":"Human + frontier model. ISC acceptance criteria written here.","tags":["Define / converge"]},
+      {"t":"🏛️ LLM Council","plain":"For hard calls, several AI models answer independently, anonymously critique each other, and a \"chair\" model synthesises one verdict — reducing any single model's bias.","cls":"council","pill":"var(--violet)","d":"fan-out → anonymous cross-review → chairman synth. glm-4.7-flash · qwen36-35b · gemma4-31b (chair).","tags":["hard strategic Q","chat.d-ma.be"]},
+      {"t":"Autoresearch Council","plain":"A parallel version of the same review that vets research findings before they're allowed through.","cls":"council","pill":"var(--violet)","d":"Sibling pipe — ratifies research before the gate.","tags":["proposed: → standalone svc"]}
+    ]},
+    {"no":"STAGE 03","title":"Spec → Gitea issue","plain_title":"Write the work order","plain":"The decision is turned into a precise, self-contained work order an AI agent can execute unsupervised — with a pass/fail definition of done, a risk rating, and a tamper-proof seal.","path":"agent-ready contract",
+     "trans_label":"Order written, sealed, agent-ready","trans":"Advances to the gate only when the spec is a complete contract: a pass/fail test, a risk tier, a regulatory note, no open human dependencies, one embedded Oath, and a valid cryptographic signature. A malformed or unsigned order fails closed and never reaches the gate.","nodes":[
+      {"t":"Contract enforced","plain":"The work order must have a clear pass/fail test, a risk rating, a regulatory-risk note, and no unfinished human dependencies before it counts as agent-ready.","d":"Binary ISC · declared risk tier · reg-risk assessment · no open human deps.","tags":["LOW / MED / HIGH"]},
+      {"t":"Admission controller","plain":"The work order is cryptographically signed when created, so any later tampering is detectable and the eventual change can be checked against it.","d":"Ed25519-sign issue body at creation (#36). Verify sig + PR alignment at infra boundary.","tags":["chain of custody"]},
+      {"t":"⚖️ var-go Oath","plain":"A machine-checkable \"definition of done\" is embedded in the work order — exactly one, or the order is rejected — later used to prove the result actually meets the spec.","cls":"oath","pill":"var(--gold)","d":"Acceptance contract embedded in the issue as a var fenced block. Exactly one — zero/multiple fail closed. Prose → typed steps; failures anchored to byte spans.","tags":["swedsl · var-go","defined here → enforced @06"]}
+    ]},
+    {"no":"STAGE 04","cls":"gate","title":"Human dispatch gate","plain_title":"Human says go","plain":"A person reviews the work order and its risk and decides whether to release it — the one and only checkpoint where work does not move on its own.","path":"the only checkpoint",
+     "trans_label":"A human said go","trans":"The hard stop. Nothing crosses automatically — a person must review the plan and risk and explicitly release it, and the repo must be on the allow-list, before any agent starts. This is the single human checkpoint in the whole pipeline.","nodes":[
+      {"t":"Human triggers execution","plain":"A person confirms the plan and its risk level, then releases the work — nothing runs until they do.","cls":"gateway","pill":"var(--amber)","d":"Ratify proposed-plan + risk tier, then dispatch.","gate":true},
+      {"t":"Session-Dispatch bridge","plain":"The approval flips a switch that hands the signed work order over to the agents to start execution.","cls":"bridge","pill":"var(--blue)","d":"claude.ai MCP → gitea:workflow_run_trigger → cad-dispatch.yml → agentsquad. The final design→execution bridge.","tags":["workflow_dispatch"]}
+    ]},
+    {"no":"STAGE 05","cls":"exec","title":"Execute · agentsquad","plain_title":"Agents do the work","plain":"AI agents actually build the thing — one writes, a second independent one reviews it to avoid marking its own homework — and every step is logged for the audit trail.","path":"koala · cmd/agentsquad-serve",
+     "trans_label":"Agents produced a change","trans":"Advances when the agents finish and open a proposed change (a PR) with its audit log attached. Until there's a concrete change to test, nothing moves.","nodes":[
+      {"t":"Task API","plain":"A request kicks off a job and hands back an id you can poll for progress.","pill":"var(--coral)","d":"POST /tasks → job id · GET /tasks/{id}. taskqueue + serve (v0.12+).","tags":["single agentsquad.yaml"]},
+      {"t":"Executor + reviewer loop","plain":"One agent does the work; a second, independent agent on a different model reviews it — so nothing marks its own homework.","cls":"win","pill":"var(--coral)","d":"ADK Go + LiteLLM. Frontier models (local qwen spirals). Reviewer on distinct tier — echo-chamber prevention.","risk":true},
+      {"t":"dma-cli · routing + scope","plain":"A router sends each agent to the right AI backend and enforces what it is and isn't allowed to touch, with a confirmation gate as a guardrail.","cls":"bridge","pill":"var(--blue)","d":"Harness-config arm: routes agents to the right LLM backend. Three-layer scope policy + confirmation gate = CAD guardrail.","tags":["backend routing","scope guardrail"]},
+      {"t":"assessor-loop ledger","plain":"Every step is recorded in a tamper-evident log so the whole run can be audited afterwards.","d":"Attestation ledger (audit trail) + brain session_log on completion.","tags":["audit package"]}
+    ]},
+    {"no":"STAGE 06","title":"PR → CI","plain_title":"Automatic quality checks","plain":"The proposed change is run through automated tests and safety checks — including a check that it actually satisfies the work order's definition of done — and only a clean pass lets it continue.","path":"Gitea Actions · cd.yml (live)","generate":"ci-jobs",
+     "trans_label":"All checks green","trans":"Advances only if every automated check passes — tests, linters, security scan, and the Oath check proving it meets the original work order. Any red gate stops it here; a passing reviewer is not enough to override a failed Oath.","nodes":[]},
+    {"no":"STAGE 07","cls":"cd","title":"CD → pod","plain_title":"Ship it","plain":"Once everything is green, the change is deployed automatically to the live server — with the rule that merging code alone doesn't ship it; the release has to be pointed at the new version.","path":"Flux GitOps → k3s","generate":"deploy-state",
+     "trans_label":"It's live","trans":"Once the new version is actually running on the server, the deployed outcome becomes the input to scoring. Advancing means shipped and observable, not just merged.","nodes":[
+      {"t":"Deploy on green","plain":"When all checks pass, the release system rolls the new version onto the live server automatically — but only once the release is pointed at that version (merging code alone doesn't ship it).","pill":"var(--green)","d":"Flux reconciles image → k3s pod on koala. Push ≠ deploy: bump tag in mathias/infra.","tags":["ntfy on deploy"]}
+    ]},
+    {"no":"STAGE 08","cls":"telos","title":"Loop back","plain_title":"Did it work?","plain":"The result is scored against the goal that started it and fed back into the mission board, so the next round of planning learns from what shipped.","path":"→ TELOS (feedback bus)",
+     "trans_label":"What did we learn?","trans":"The scored outcome flows back into the mission board so goals, problems, and priorities update — the loop that makes the pipeline a cycle rather than a line. Partly manual today; an explicit improvement target.","nodes":[
+      {"t":"Close the loop","plain":"The outcome is scored against the goal that started it and written back to the mission board, so future planning learns from what actually shipped.","pill":"var(--violet)","d":"session_log + attestation → brain. Score deploy outcome vs originating goal. (arc partly manual — improvement target.)","tags":["continuous"]}
+    ]}
+  ]
+}

internal/atlas/build_test.go

@@ -0,0 +1,71 @@
+package atlas_test
+
+import (
+	"reflect"
+	"testing"
+
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
+)
+
+func TestBuild_OverlaysCIStageNodesFromWorkflow(t *testing.T) {
+	atlasJSON := []byte(`{
+	  "substrate": [{"n":"koala","k":"gpu"}],
+	  "stages": [
+	    {"no":"STAGE 06","title":"PR → CI","generate":"ci-jobs","nodes":[]},
+	    {"no":"STAGE 01","title":"TELOS","nodes":[{"t":"Intention substrate"}]}
+	  ]}`)
+	workflow := []byte("jobs:\n  check:\n    a: 1\n  build:\n    b: 2\n  deploy:\n    c: 3\n")
+
+	a, err := atlas.Build(atlasJSON, workflow)
+	if err != nil {
+		t.Fatalf("Build: %v", err)
+	}
+
+	// CI stage nodes generated from the workflow's jobs, in order.
+	var titles []string
+	for _, n := range a.Stages[0].Nodes {
+		titles = append(titles, n.Title)
+	}
+	if want := []string{"check", "build", "deploy"}; !reflect.DeepEqual(titles, want) {
+		t.Fatalf("CI stage nodes = %v, want %v", titles, want)
+	}
+
+	// Non-generated stage is untouched.
+	if a.Stages[1].Nodes[0].Title != "Intention substrate" {
+		t.Fatalf("authored stage was altered: %+v", a.Stages[1])
+	}
+
+	// Substrate preserved.
+	if len(a.Substrate) != 1 || a.Substrate[0].Name != "koala" {
+		t.Fatalf("substrate not preserved: %+v", a.Substrate)
+	}
+}
+
+func TestDefault_HasPlainLayerAndTransitions(t *testing.T) {
+	a, err := atlas.Build(atlas.DataJSON, []byte("jobs:\n  guard:\n    a: 1\n"))
+	if err != nil {
+		t.Fatalf("Build embedded atlas: %v", err)
+	}
+	for _, s := range a.Stages {
+		if s.PlainTitle == "" {
+			t.Fatalf("stage %s missing plain_title", s.No)
+		}
+		if s.TransLabel == "" {
+			t.Fatalf("stage %s missing trans_label", s.No)
+		}
+		if s.Generate == "ci-jobs" {
+			continue // nodes are generated live, no authored plain
+		}
+		for _, n := range s.Nodes {
+			if n.Plain == "" {
+				t.Fatalf("stage %s node %q missing plain", s.No, n.Title)
+			}
+		}
+	}
+}
+
+func TestBuild_ErrorsOnBadAtlasJSON(t *testing.T) {
+	if _, err := atlas.Build([]byte("{not json"), []byte("jobs:\n  x:\n    a: 1\n")); err == nil {
+		t.Fatal("expected error on bad atlas JSON, got nil")
+	}
+}

internal/atlas/cluster.go

@@ -0,0 +1,272 @@
+package atlas
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"sort"
+	"strconv"
+	"strings"
+)
+
+// HostsFromNodes parses a Kubernetes `/api/v1/nodes` list response into
+// substrate Host entries, so the atlas machines reflect the live cluster.
+func HostsFromNodes(nodesJSON []byte) ([]Host, error) {
+	var list struct {
+		Items []struct {
+			Metadata struct {
+				Name string `json:"name"`
+			} `json:"metadata"`
+			Status struct {
+				Capacity map[string]string `json:"capacity"`
+				NodeInfo struct {
+					Architecture   string `json:"architecture"`
+					KubeletVersion string `json:"kubeletVersion"`
+				} `json:"nodeInfo"`
+			} `json:"status"`
+		} `json:"items"`
+	}
+	if err := json.Unmarshal(nodesJSON, &list); err != nil {
+		return nil, fmt.Errorf("parse nodes: %w", err)
+	}
+
+	hosts := make([]Host, 0, len(list.Items))
+	for _, it := range list.Items {
+		var parts []string
+		if a := it.Status.NodeInfo.Architecture; a != "" {
+			parts = append(parts, a)
+		}
+		if c := it.Status.Capacity["cpu"]; c != "" {
+			parts = append(parts, c+" cpu")
+		}
+		if gi := memGi(it.Status.Capacity["memory"]); gi != "" {
+			parts = append(parts, gi+"Gi")
+		}
+		if g := it.Status.Capacity["nvidia.com/gpu"]; g != "" && g != "0" {
+			parts = append(parts, g+"× GPU")
+		}
+		if v := k3sVersion(it.Status.NodeInfo.KubeletVersion); v != "" {
+			parts = append(parts, v)
+		}
+		hosts = append(hosts, Host{Name: it.Metadata.Name, Spec: strings.Join(parts, " · ")})
+	}
+	return hosts, nil
+}
+
+// Flux is the reconciliation state of a Flux Kustomization.
+type Flux struct {
+	Ready    bool
+	Reason   string
+	Revision string
+}
+
+// FluxStatus parses a Flux Kustomization object into its reconcile state.
+func FluxStatus(kustJSON []byte) (Flux, error) {
+	var k struct {
+		Status struct {
+			Conditions []struct {
+				Type   string `json:"type"`
+				Status string `json:"status"`
+				Reason string `json:"reason"`
+			} `json:"conditions"`
+			LastAppliedRevision string `json:"lastAppliedRevision"`
+		} `json:"status"`
+	}
+	if err := json.Unmarshal(kustJSON, &k); err != nil {
+		return Flux{}, fmt.Errorf("parse kustomization: %w", err)
+	}
+	f := Flux{Revision: shortRev(k.Status.LastAppliedRevision)}
+	for _, c := range k.Status.Conditions {
+		if c.Type == "Ready" {
+			f.Ready = c.Status == "True"
+			f.Reason = c.Reason
+		}
+	}
+	return f, nil
+}
+
+// FluxNode renders the Flux reconcile state as a stage node.
+func FluxNode(f Flux) Node {
+	if f.Ready {
+		return Node{
+			Title: "⟳ Flux · reconciled · " + f.Revision,
+			Pill:  "var(--green)",
+			Tags:  []string{"live · k8s"},
+		}
+	}
+	return Node{
+		Title: "⟳ Flux · " + f.Reason,
+		Pill:  "var(--coral)",
+		Tags:  []string{"live · k8s"},
+	}
+}
+
+// shortRev turns a Flux revision "main@sha1:<full>" into "main@<short>".
+func shortRev(rev string) string {
+	at := strings.Index(rev, "@")
+	if at < 0 {
+		return rev
+	}
+	branch, sha := rev[:at], rev[at+1:]
+	if c := strings.LastIndex(sha, ":"); c >= 0 {
+		sha = sha[c+1:]
+	}
+	if len(sha) > 7 {
+		sha = sha[:7]
+	}
+	return branch + "@" + sha
+}
+
+// Deploy is the live state of a Kubernetes Deployment.
+type Deploy struct {
+	Image   string
+	Ready   int
+	Desired int
+}
+
+// Tag is the image tag (substring after the last ":").
+func (d Deploy) Tag() string {
+	if i := strings.LastIndex(d.Image, ":"); i >= 0 {
+		return d.Image[i+1:]
+	}
+	return d.Image
+}
+
+// DeployState parses a Kubernetes Deployment object into its live state.
+func DeployState(deployJSON []byte) (Deploy, error) {
+	var dep struct {
+		Spec struct {
+			Replicas int `json:"replicas"`
+			Template struct {
+				Spec struct {
+					Containers []struct {
+						Image string `json:"image"`
+					} `json:"containers"`
+				} `json:"spec"`
+			} `json:"template"`
+		} `json:"spec"`
+		Status struct {
+			ReadyReplicas int `json:"readyReplicas"`
+		} `json:"status"`
+	}
+	if err := json.Unmarshal(deployJSON, &dep); err != nil {
+		return Deploy{}, fmt.Errorf("parse deployment: %w", err)
+	}
+	d := Deploy{Ready: dep.Status.ReadyReplicas, Desired: dep.Spec.Replicas}
+	if len(dep.Spec.Template.Spec.Containers) > 0 {
+		d.Image = dep.Spec.Template.Spec.Containers[0].Image
+	}
+	return d, nil
+}
+
+// DeployNode renders the live deploy state as a stage node (green when the
+// rollout is fully ready, amber otherwise).
+func DeployNode(d Deploy) Node {
+	pill := "var(--amber)"
+	if d.Desired > 0 && d.Ready == d.Desired {
+		pill = "var(--green)"
+	}
+	img := d.Image
+	if i := strings.LastIndex(img, "/"); i >= 0 {
+		img = img[i+1:] // drop registry host
+	}
+	return Node{
+		Title: fmt.Sprintf("◆ deployed · %s · %d/%d ready", img, d.Ready, d.Desired),
+		Pill:  pill,
+		Tags:  []string{"live · k8s"},
+	}
+}
+
+// NamespaceSummary parses a Kubernetes `/api/v1/namespaces` list into a compact
+// "ns: a · b · c" line for the substrate, dropping system namespaces, sorting,
+// and capping the count (with "+N more" when it overflows).
+func NamespaceSummary(nsJSON []byte) (string, error) {
+	var list struct {
+		Items []struct {
+			Metadata struct {
+				Name string `json:"name"`
+			} `json:"metadata"`
+		} `json:"items"`
+	}
+	if err := json.Unmarshal(nsJSON, &list); err != nil {
+		return "", fmt.Errorf("parse namespaces: %w", err)
+	}
+
+	var names []string
+	for _, it := range list.Items {
+		n := it.Metadata.Name
+		if strings.HasPrefix(n, "kube-") || n == "default" || n == "flux-system" {
+			continue
+		}
+		names = append(names, n)
+	}
+	sort.Strings(names)
+
+	const maxShown = 12
+	more := 0
+	if len(names) > maxShown {
+		more = len(names) - maxShown
+		names = names[:maxShown]
+	}
+	if len(names) == 0 {
+		return "ns: (none)", nil
+	}
+	s := "ns: " + strings.Join(names, " · ")
+	if more > 0 {
+		s += " · +" + strconv.Itoa(more) + " more"
+	}
+	return s, nil
+}
+
+// MergeSubstrate overlays live node specs onto the authored substrate: an
+// authored host is replaced by the live node of the same name (fresh specs),
+// authored-only machines (non-cluster: iguana/flamingo/piguard) are kept, and
+// live-only nodes (new to the cluster) are appended.
+func MergeSubstrate(authored, live []Host) []Host {
+	liveByName := make(map[string]Host, len(live))
+	for _, h := range live {
+		liveByName[h.Name] = h
+	}
+	seen := make(map[string]bool, len(authored))
+	out := make([]Host, 0, len(authored)+len(live))
+	for _, a := range authored {
+		if l, ok := liveByName[a.Name]; ok {
+			out = append(out, l)
+		} else {
+			out = append(out, a)
+		}
+		seen[a.Name] = true
+	}
+	for _, l := range live {
+		if !seen[l.Name] {
+			out = append(out, l)
+		}
+	}
+	return out
+}
+
+// memGi converts a Kubernetes memory quantity in Ki (e.g. "67108864Ki") to a
+// rounded Gi string. Returns "" if unparseable.
+func memGi(ki string) string {
+	n, err := strconv.ParseFloat(strings.TrimSuffix(ki, "Ki"), 64)
+	if err != nil {
+		return ""
+	}
+	return strconv.Itoa(int(math.Round(n / 1048576)))
+}
+
+// k3sVersion trims a kubeletVersion's build metadata, labelling k3s builds.
+// "v1.31.4+k3s1" → "k3s v1.31.4"; "v1.31.4" → "v1.31.4".
+func k3sVersion(kubelet string) string {
+	if kubelet == "" {
+		return ""
+	}
+	if i := strings.Index(kubelet, "+"); i >= 0 {
+		ver, suffix := kubelet[:i], kubelet[i+1:]
+		if strings.Contains(suffix, "k3s") {
+			return "k3s " + ver
+		}
+		return ver
+	}
+	return kubelet
+}

internal/atlas/cluster_test.go

@@ -0,0 +1,165 @@
+package atlas_test
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
+)
+
+func TestHostsFromNodes_DerivesSpecFromNodeStatus(t *testing.T) {
+	nodes := []byte(`{"items":[
+	  {"metadata":{"name":"koala"},
+	   "status":{"capacity":{"cpu":"16","memory":"67108864Ki","nvidia.com/gpu":"1"},
+	             "nodeInfo":{"architecture":"amd64","kubeletVersion":"v1.31.4+k3s1"}}},
+	  {"metadata":{"name":"worker2"},
+	   "status":{"capacity":{"cpu":"8","memory":"33554432Ki"},
+	             "nodeInfo":{"architecture":"arm64","kubeletVersion":"v1.30.0+k3s1"}}}
+	]}`)
+
+	hosts, err := atlas.HostsFromNodes(nodes)
+	if err != nil {
+		t.Fatalf("HostsFromNodes: %v", err)
+	}
+
+	want := []atlas.Host{
+		{Name: "koala", Spec: "amd64 · 16 cpu · 64Gi · 1× GPU · k3s v1.31.4"},
+		{Name: "worker2", Spec: "arm64 · 8 cpu · 32Gi · k3s v1.30.0"},
+	}
+	if !reflect.DeepEqual(hosts, want) {
+		t.Fatalf("hosts = %+v\nwant   %+v", hosts, want)
+	}
+}
+
+func TestMergeSubstrate_LiveOverridesAuthoredKeepsRestAppendsNew(t *testing.T) {
+	authored := []atlas.Host{
+		{Name: "koala", Spec: "RTX 5070 · authored"},
+		{Name: "iguana", Spec: "M2 Ultra · Ollama / mlx"},
+		{Name: "piguard", Spec: "NGINX reverse-proxy · ntfy"},
+	}
+	live := []atlas.Host{
+		{Name: "koala", Spec: "amd64 · 16 cpu · 60Gi · 1× GPU · k3s v1.34.5"},
+		{Name: "worker9", Spec: "arm64 · 4 cpu"},
+	}
+
+	got := atlas.MergeSubstrate(authored, live)
+	want := []atlas.Host{
+		{Name: "koala", Spec: "amd64 · 16 cpu · 60Gi · 1× GPU · k3s v1.34.5"}, // live wins
+		{Name: "iguana", Spec: "M2 Ultra · Ollama / mlx"},                     // authored kept
+		{Name: "piguard", Spec: "NGINX reverse-proxy · ntfy"},                 // authored kept
+		{Name: "worker9", Spec: "arm64 · 4 cpu"},                              // live-only appended
+	}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("merge = %+v\nwant   %+v", got, want)
+	}
+}
+
+func TestMergeSubstrate_NoLiveReturnsAuthored(t *testing.T) {
+	authored := []atlas.Host{{Name: "koala", Spec: "authored"}}
+	got := atlas.MergeSubstrate(authored, nil)
+	if !reflect.DeepEqual(got, authored) {
+		t.Fatalf("merge with no live = %+v, want %+v", got, authored)
+	}
+}
+
+func TestNamespaceSummary_FiltersSystemSortsAndJoins(t *testing.T) {
+	ns := []byte(`{"items":[
+	  {"metadata":{"name":"gitea"}},
+	  {"metadata":{"name":"kube-system"}},
+	  {"metadata":{"name":"ai-stack"}},
+	  {"metadata":{"name":"default"}},
+	  {"metadata":{"name":"flux-system"}},
+	  {"metadata":{"name":"brain"}}
+	]}`)
+
+	got, err := atlas.NamespaceSummary(ns)
+	if err != nil {
+		t.Fatalf("NamespaceSummary: %v", err)
+	}
+	if want := "ns: ai-stack · brain · gitea"; got != want {
+		t.Fatalf("summary = %q, want %q", got, want)
+	}
+}
+
+func TestNamespaceSummary_CapsWithMore(t *testing.T) {
+	var items []string
+	for i := 0; i < 15; i++ {
+		items = append(items, `{"metadata":{"name":"app`+string(rune('a'+i))+`"}}`)
+	}
+	ns := []byte(`{"items":[` + strings.Join(items, ",") + `]}`)
+
+	got, err := atlas.NamespaceSummary(ns)
+	if err != nil {
+		t.Fatalf("NamespaceSummary: %v", err)
+	}
+	if !strings.HasSuffix(got, "· +3 more") {
+		t.Fatalf("expected cap suffix, got %q", got)
+	}
+}
+
+func TestFluxStatus_ParsesReadyReasonAndShortRevision(t *testing.T) {
+	k := []byte(`{"status":{
+	  "conditions":[{"type":"Ready","status":"True","reason":"ReconciliationSucceeded","message":"Applied revision: main@sha1:7a51d1d13944"}],
+	  "lastAppliedRevision":"main@sha1:7a51d1d13944c481b04fc434d9a2aca2a25632e4"}}`)
+
+	f, err := atlas.FluxStatus(k)
+	if err != nil {
+		t.Fatalf("FluxStatus: %v", err)
+	}
+	if !f.Ready || f.Reason != "ReconciliationSucceeded" || f.Revision != "main@7a51d1d" {
+		t.Fatalf("flux = %+v", f)
+	}
+}
+
+func TestFluxNode_GreenWhenReadyCoralWhenNot(t *testing.T) {
+	ready := atlas.FluxNode(atlas.Flux{Ready: true, Reason: "ReconciliationSucceeded", Revision: "main@7a51d1d"})
+	if ready.Pill != "var(--green)" {
+		t.Fatalf("ready pill = %q", ready.Pill)
+	}
+	if ready.Title != "⟳ Flux · reconciled · main@7a51d1d" {
+		t.Fatalf("title = %q", ready.Title)
+	}
+	notReady := atlas.FluxNode(atlas.Flux{Ready: false, Reason: "BuildFailed"})
+	if notReady.Pill != "var(--coral)" {
+		t.Fatalf("not-ready pill = %q", notReady.Pill)
+	}
+}
+
+func TestDeployState_ParsesImageAndReplicas(t *testing.T) {
+	dep := []byte(`{"spec":{"replicas":2,"template":{"spec":{"containers":[
+	  {"name":"cad-atlas","image":"localhost:5000/cad-atlas:3ff922a"}]}}},
+	  "status":{"readyReplicas":1,"replicas":2}}`)
+
+	d, err := atlas.DeployState(dep)
+	if err != nil {
+		t.Fatalf("DeployState: %v", err)
+	}
+	if d.Image != "localhost:5000/cad-atlas:3ff922a" || d.Ready != 1 || d.Desired != 2 {
+		t.Fatalf("deploy = %+v", d)
+	}
+	if d.Tag() != "3ff922a" {
+		t.Fatalf("tag = %q, want 3ff922a", d.Tag())
+	}
+}
+
+func TestDeployNode_GreenOnlyWhenFullyReady(t *testing.T) {
+	ready := atlas.DeployNode(atlas.Deploy{Image: "localhost:5000/cad-atlas:abc", Ready: 1, Desired: 1})
+	if ready.Pill != "var(--green)" {
+		t.Fatalf("ready pill = %q, want green", ready.Pill)
+	}
+	if ready.Title != "◆ deployed · cad-atlas:abc · 1/1 ready" {
+		t.Fatalf("title = %q", ready.Title)
+	}
+
+	rolling := atlas.DeployNode(atlas.Deploy{Image: "x/cad-atlas:def", Ready: 0, Desired: 1})
+	if rolling.Pill != "var(--amber)" {
+		t.Fatalf("rolling pill = %q, want amber", rolling.Pill)
+	}
+}
+
+func TestHostsFromNodes_ErrorsOnBadJSON(t *testing.T) {
+	if _, err := atlas.HostsFromNodes([]byte("{not json")); err == nil {
+		t.Fatal("expected error on bad JSON, got nil")
+	}
+}

internal/atlas/embed.go

@@ -0,0 +1,15 @@
+package atlas
+
+import _ "embed"
+
+// DataJSON is the authored atlas content (substrate + stage narrative).
+// Sourced/generated facts are overlaid onto it at Build time.
+//
+//go:embed atlas.json
+var DataJSON []byte
+
+// Default builds the atlas from the embedded authored data, overlaying the
+// given CI workflow (for the stage marked generate:ci-jobs).
+func Default(workflow []byte) (Atlas, error) {
+	return Build(DataJSON, workflow)
+}

internal/atlas/model.go

@@ -0,0 +1,77 @@
+package atlas
+
+import (
+	"encoding/json"
+	"fmt"
+)
+
+// Host is a substrate machine/entry (koala, iguana, …).
+type Host struct {
+	Name string `json:"n"`
+	Spec string `json:"k"`
+}
+
+// Node is a card within a stage. Plain is the jargon-free default text; Desc is
+// the technical detail shown on demand.
+type Node struct {
+	Title string   `json:"t"`
+	Plain string   `json:"plain,omitempty"`
+	Desc  string   `json:"d,omitempty"`
+	Pill  string   `json:"pill,omitempty"`
+	Cls   string   `json:"cls,omitempty"`
+	Tags  []string `json:"tags,omitempty"`
+	Risk  bool     `json:"risk,omitempty"`
+	Gate  bool     `json:"gate,omitempty"`
+}
+
+// Stage is one column of the pipeline. PlainTitle/Plain are the plain-language
+// default layer; Title/Path/Nodes[].Desc are the technical layer. TransLabel/
+// Trans annotate the outgoing transition (the arrow to the next stage): what
+// moves work forward and what must be true to advance. When Generate is set,
+// Nodes are derived from a live source at Build time.
+type Stage struct {
+	No         string `json:"no"`
+	Title      string `json:"title"`
+	PlainTitle string `json:"plain_title,omitempty"`
+	Plain      string `json:"plain,omitempty"`
+	Path       string `json:"path,omitempty"`
+	Cls        string `json:"cls,omitempty"`
+	Generate   string `json:"generate,omitempty"`
+	TransLabel string `json:"trans_label,omitempty"`
+	Trans      string `json:"trans,omitempty"`
+	Nodes      []Node `json:"nodes"`
+}
+
+// Atlas is the full data model the frontend renders.
+type Atlas struct {
+	Version   string   `json:"version,omitempty"`
+	Substrate []Host   `json:"substrate"`
+	NS        string   `json:"ns,omitempty"`
+	Timeline  []RunDot `json:"timeline,omitempty"`
+	Stages    []Stage  `json:"stages"`
+}
+
+// Build unmarshals the authored atlas JSON and overlays generated facts from
+// real sources, so sourced parts can't drift. Currently: any stage marked
+// `"generate":"ci-jobs"` gets its Nodes replaced by the workflow's job list.
+func Build(atlasJSON, workflow []byte) (Atlas, error) {
+	var a Atlas
+	if err := json.Unmarshal(atlasJSON, &a); err != nil {
+		return Atlas{}, fmt.Errorf("parse atlas data: %w", err)
+	}
+	for i := range a.Stages {
+		if a.Stages[i].Generate != "ci-jobs" {
+			continue
+		}
+		jobs, err := JobsFromWorkflow(workflow)
+		if err != nil {
+			return Atlas{}, fmt.Errorf("stage %s: %w", a.Stages[i].No, err)
+		}
+		nodes := make([]Node, 0, len(jobs))
+		for _, j := range jobs {
+			nodes = append(nodes, Node{Title: j})
+		}
+		a.Stages[i].Nodes = nodes
+	}
+	return a, nil
+}

internal/atlas/runs.go

@@ -0,0 +1,153 @@
+package atlas
+
+import (
+	"encoding/json"
+	"fmt"
+)
+
+// Job is one job within a workflow run (a Gitea Actions "task").
+type Job struct {
+	Name       string
+	Status     string
+	Conclusion string
+}
+
+// State is the effective outcome: conclusion if set, else status.
+func (j Job) State() string {
+	if j.Conclusion != "" {
+		return j.Conclusion
+	}
+	return j.Status
+}
+
+// RunSummary is the newest workflow run and its per-job outcomes.
+type RunSummary struct {
+	Number int
+	SHA    string
+	Title  string
+	Jobs   []Job
+}
+
+// State aggregates the jobs: failure if any failed, running if any not yet
+// succeeded, else success.
+func (s RunSummary) State() string {
+	return aggregateState(s.Jobs)
+}
+
+// RunDot is one run's aggregate outcome for the recent-runs timeline.
+type RunDot struct {
+	Number int    `json:"number"`
+	State  string `json:"state"`
+}
+
+// RecentRuns parses a Gitea `/actions/tasks` response (per-job, newest first)
+// into up to n most-recent runs with their aggregate outcome, newest first.
+func RecentRuns(tasksJSON []byte, n int) ([]RunDot, error) {
+	var resp struct {
+		Tasks []struct {
+			RunNumber  int    `json:"run_number"`
+			Status     string `json:"status"`
+			Conclusion string `json:"conclusion"`
+		} `json:"workflow_runs"`
+	}
+	if err := json.Unmarshal(tasksJSON, &resp); err != nil {
+		return nil, fmt.Errorf("parse tasks: %w", err)
+	}
+	var order []int
+	jobsByRun := map[int][]Job{}
+	for _, t := range resp.Tasks {
+		if _, seen := jobsByRun[t.RunNumber]; !seen {
+			order = append(order, t.RunNumber)
+		}
+		jobsByRun[t.RunNumber] = append(jobsByRun[t.RunNumber], Job{Status: t.Status, Conclusion: t.Conclusion})
+	}
+	dots := make([]RunDot, 0, n)
+	for _, rn := range order {
+		if len(dots) >= n {
+			break
+		}
+		dots = append(dots, RunDot{Number: rn, State: aggregateState(jobsByRun[rn])})
+	}
+	return dots, nil
+}
+
+// aggregateState folds per-job outcomes into a run outcome.
+func aggregateState(jobs []Job) string {
+	pending := false
+	for _, j := range jobs {
+		switch j.State() {
+		case "failure", "cancelled", "error":
+			return "failure"
+		case "success", "skipped": // completed OK — skipped (e.g. deploy on a tag push) doesn't block
+		default:
+			pending = true // running / in_progress / waiting / queued / unknown
+		}
+	}
+	if pending {
+		return "running"
+	}
+	return "success"
+}
+
+// LatestRunJobs parses a Gitea `/actions/tasks` response (per-job entries,
+// newest first) and returns the newest run with its jobs in pipeline order.
+func LatestRunJobs(tasksJSON []byte) (RunSummary, error) {
+	var resp struct {
+		Tasks []struct {
+			RunNumber  int    `json:"run_number"`
+			Name       string `json:"name"`
+			Status     string `json:"status"`
+			Conclusion string `json:"conclusion"`
+			SHA        string `json:"head_sha"`
+			Title      string `json:"display_title"`
+		} `json:"workflow_runs"`
+	}
+	if err := json.Unmarshal(tasksJSON, &resp); err != nil {
+		return RunSummary{}, fmt.Errorf("parse tasks: %w", err)
+	}
+	if len(resp.Tasks) == 0 {
+		return RunSummary{}, fmt.Errorf("no workflow tasks")
+	}
+
+	latest := resp.Tasks[0]
+	s := RunSummary{Number: latest.RunNumber, SHA: latest.SHA, Title: latest.Title}
+	for _, t := range resp.Tasks {
+		if t.RunNumber == latest.RunNumber {
+			s.Jobs = append(s.Jobs, Job{Name: t.Name, Status: t.Status, Conclusion: t.Conclusion})
+		}
+	}
+	// Gitea lists newest (last-finished) first; reverse to pipeline order.
+	for i, j := 0, len(s.Jobs)-1; i < j; i, j = i+1, j-1 {
+		s.Jobs[i], s.Jobs[j] = s.Jobs[j], s.Jobs[i]
+	}
+	return s, nil
+}
+
+// RunNodes renders a run as stage nodes: a summary node followed by one node
+// per job, each coloured by outcome.
+func RunNodes(s RunSummary) []Node {
+	sha := s.SHA
+	if len(sha) > 7 {
+		sha = sha[:7]
+	}
+	nodes := []Node{{
+		Title: fmt.Sprintf("▶ run #%d · %s", s.Number, s.State()),
+		Desc:  s.Title,
+		Pill:  statePill(s.State()),
+		Tags:  []string{"live · Gitea Actions", sha},
+	}}
+	for _, j := range s.Jobs {
+		nodes = append(nodes, Node{Title: j.Name, Pill: statePill(j.State())})
+	}
+	return nodes
+}
+
+func statePill(state string) string {
+	switch state {
+	case "success":
+		return "var(--green)"
+	case "failure", "cancelled", "error":
+		return "var(--coral)"
+	}
+	return "var(--amber)"
+}

internal/atlas/runs_test.go

@@ -0,0 +1,106 @@
+package atlas_test
+
+import (
+	"testing"
+
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
+)
+
+func TestLatestRunJobs_GroupsNewestRunReversedToPipelineOrder(t *testing.T) {
+	// Gitea returns tasks newest-first (deploy finished last → appears first).
+	tasks := []byte(`{"workflow_runs":[
+	  {"run_number":28,"name":"Deploy via GitOps","status":"success","head_sha":"633ba153f26abc","display_title":"feat: x"},
+	  {"run_number":28,"name":"Build & Import","status":"success"},
+	  {"run_number":28,"name":"Lint / Test / Vet","status":"success"},
+	  {"run_number":27,"name":"Deploy via GitOps","status":"failure"}
+	]}`)
+
+	s, err := atlas.LatestRunJobs(tasks)
+	if err != nil {
+		t.Fatalf("LatestRunJobs: %v", err)
+	}
+	if s.Number != 28 || s.SHA != "633ba153f26abc" || s.Title != "feat: x" {
+		t.Fatalf("summary = %+v", s)
+	}
+	// only run 28's jobs, reversed to pipeline order (Lint → Build → Deploy)
+	got := []string{}
+	for _, j := range s.Jobs {
+		got = append(got, j.Name)
+	}
+	want := []string{"Lint / Test / Vet", "Build & Import", "Deploy via GitOps"}
+	if len(got) != 3 || got[0] != want[0] || got[2] != want[2] {
+		t.Fatalf("jobs = %v, want %v", got, want)
+	}
+	if s.State() != "success" {
+		t.Fatalf("state = %q, want success", s.State())
+	}
+}
+
+func TestRunSummary_StateFailsIfAnyJobFailed(t *testing.T) {
+	s := atlas.RunSummary{Jobs: []atlas.Job{
+		{Status: "success"}, {Status: "completed", Conclusion: "failure"},
+	}}
+	if s.State() != "failure" {
+		t.Fatalf("state = %q, want failure", s.State())
+	}
+}
+
+func TestRunNodes_SummaryThenPerJobColoured(t *testing.T) {
+	s := atlas.RunSummary{Number: 28, SHA: "633ba153f26", Title: "feat: x", Jobs: []atlas.Job{
+		{Name: "Lint / Test / Vet", Status: "success"},
+		{Name: "Deploy via GitOps", Status: "failure"},
+	}}
+	nodes := atlas.RunNodes(s)
+	if len(nodes) != 3 {
+		t.Fatalf("want 3 nodes (summary + 2 jobs), got %d", len(nodes))
+	}
+	if nodes[0].Title != "▶ run #28 · failure" || nodes[0].Pill != "var(--coral)" {
+		t.Fatalf("summary node = %+v", nodes[0])
+	}
+	if nodes[1].Title != "Lint / Test / Vet" || nodes[1].Pill != "var(--green)" {
+		t.Fatalf("job node 1 = %+v", nodes[1])
+	}
+	if nodes[2].Pill != "var(--coral)" {
+		t.Fatalf("failed job pill = %q", nodes[2].Pill)
+	}
+}
+
+func TestRecentRuns_GroupsRunsNewestFirstWithAggregateState(t *testing.T) {
+	tasks := []byte(`{"workflow_runs":[
+	  {"run_number":28,"name":"Deploy","status":"success"},
+	  {"run_number":28,"name":"Build","status":"success"},
+	  {"run_number":27,"name":"Deploy","status":"completed","conclusion":"failure"},
+	  {"run_number":26,"name":"Build","status":"running"},
+	  {"run_number":25,"name":"Deploy","status":"success"}
+	]}`)
+
+	dots, err := atlas.RecentRuns(tasks, 3)
+	if err != nil {
+		t.Fatalf("RecentRuns: %v", err)
+	}
+	if len(dots) != 3 {
+		t.Fatalf("want 3 dots (capped), got %d: %+v", len(dots), dots)
+	}
+	want := []atlas.RunDot{{28, "success"}, {27, "failure"}, {26, "running"}}
+	for i := range want {
+		if dots[i] != want[i] {
+			t.Fatalf("dot[%d] = %+v, want %+v", i, dots[i], want[i])
+		}
+	}
+}
+
+func TestRunState_SkippedJobsCountAsOK(t *testing.T) {
+	// A tag-push run skips the deploy job; the run still succeeded.
+	s := atlas.RunSummary{Jobs: []atlas.Job{
+		{Status: "skipped"}, {Status: "success"}, {Status: "success"},
+	}}
+	if s.State() != "success" {
+		t.Fatalf("skipped+success run state = %q, want success", s.State())
+	}
+}
+
+func TestLatestRunJobs_ErrorsWhenEmpty(t *testing.T) {
+	if _, err := atlas.LatestRunJobs([]byte(`{"workflow_runs":[]}`)); err == nil {
+		t.Fatal("expected error on empty, got nil")
+	}
+}

internal/atlas/workflow.go

@@ -0,0 +1,48 @@
+// Package atlas builds the CAD Atlas data model, deriving parts from real
+// sources (the repo's own CI workflow, infra manifests, the live cluster)
+// so the visualization can't drift from reality.
+package atlas
+
+import (
+	"fmt"
+
+	"gopkg.in/yaml.v3"
+)
+
+// JobsFromWorkflow extracts the job names, in document order, from a Gitea
+// Actions / GitHub Actions workflow YAML. Used to generate the CI/CD stage of
+// the atlas from the pipeline that actually runs, rather than hand-authoring it.
+func JobsFromWorkflow(workflow []byte) ([]string, error) {
+	var doc yaml.Node
+	if err := yaml.Unmarshal(workflow, &doc); err != nil {
+		return nil, fmt.Errorf("parse workflow: %w", err)
+	}
+	if len(doc.Content) == 0 || doc.Content[0].Kind != yaml.MappingNode {
+		return nil, fmt.Errorf("workflow: expected a top-level mapping")
+	}
+	jobs := mappingValue(doc.Content[0], "jobs")
+	if jobs == nil {
+		return nil, fmt.Errorf("workflow: no jobs block")
+	}
+	if jobs.Kind != yaml.MappingNode {
+		return nil, fmt.Errorf("workflow: jobs is not a mapping")
+	}
+
+	// A mapping node stores keys and values as alternating Content entries;
+	// keys are the even indices, in document order.
+	names := make([]string, 0, len(jobs.Content)/2)
+	for i := 0; i+1 < len(jobs.Content); i += 2 {
+		names = append(names, jobs.Content[i].Value)
+	}
+	return names, nil
+}
+
+// mappingValue returns the value node for key in a YAML mapping node, or nil.
+func mappingValue(m *yaml.Node, key string) *yaml.Node {
+	for i := 0; i+1 < len(m.Content); i += 2 {
+		if m.Content[i].Value == key {
+			return m.Content[i+1]
+		}
+	}
+	return nil
+}

internal/atlas/workflow_test.go

@@ -0,0 +1,45 @@
+package atlas_test
+
+import (
+	"reflect"
+	"testing"
+
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
+)
+
+func TestJobsFromWorkflow_ReturnsJobNamesInDocumentOrder(t *testing.T) {
+	workflow := []byte(`
+name: CD
+"on":
+  push:
+    branches: [main]
+jobs:
+  guard:
+    runs-on: self-hosted
+  check:
+    needs: guard
+    runs-on: self-hosted
+  build:
+    needs: [guard, check]
+    runs-on: self-hosted
+  deploy:
+    needs: [guard, build]
+    runs-on: self-hosted
+`)
+
+	jobs, err := atlas.JobsFromWorkflow(workflow)
+	if err != nil {
+		t.Fatalf("JobsFromWorkflow: %v", err)
+	}
+
+	want := []string{"guard", "check", "build", "deploy"}
+	if !reflect.DeepEqual(jobs, want) {
+		t.Fatalf("jobs = %v, want %v", jobs, want)
+	}
+}
+
+func TestJobsFromWorkflow_ErrorsOnMalformedYAML(t *testing.T) {
+	if _, err := atlas.JobsFromWorkflow([]byte("jobs: [this is: not valid")); err == nil {
+		t.Fatal("expected error on malformed YAML, got nil")
+	}
+}

internal/cluster/client.go

@@ -0,0 +1,80 @@
+// Package cluster reads live k3s state from inside a pod via the Kubernetes
+// API, using the mounted ServiceAccount credentials. No client-go: the queries
+// are read-only and few, so stdlib net/http keeps the dependency surface small.
+package cluster
+
+import (
+	"crypto/tls"
+	"crypto/x509"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"time"
+)
+
+const (
+	tokenPath = "/var/run/secrets/kubernetes.io/serviceaccount/token" //nolint:gosec // well-known in-cluster path, not a secret literal
+	caPath    = "/var/run/secrets/kubernetes.io/serviceaccount/ca.crt"
+)
+
+// Nodes returns the raw /api/v1/nodes JSON from the in-cluster API server.
+func Nodes() ([]byte, error) { return get("/api/v1/nodes") }
+
+// Namespaces returns the raw /api/v1/namespaces JSON from the in-cluster API server.
+func Namespaces() ([]byte, error) { return get("/api/v1/namespaces") }
+
+// Deployment returns the raw JSON for the cad-atlas Deployment (its own live state).
+func Deployment() ([]byte, error) {
+	return get("/apis/apps/v1/namespaces/cad-atlas/deployments/cad-atlas")
+}
+
+// FluxKustomization returns the raw JSON for the Flux `apps` Kustomization that
+// reconciles this repo's manifests.
+func FluxKustomization() ([]byte, error) {
+	return get("/apis/kustomize.toolkit.fluxcd.io/v1/namespaces/flux-system/kustomizations/apps")
+}
+
+func get(path string) ([]byte, error) {
+	host, port := os.Getenv("KUBERNETES_SERVICE_HOST"), os.Getenv("KUBERNETES_SERVICE_PORT")
+	if host == "" || port == "" {
+		return nil, fmt.Errorf("not in-cluster: KUBERNETES_SERVICE_HOST unset")
+	}
+	token, err := os.ReadFile(tokenPath)
+	if err != nil {
+		return nil, fmt.Errorf("read sa token: %w", err)
+	}
+	ca, err := os.ReadFile(caPath)
+	if err != nil {
+		return nil, fmt.Errorf("read ca: %w", err)
+	}
+	pool := x509.NewCertPool()
+	if !pool.AppendCertsFromPEM(ca) {
+		return nil, fmt.Errorf("invalid cluster CA cert")
+	}
+	client := &http.Client{
+		Timeout: 5 * time.Second,
+		Transport: &http.Transport{
+			TLSClientConfig: &tls.Config{RootCAs: pool, MinVersion: tls.VersionTLS12},
+		},
+	}
+	req, err := http.NewRequest(http.MethodGet, "https://"+host+":"+port+path, nil)
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Authorization", "Bearer "+string(token))
+
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	body, err := io.ReadAll(io.LimitReader(resp.Body, 8<<20))
+	if err != nil {
+		return nil, err
+	}
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("k8s API GET %s: %s", path, resp.Status)
+	}
+	return body, nil
+}

internal/gitea/client.go

@@ -0,0 +1,39 @@
+// Package gitea reads the repo's own Gitea Actions run history from the
+// in-cluster Gitea service (public read — no token), so the atlas can show the
+// pipeline's live executions.
+package gitea
+
+import (
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"time"
+)
+
+// base is the in-cluster Gitea service by default; override with GITEA_BASE.
+func base() string {
+	if b := os.Getenv("GITEA_BASE"); b != "" {
+		return b
+	}
+	return "http://gitea-http.gitea.svc.cluster.local:3000"
+}
+
+// Runs returns the raw /actions/tasks JSON for mathias/cad-atlas (newest first).
+func Runs() ([]byte, error) {
+	url := base() + "/api/v1/repos/mathias/cad-atlas/actions/tasks?limit=50"
+	client := &http.Client{Timeout: 5 * time.Second}
+	resp, err := client.Get(url) //nolint:noctx // short-lived, timeout on the client
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	body, err := io.ReadAll(io.LimitReader(resp.Body, 4<<20))
+	if err != nil {
+		return nil, err
+	}
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("gitea runs: %s", resp.Status)
+	}
+	return body, nil
+}

internal/version/version.go

@@ -0,0 +1,6 @@
+// Package version holds the build version, injected at build time via
+// -ldflags "-X .../internal/version.Value=$(git describe --tags --always)".
+package version
+
+// Value is the build version. Defaults to "dev" for local/un-injected builds.
+var Value = "dev"

internal/web/handler.go

@@ -1,20 +1,28 @@
 package web
 
 import (
-	"context"
 	_ "embed"
+	"encoding/json"
 	"net/http"
+	"sync"
+	"time"
+
+	cadatlas "git.d-ma.be/mathias/cad-atlas"
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
+	"git.d-ma.be/mathias/cad-atlas/internal/cluster"
+	"git.d-ma.be/mathias/cad-atlas/internal/gitea"
+	"git.d-ma.be/mathias/cad-atlas/internal/version"
 )
 
-// atlasHTML is the Phase-A static hero visualization. Phase C replaces this
+// atlasHTML is the Phase-A/B static shell. It fetches /api/atlas.json at load
-// self-contained file with a Templ view hydrated from live CAD trace data
+// and renders from that data (no inline arrays), so the content is sourced.
-// (assessor-loop ledger, session_log, Gitea run API, Flux events).
 //
 //go:embed static/cad-atlas.html
 var atlasHTML []byte
 
-// NewHandler serves the CAD Atlas. Root ("/") returns the static atlas;
+// NewHandler serves the CAD Atlas: the shell at "/", and the sourced data at
-// /api/hello is a leftover template probe kept until Phase C wires real endpoints.
+// "/api/atlas.json" (authored data + CI stage generated from the real cd.yml +
+// substrate overlaid from the live cluster when reachable).
 func NewHandler() http.Handler {
 	mux := http.NewServeMux()
 	mux.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
@@ -25,8 +33,106 @@ func NewHandler() http.Handler {
 		w.Header().Set("Content-Type", "text/html; charset=utf-8")
 		_, _ = w.Write(atlasHTML)
 	})
-	mux.HandleFunc("/api/hello", func(w http.ResponseWriter, r *http.Request) {
+	mux.HandleFunc("/api/atlas.json", func(w http.ResponseWriter, r *http.Request) {
-		_ = Hello("world").Render(context.Background(), w)
+		a, err := atlas.Default(cadatlas.CDWorkflow)
+		if err != nil {
+			http.Error(w, "atlas build failed", http.StatusInternalServerError)
+			return
+		}
+		a.Version = version.Value
+		ld := liveOverlay()
+		if len(ld.hosts) > 0 {
+			a.Substrate = atlas.MergeSubstrate(a.Substrate, ld.hosts)
+		}
+		if ld.ns != "" {
+			a.NS = "Tailscale mesh · " + ld.ns
+		}
+		a.Timeline = ld.timeline
+		if ld.run != nil {
+			nodes := atlas.RunNodes(*ld.run)
+			for i := range a.Stages {
+				if a.Stages[i].Generate == "ci-jobs" {
+					a.Stages[i].Nodes = nodes
+				}
+			}
+		}
+		if ld.deploy != nil || ld.flux != nil {
+			var live []atlas.Node
+			if ld.deploy != nil {
+				live = append(live, atlas.DeployNode(*ld.deploy))
+			}
+			if ld.flux != nil {
+				live = append(live, atlas.FluxNode(*ld.flux))
+			}
+			for i := range a.Stages {
+				if a.Stages[i].Generate == "deploy-state" {
+					a.Stages[i].Nodes = append(live, a.Stages[i].Nodes...)
+				}
+			}
+		}
+		w.Header().Set("Content-Type", "application/json; charset=utf-8")
+		_ = json.NewEncoder(w).Encode(a)
 	})
 	return mux
 }
+
+// liveOverlay holds cluster-sourced substrate facts (node specs + namespace line).
+type liveOverlayData struct {
+	hosts    []atlas.Host
+	ns       string
+	run      *atlas.RunSummary
+	deploy   *atlas.Deploy
+	flux     *atlas.Flux
+	timeline []atlas.RunDot
+}
+
+// live cache: query the cluster at most once per TTL; fall back to the authored
+// substrate/ns (zero values) whenever the cluster is unreachable.
+var (
+	liveMu    sync.Mutex
+	liveCache liveOverlayData
+	liveAt    time.Time
+)
+
+const liveTTL = 30 * time.Second
+
+func liveOverlay() liveOverlayData {
+	liveMu.Lock()
+	defer liveMu.Unlock()
+	if !liveAt.IsZero() && time.Since(liveAt) < liveTTL {
+		return liveCache
+	}
+	liveAt = time.Now()
+
+	var d liveOverlayData
+	if raw, err := cluster.Nodes(); err == nil {
+		if hosts, err := atlas.HostsFromNodes(raw); err == nil {
+			d.hosts = hosts
+		}
+	}
+	if raw, err := cluster.Namespaces(); err == nil {
+		if s, err := atlas.NamespaceSummary(raw); err == nil {
+			d.ns = s
+		}
+	}
+	if raw, err := gitea.Runs(); err == nil {
+		if r, err := atlas.LatestRunJobs(raw); err == nil {
+			d.run = &r
+		}
+		if dots, err := atlas.RecentRuns(raw, 12); err == nil {
+			d.timeline = dots
+		}
+	}
+	if raw, err := cluster.Deployment(); err == nil {
+		if dep, err := atlas.DeployState(raw); err == nil {
+			d.deploy = &dep
+		}
+	}
+	if raw, err := cluster.FluxKustomization(); err == nil {
+		if f, err := atlas.FluxStatus(raw); err == nil {
+			d.flux = &f
+		}
+	}
+	liveCache = d
+	return d
+}

internal/web/handler_test.go

@@ -1,10 +1,13 @@
 package web
 
 import (
+	"encoding/json"
 	"net/http"
 	"net/http/httptest"
 	"strings"
 	"testing"
+
+	"git.d-ma.be/mathias/cad-atlas/internal/atlas"
 )
 
 func TestRootServesAtlas(t *testing.T) {
@@ -27,6 +30,43 @@ func TestRootServesAtlas(t *testing.T) {
 	}
 }
 
+func TestAtlasJSON_ServesAtlasWithGeneratedCIStage(t *testing.T) {
+	srv := httptest.NewServer(NewHandler())
+	defer srv.Close()
+
+	resp, err := http.Get(srv.URL + "/api/atlas.json")
+	if err != nil {
+		t.Fatalf("GET /api/atlas.json: %v", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("status = %d, want 200", resp.StatusCode)
+	}
+	var a atlas.Atlas
+	if err := json.NewDecoder(resp.Body).Decode(&a); err != nil {
+		t.Fatalf("decode atlas: %v", err)
+	}
+	if len(a.Stages) == 0 {
+		t.Fatal("atlas has no stages")
+	}
+
+	// The generate:ci-jobs stage must be populated from the real cd.yml jobs.
+	got := map[string]bool{}
+	for _, s := range a.Stages {
+		if s.No == "STAGE 06" {
+			for _, n := range s.Nodes {
+				got[n.Title] = true
+			}
+		}
+	}
+	for _, want := range []string{"guard", "check", "build", "deploy"} {
+		if !got[want] {
+			t.Fatalf("CI stage missing generated job %q (got %v)", want, got)
+		}
+	}
+}
+
 func TestUnknownPath404(t *testing.T) {
 	srv := httptest.NewServer(NewHandler())
 	defer srv.Close()

internal/web/static/cad-atlas.html

@@ -52,6 +52,22 @@
   .host .k{color:var(--dim);font-size:11px}
   .host.mesh{border-style:dashed;color:var(--dim)}
 
+  .tl{display:none;gap:5px;align-items:center;flex-wrap:wrap;
+    padding:8px 22px;border-bottom:1px solid var(--line);background:var(--panel)}
+  .tl .lbl{color:var(--dim);font-size:11px;letter-spacing:1.5px;margin-right:4px}
+  .rundot{width:22px;height:16px;border-radius:4px;border:1px solid rgba(0,0,0,.35);
+    display:inline-flex;align-items:center;justify-content:center;
+    font-size:9px;color:#08121f;font-weight:600}
+
+  /* progressive disclosure */
+  .tech-sub{color:var(--dim);font-size:11px;font-family:ui-monospace,SFMono-Regular,monospace;margin:1px 0 5px}
+  .plain-what{color:var(--ink);font-size:12.5px;line-height:1.45;margin-bottom:2px;opacity:.92}
+  .translabel{fill:var(--mono);font-size:10px;font-family:ui-monospace,SFMono-Regular,monospace}
+  .translabel-gate{fill:var(--gold);font-weight:700}
+  .stagehead{min-height:64px}
+  body.plain .stagehead{min-height:188px}
+  .stage .node:first-of-type{margin-top:24px}
+
   .scroll{overflow-x:auto;padding:24px 22px 20px}
   .track{position:relative;display:flex;align-items:flex-start;min-width:max-content}
   svg.spine{position:absolute;left:0;top:0;z-index:0;pointer-events:none;overflow:visible}
@@ -114,14 +130,16 @@
 <body>
 <header>
   <h1><b>CAD</b> Atlas · From Signal to Pod</h1>
-  <span class="sub mono">one human gate · everything up- and downstream is agents · <em>v0.3 static snapshot (→ live in Phase C)</em></span>
+  <span class="sub mono">one human gate · everything up- and downstream is agents · <em id="ver">dev</em></span>
   <div class="controls">
+    <button id="mode" class="on">View · <span id="modeState">Plain</span></button>
     <button id="replay"><span class="dot"></span> Replay</button>
     <button id="slowmo">Slow-mo · <span id="slowState">off</span></button>
   </div>
 </header>
 
 <div class="substrate" id="substrate"><span class="lbl mono">SUBSTRATE</span></div>
+<div class="tl mono" id="timeline"></div>
 
 <div class="scroll">
   <div class="track" id="track">
@@ -137,6 +155,7 @@
             stroke-dasharray="5 5" opacity=".7"></path>
       <text id="loopLbl" fill="#9b8cff" font-size="11"
             font-family="ui-monospace,monospace" opacity=".85"></text>
+      <g id="translabels"></g>
     </svg>
     <div class="pulse" id="pulse"></div>
   </div>
@@ -145,93 +164,79 @@
 <footer class="mono">
   CAD → CI → CD · intent→specify→dispatch · build→test→validate · deploy→ship.
   Dashed violet = feedback bus (stage 08 → TELOS: deploy outcome scored vs originating goal).
-  Data: static inventory from <code>brain</code> (2026-07-19). Phase C swaps these arrays for live reads of
+  Data served from <code>/api/atlas.json</code> (authored <code>atlas.json</code> + CI stage from the live <code>cd.yml</code> + substrate from the live cluster nodes).
-  <code>assessor-loop</code> ledger · <code>session_log</code> · Gitea run API · Flux events.
+  Phase C: <code>Gitea run API</code> live (latest run on the CI stage). Pending: <code>assessor-loop</code> ledger · <code>session_log</code> · per-job status · Flux events.
 </footer>
 
 <script>
-const SUBSTRATE=[
+// Single source of truth: /api/atlas.json (Go app — authored atlas.json + CI
-  {n:"koala",  k:"RTX 5070 · k3s control-plane · Gitea · LiteLLM :30401 · llama-swap :31234 · searxng"},
+// stage from the live cd.yml + substrate from the live cluster). These start
-  {n:"iguana", k:"M2 Ultra · Ollama / mlx"},
+// empty and are filled by init()'s fetch; on failure the page shows an error
-  {n:"flamingo",k:"daily driver · ~/dev"},
+// banner rather than stale inline data.
-  {n:"piguard",k:"NGINX reverse-proxy · ntfy"},
+let SUBSTRATE=[], NS="", STAGES=[], TIMELINE=[];
-];
+let MODE = localStorage.getItem('atlas-mode') || 'plain';   // 'plain' | 'technical'
-const NS="Tailscale mesh · ns: ai-stack · supervisor(→brain) · gitea-mcp · infra-mcp · council";
-
-const STAGES=[
- {no:"STAGE 00",cls:"",title:"Signals",path:"→ mathias/signals",nodes:[
-   {t:"Applied AI Radar",d:"Daily Tier-1 + weekly Tier-2 deep pass. Verified-primary bar (paper/benchmark/code/named-lab).",tags:["cron · daily/weekly","→ signals #1–26+"]},
-   {t:"Manual capture",d:"claude.ai strategic drop · brain capture tool.",tags:["ad-hoc"]},
-   {t:"Aspirational surfaces",pill:"var(--dim)",d:"Telegram / voice / URL → inbox. NOT built.",tags:["gap"]},
- ]},
- {no:"STAGE 01",cls:"telos",title:"TELOS",path:"wiki/telos/",nodes:[
-   {t:"Intention substrate",pill:"var(--violet)",d:"Mission · goals · problems · strategies · status. Every downstream item traces to a goal.",tags:["brain_query wing=telos"]},
- ]},
- {no:"STAGE 02",cls:"",title:"Strategic session",path:"claude.ai frontier + brain MCP",nodes:[
-   {t:"Design · ADRs · specs",d:"Human + frontier model. ISC acceptance criteria written here.",tags:["Define / converge"]},
-   {t:"🏛️ LLM Council",cls:"council",pill:"var(--violet)",d:"fan-out → anonymous cross-review → chairman synth. glm-4.7-flash · qwen36-35b · gemma4-31b (chair).",tags:["hard strategic Q","chat.d-ma.be"]},
-   {t:"Autoresearch Council",cls:"council",pill:"var(--violet)",d:"Sibling pipe — ratifies research before the gate.",tags:["proposed: → standalone svc"]},
- ]},
- {no:"STAGE 03",cls:"",title:"Spec → Gitea issue",path:"agent-ready contract",nodes:[
-   {t:"Contract enforced",d:"Binary ISC · declared risk tier · reg-risk assessment · no open human deps.",tags:["LOW / MED / HIGH"]},
-   {t:"Admission controller",d:"Ed25519-sign issue body at creation (#36). Verify sig + PR alignment at infra boundary.",tags:["chain of custody"]},
-   {t:"⚖️ var-go Oath",cls:"oath",pill:"var(--gold)",d:"Acceptance contract embedded in the issue as a ```var fenced block. Exactly one — zero/multiple fail closed. Prose → typed steps; failures anchored to byte spans.",tags:["swedsl · var-go","defined here → enforced @06"]},
- ]},
- {no:"STAGE 04",cls:"gate",title:"Human dispatch gate",path:"the only checkpoint",nodes:[
-   {t:"Human triggers execution",cls:"gateway",pill:"var(--amber)",d:"Ratify proposed-plan + risk tier, then dispatch.",gate:true},
-   {t:"Session-Dispatch bridge",cls:"bridge",pill:"var(--blue)",d:"claude.ai MCP → gitea:workflow_run_trigger → cad-dispatch.yml → agentsquad. The final design→execution bridge.",tags:["workflow_dispatch"]},
- ]},
- {no:"STAGE 05",cls:"exec",title:"Execute · agentsquad",path:"koala · cmd/agentsquad-serve",nodes:[
-   {t:"Task API",pill:"var(--coral)",d:"POST /tasks → job id · GET /tasks/{id}. taskqueue + serve (v0.12+).",tags:["single agentsquad.yaml"]},
-   {t:"Executor + reviewer loop",cls:"win",pill:"var(--coral)",d:"ADK Go + LiteLLM. Frontier models (local qwen spirals). Reviewer on distinct tier — echo-chamber prevention.",risk:true},
-   {t:"dma-cli · routing + scope",cls:"bridge",pill:"var(--blue)",d:"Harness-config arm: routes agents to the right LLM backend. Three-layer scope policy + confirmation gate = CAD guardrail.",tags:["backend routing","scope guardrail"]},
-   {t:"assessor-loop ledger",d:"Attestation ledger (audit trail) + brain session_log on completion.",tags:["audit package"]},
- ]},
- {no:"STAGE 06",cls:"",title:"PR → CI",path:"Gitea Actions",nodes:[
-   {t:"PR + label",d:"Gitea PR · agent-done / agent-blocked label.",tags:[]},
-   {t:"Mechanical gate",d:"go test · vet · lint · govulncheck. ISC verified mechanically.",tags:["green = proceed"]},
-   {t:"⚖️ var-go/oath gate",cls:"oath",pill:"var(--gold)",d:"cmd/vargo-gate runs in CI → posts commit status context=var-go/oath. Authoritative FLOOR: failed Oath blocks regardless of reviewer approval (#55 anti-rubber-stamp).",tags:["branch-protection req","enforces @03 Oath"]},
- ]},
- {no:"STAGE 07",cls:"cd",title:"CD → pod",path:"Flux GitOps → k3s",nodes:[
-   {t:"Deploy on green",pill:"var(--green)",d:"Flux reconciles image → k3s pod on koala. Push ≠ deploy: bump tag in mathias/infra.",tags:["ntfy on deploy"]},
- ]},
- {no:"STAGE 08",cls:"telos",title:"Loop back",path:"→ TELOS (feedback bus)",nodes:[
-   {t:"Close the loop",pill:"var(--violet)",d:"session_log + attestation → brain. Score deploy outcome vs originating goal. (arc partly manual — improvement target.)",tags:["continuous"]},
- ]},
-];
-
-const sub=document.getElementById('substrate');
-SUBSTRATE.forEach(h=>{const el=document.createElement('div');el.className='host';
-  el.innerHTML=`<b>${h.n}</b><span class="k mono">${h.k}</span>`;sub.appendChild(el);});
-const mesh=document.createElement('div');mesh.className='host mesh mono';mesh.textContent=NS;sub.appendChild(mesh);
 
 const track=document.getElementById('track');
-const stageEls=[];
+let stageEls=[];
-STAGES.forEach(s=>{
+function renderAtlas(){
-  const st=document.createElement('div');st.className='stage '+s.cls;
+  const sub=document.getElementById('substrate');
-  let h=`<div class="no mono">${s.no}</div><h2>${s.title}</h2><div class="path mono">${s.path||''}</div>`;
+  sub.querySelectorAll('.host').forEach(el=>el.remove());
-  s.nodes.forEach(n=>{
+  SUBSTRATE.forEach(h=>{const el=document.createElement('div');el.className='host';
+    el.innerHTML=`<b>${h.n}</b><span class="k mono">${h.k}</span>`;sub.appendChild(el);});
+  const mesh=document.createElement('div');mesh.className='host mesh mono';mesh.textContent=NS;sub.appendChild(mesh);
+  stageEls=[];
+  track.querySelectorAll('.stage').forEach(el=>el.remove());
+  const plain = MODE==='plain';
+  STAGES.forEach(s=>{
+  const st=document.createElement('div');st.className='stage '+(s.cls||'');
+  const head = plain
+    ? `<div class="no mono">${s.no}</div><h2>${s.plain_title||s.title}</h2>`+
+      `<div class="tech-sub">${s.title}</div>`+(s.plain?`<div class="plain-what">${s.plain}</div>`:'')
+    : `<div class="no mono">${s.no}</div><h2>${s.title}</h2><div class="path mono">${s.path||''}</div>`;
+  let h = `<div class="stagehead">${head}</div>`;
+  (s.nodes||[]).forEach(n=>{
     const pill=n.pill?`<span class="pill" style="background:${n.pill}"></span>`:'';
-    let inner=`<div class="t">${pill}${n.t}</div><div class="d">${n.d}</div>`;
+    const body = plain ? (n.plain||n.d||'') : (n.d||'');
+    let inner=`<div class="t">${pill}${n.t}</div>`+(body?`<div class="d">${body}</div>`:'');
+    if(!plain){
       if(n.tags&&n.tags.length)inner+=n.tags.map(t=>`<span class="tag mono">${t}</span>`).join('');
       if(n.risk)inner+=`<div class="risk mono"><span class="lo">LOW · auto</span><span class="md">MED · ntfy gate</span><span class="hi">HIGH · blocked</span></div>`;
+    }
     if(n.gate)inner+=`<div class="gatebtns mono"><div class="g ok">✓ approve</div><div class="g no">✕ reject</div></div>`;
     h+=`<div class="node ${n.cls||''}">${inner}</div>`;
   });
   st.innerHTML=h;track.appendChild(st);stageEls.push(st);
-});
+  });
+}
+
+function renderTimeline(){
+  const tl=document.getElementById('timeline');
+  if(!TIMELINE.length){tl.style.display='none';return;}
+  tl.innerHTML='<span class="lbl">RECENT RUNS</span>';
+  TIMELINE.forEach(d=>{
+    const c=d.state==='success'?'var(--green)':(d.state==='failure'?'var(--coral)':'var(--amber)');
+    const el=document.createElement('span');
+    el.className='rundot';el.style.background=c;
+    el.title='run #'+d.number+' · '+d.state;el.textContent=d.number;
+    tl.appendChild(el);
+  });
+  tl.style.display='flex';
+}
 
 /* ---- geometry ---- */
 const spine=document.getElementById('spine'), spinePath=document.getElementById('spinePath'),
       loopPath=document.getElementById('loopPath'), loopLbl=document.getElementById('loopLbl'),
       pulse=document.getElementById('pulse');
-const RAILY=70;
+let RAILY=70;
 let cs=[],loopY=0,spineLen=0,loopLen=0,slow=false,raf=null,t0=null,mobile=false;
 
 function build(){
   mobile=window.matchMedia('(max-width:820px)').matches;
   if(mobile)return;
   cs=stageEls.map(s=>s.offsetLeft+s.offsetWidth/2);
+  // spine sits in the band between the (uniform) stage headers and the first node
+  const firstTops=stageEls.map(s=>{const n=s.querySelector('.node');return n?s.offsetTop+n.offsetTop:120;});
+  RAILY=Math.max(66, Math.min(...firstTops)-16);
   const maxBottom=Math.max(...stageEls.map(s=>s.offsetTop+s.offsetHeight));
   loopY=maxBottom+40;
   spine.setAttribute('width',track.scrollWidth);
@@ -241,7 +246,22 @@ function build(){
   const lastX=cs[cs.length-1], telosX=cs[1];
   loopPath.setAttribute('d',`M ${lastX} ${RAILY} L ${lastX} ${loopY} L ${telosX} ${loopY} L ${telosX} ${RAILY}`);
   loopLbl.setAttribute('x',(telosX+lastX)/2-90);loopLbl.setAttribute('y',loopY-8);
-  loopLbl.textContent='feedback bus · outcome → goal';
+  loopLbl.textContent=(STAGES[8]&&STAGES[8].trans_label?STAGES[8].trans_label+' · ':'')+'feedback bus';
+  // transition labels on the spine — the "what must be true to advance" story.
+  // Always visible (default layer). Gate hops (04→05 human, 06→07 CI) in gold.
+  const tg=document.getElementById('translabels'); tg.innerHTML='';
+  for(let i=0;i<cs.length-1;i++){
+    const s=STAGES[i]; if(!s||!s.trans_label) continue;
+    const gate = (i===4||i===6);
+    const t=document.createElementNS('http://www.w3.org/2000/svg','text');
+    t.setAttribute('x',(cs[i]+cs[i+1])/2);t.setAttribute('y',RAILY-9);
+    t.setAttribute('text-anchor','middle');
+    t.setAttribute('class',gate?'translabel translabel-gate':'translabel');
+    t.textContent=(gate?'▸ ':'')+s.trans_label;
+    const ttl=document.createElementNS('http://www.w3.org/2000/svg','title');
+    ttl.textContent=s.trans||''; t.appendChild(ttl);
+    tg.appendChild(t);
+  }
   spineLen=spinePath.getTotalLength();loopLen=loopPath.getTotalLength();
 }
 function run(ts){
@@ -264,9 +284,41 @@ function replay(){cancelAnimationFrame(raf);t0=null;build();if(!mobile)raf=reque
 document.getElementById('replay').onclick=replay;
 document.getElementById('slowmo').onclick=e=>{slow=!slow;e.currentTarget.classList.toggle('on',slow);
   document.getElementById('slowState').textContent=slow?'on':'off';replay();};
+function applyMode(){
+  document.getElementById('modeState').textContent = MODE==='plain'?'Plain':'Technical';
+  document.getElementById('mode').classList.toggle('on', MODE==='plain');
+  document.body.classList.toggle('plain', MODE==='plain');
+}
+document.getElementById('mode').onclick=()=>{
+  MODE = MODE==='plain' ? 'technical' : 'plain';
+  localStorage.setItem('atlas-mode',MODE);
+  applyMode(); renderAtlas(); replay();
+};
 window.addEventListener('resize',()=>{clearTimeout(window._r);window._r=setTimeout(replay,150);});
-window.addEventListener('load',replay);
+async function init(){
-build();
+  try{
+    const r=await fetch('/api/atlas.json');
+    if(!r.ok) throw new Error('atlas.json '+r.status);
+    const data=await r.json();
+    if(Array.isArray(data.substrate)) SUBSTRATE=data.substrate;
+    if(typeof data.ns==='string') NS=data.ns;
+    if(Array.isArray(data.stages)) STAGES=data.stages;
+    if(Array.isArray(data.timeline)) TIMELINE=data.timeline;
+    if(data.version) document.getElementById('ver').textContent=data.version;
+  }catch(e){
+    console.error('atlas: failed to load /api/atlas.json —',e);
+    document.getElementById('track').insertAdjacentHTML('beforeend',
+      '<div class="stage"><div class="no mono">ERROR</div>'+
+      '<h2 style="color:var(--coral)">Data unavailable</h2>'+
+      '<div class="path mono">/api/atlas.json failed to load</div></div>');
+    return;
+  }
+  applyMode();
+  renderAtlas();
+  renderTimeline();
+  replay();
+}
+window.addEventListener('load',init);
 </script>
 </body>
 </html>