diff --git a/soleprint/atlas2/docgen/emitters/notebook.py b/soleprint/atlas2/docgen/emitters/notebook.py
index 75abe4b..32df47f 100644
--- a/soleprint/atlas2/docgen/emitters/notebook.py
+++ b/soleprint/atlas2/docgen/emitters/notebook.py
@@ -146,20 +146,41 @@ def _call_cell(step: dict) -> str:
     """The generated call. An overlay's `code` replaces this wholesale."""
     method, path = step.get("method", "GET"), step.get("path", "/")
     params = step.get("path_params") or []
+    if step.get("graphql"):
+        # One endpoint carries every operation, so the operation name is the
+        # thing worth showing, not the path.
+        variables = {f["name"]: f"<{f['name']}>" for f in (step.get("body_fields") or [])}
+        return (
+            f'QUERY = """{step.get("title", "query")} {{ ... }}"""   '
+            "# fill in the selection set\n"
+            + (f"VARIABLES = {variables!r}\n" if variables else "")
+            + f'show(call("POST", "{path}", body={{"query": QUERY'
+            + (", \"variables\": VARIABLES" if variables else "")
+            + '}))'
+        )
     lines = [f'{p.upper()} = "<{p}>"      # path parameter' for p in params]
     call_path = path
     for p in params:
         call_path = call_path.replace("{" + p + "}", f'" + str({p.upper()}) + "')
+    # Trim the empty concatenations a placeholder at either end leaves behind.
     expr = f'"{call_path}"' if params else f'"{path}"'
+    expr = expr.replace(' + ""', "").replace('"" + ', "")
+
+    # Parameters that were *always* sent are not optional in practice, whatever
+    # the spec calls them.
+    always = step.get("params_always") or []
+    if always:
+        lines.append("PARAMS = " + repr({p: f"<{p}>" for p in always}))
+    arg = ", params=PARAMS" if always else ""
 
     if step.get("body_fields"):
         lines.append("BODY = " + _example(step["body_fields"]))
         lines.append("")
-        lines.append(f'show(call("{method}", {expr}, body=BODY))')
+        lines.append(f'show(call("{method}", {expr}{arg}, body=BODY))')
     else:
         if lines:
             lines.append("")
-        lines.append(f'show(call("{method}", {expr}))')
+        lines.append(f'show(call("{method}", {expr}{arg}))')
     return "\n".join(lines)
 
 
@@ -175,6 +196,15 @@ def _call_md(step: dict) -> str:
         facts.append(f'accepts **{step["request_model"]}**')
     if step.get("status"):
         facts.append(f'expects `{step["status"]}`')
+    if step.get("statuses"):
+        # What really came back, which is usually more than the spec promises.
+        facts.append("seen: " + ", ".join(f'`{c}`' for c in step["statuses"]))
+    if step.get("calls"):
+        facts.append(f'called {step["calls"]}×')
+    if step.get("params_sometimes"):
+        facts.append("sometimes sends " + ", ".join(f'`{p}`' for p in step["params_sometimes"]))
+    if step.get("id_formats"):
+        facts.append("id as " + "/".join(step["id_formats"]))
     if facts:
         out += ["", " · ".join(facts)]
     if step.get("note"):
diff --git a/soleprint/atlas2/docgen/extractors/__main__.py b/soleprint/atlas2/docgen/extractors/__main__.py
index 4d38704..799c481 100644
--- a/soleprint/atlas2/docgen/extractors/__main__.py
+++ b/soleprint/atlas2/docgen/extractors/__main__.py
@@ -1,4 +1,4 @@
-"""    python3 -m docgen.extractors <db|openapi> [options]"""
+"""    python3 -m docgen.extractors <db|openapi|usage> [options]"""
 
 import sys
 
@@ -8,6 +8,9 @@ def main(argv=None):
     if argv and argv[0] == "openapi":
         from .openapi_main import main as run
         return run(argv[1:])
+    if argv and argv[0] == "usage":
+        from .usage_main import main as run
+        return run(argv[1:])
     if argv and argv[0] == "db":
         from .db_main import main as run
         return run(argv[1:])
diff --git a/soleprint/atlas2/docgen/extractors/usage.py b/soleprint/atlas2/docgen/extractors/usage.py
new file mode 100644
index 0000000..0e0bca3
--- /dev/null
+++ b/soleprint/atlas2/docgen/extractors/usage.py
@@ -0,0 +1,247 @@
+"""
+Recorded traffic -> IR. What callers actually do, rather than what exists.
+
+Reads a **HAR** file — the format browser devtools, mitmproxy, Charles and
+Insomnia all export. Standard, JSON, stdlib-parseable, and already sitting on
+most people's disk after ten minutes of using the thing they want documented.
+
+    python3 -m docgen.extractors usage --har session.har -o ir.json
+
+## Why this exists
+
+An OpenAPI document says what endpoints *are*. It does not say how to use them,
+and the gap is where the whole difficulty lives:
+
+- **The order.** Which call has to happen first. A spec is a set; usage is a
+  sequence, and the sequence is most of what a newcomer needs.
+- **Which parameters matter.** A spec lists twenty optional query parameters.
+  Traffic shows the two that are always sent.
+- **What a real payload looks like** — as opposed to a shape with every field
+  present and none of them meaning anything.
+- **The endpoints that are not in the document at all**, which for a GraphQL
+  endpoint beside a REST surface is the normal case rather than an oversight.
+- **Which responses actually happen.** A spec promises 200 and 404; traffic
+  shows the 422 that everyone hits.
+
+None of that is recoverable by reading harder. It is only in the traffic.
+
+## The two inferences, and their limits
+
+**Path templating.** `/pets/123` and `/pets/456` are one endpoint. Any segment
+that looks like an identifier — digits, a UUID, a long hex string — becomes
+`{id}`, *whatever its format*: a route taking a numeric id on one call and a
+UUID on the next is one route, and splitting it by format invents an endpoint
+that does not exist. This is still a guess — a genuine path segment that happens
+to be numeric gets templated wrongly — so `attrs.observed_paths` and
+`attrs.id_formats` keep what was actually seen beside it.
+
+**Sequence.** Consecutive calls become `follows` edges carrying how often that
+pair occurred. Consecutive is not *caused by*, and one session's order is not
+the only order — the weight is what separates a habit from an accident, and a
+single recording will not tell you which.
+
+Nothing here reads a response body's values beyond its shape, and no header is
+copied into the IR: a HAR is full of cookies and bearer tokens, and none of them
+belong in a document that gets committed.
+"""
+
+import json
+import re
+from collections import Counter
+from pathlib import Path
+from urllib.parse import parse_qs, urlsplit
+
+from ..ir import Graph, Meta
+
+UUID = re.compile(r"^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$", re.I)
+LONG_HEX = re.compile(r"^[0-9a-f]{16,}$", re.I)
+DIGITS = re.compile(r"^\d+$")
+
+# Never copied into the IR. A HAR carries live credentials, and a document that
+# gets committed must not.
+SECRET_HEADERS = {"authorization", "cookie", "set-cookie", "x-api-key", "proxy-authorization"}
+
+
+def _template(path: str) -> tuple[str, set[str]]:
+    """`/pets/123` -> `/pets/{id}`. Returns (templated, id formats seen).
+
+    Every identifier-looking segment becomes `{id}`, whatever its format. A
+    caller that passes a numeric id on one call and a UUID on the next is using
+    **one** route, and templating them to `{id}` and `{uuid}` splits it into two
+    endpoints that do not exist — which is worse than the imprecision it avoids.
+    The formats are recorded instead, so "this route takes both" stays visible.
+    """
+    out, formats = [], set()
+    for seg in path.split("/"):
+        if not seg:
+            out.append(seg)
+            continue
+        fmt = None
+        if DIGITS.match(seg):
+            fmt = "numeric"
+        elif UUID.match(seg):
+            fmt = "uuid"
+        elif LONG_HEX.match(seg):
+            fmt = "hash"
+        if fmt:
+            out.append("{id}")
+            formats.add(fmt)
+        else:
+            out.append(seg)
+    return "/".join(out), formats
+
+
+def _body(entry: dict) -> dict | None:
+    post = (entry.get("request") or {}).get("postData") or {}
+    text = post.get("text")
+    if not text:
+        return None
+    try:
+        parsed = json.loads(text)
+    except (json.JSONDecodeError, TypeError):
+        return None
+    return parsed if isinstance(parsed, dict) else None
+
+
+def _shape(value) -> str:
+    """The type of a value, never the value. A payload is full of real data."""
+    if isinstance(value, bool):
+        return "bool"
+    if isinstance(value, int):
+        return "int"
+    if isinstance(value, float):
+        return "float"
+    if isinstance(value, list):
+        return "list"
+    if isinstance(value, dict):
+        return "object"
+    if value is None:
+        return "null"
+    return "str"
+
+
+def _graphql(body: dict | None) -> tuple[str | None, str | None]:
+    """(operation name, operation type) if this is a GraphQL call."""
+    if not body or "query" not in body:
+        return None, None
+    query = body.get("query")
+    if not isinstance(query, str):
+        return None, None
+    name = body.get("operationName")
+    m = re.match(r"\s*(query|mutation|subscription)?\s*([A-Za-z_]\w*)?", query)
+    op_type = (m.group(1) if m else None) or "query"
+    return (name or (m.group(2) if m and m.group(2) else "anonymous")), op_type
+
+
+def extract(har_path, source: str = "usage") -> Graph:
+    """A HAR file -> IR of what was actually called."""
+    path = Path(har_path)
+    har = json.loads(path.read_text())
+    entries = (har.get("log") or {}).get("entries") or []
+    entries = sorted(entries, key=lambda e: e.get("startedDateTime", ""))
+
+    calls = []                                   # (key, kind, facts) in order
+    seen: dict[str, dict] = {}
+
+    for entry in entries:
+        req = entry.get("request") or {}
+        res = entry.get("response") or {}
+        method = (req.get("method") or "GET").upper()
+        url = req.get("url") or ""
+        split = urlsplit(url)
+        templated, id_formats = _template(split.path or "/")
+        body = _body(entry)
+
+        op, op_type = _graphql(body)
+        if op:
+            key = f"{op_type} {op}"
+            kind = "operation"
+        else:
+            key = f"{method} {templated}"
+            kind = "endpoint"
+
+        record = seen.setdefault(
+            key,
+            {
+                "kind": kind,
+                "calls": 0,
+                "statuses": Counter(),
+                "params": Counter(),
+                "body_fields": Counter(),
+                "field_types": {},
+                "paths": Counter(),
+                "method": method,
+                "path": templated,
+                "id_formats": set(),
+                "op_type": op_type,
+                "host": split.netloc,
+            },
+        )
+        record["calls"] += 1
+        record["id_formats"] |= id_formats
+        status = res.get("status")
+        if status:
+            record["statuses"][int(status)] += 1
+        if split.path:
+            record["paths"][split.path] += 1
+        for name in parse_qs(split.query or ""):
+            record["params"][name] += 1
+        if body and not op:
+            for name, value in body.items():
+                record["body_fields"][name] += 1
+                record["field_types"].setdefault(name, _shape(value))
+        if op and isinstance(body.get("variables"), dict):
+            for name, value in body["variables"].items():
+                record["body_fields"][name] += 1
+                record["field_types"].setdefault(name, _shape(value))
+
+        calls.append(key)
+
+    g = Graph(Meta(source=source, root=path.name))
+
+    order = {}
+    for i, key in enumerate(calls):
+        order.setdefault(key, i)
+
+    for key, r in sorted(seen.items(), key=lambda kv: order[kv[0]]):
+        n = r["calls"]
+        attrs = {
+            "calls": n,
+            "statuses": sorted(r["statuses"]),
+            "first_seen": order[key],
+        }
+        if r["kind"] == "endpoint":
+            attrs["method"] = r["method"]
+            attrs["path"] = r["path"]
+        else:
+            attrs["protocol"] = "graphql"
+            attrs["operation"] = r["op_type"]
+        if r["host"]:
+            attrs["host"] = r["host"]
+        # Always sent vs sometimes sent is the distinction a spec cannot make.
+        always = sorted(p for p, c in r["params"].items() if c == n)
+        sometimes = sorted(p for p, c in r["params"].items() if c < n)
+        if always:
+            attrs["params_always"] = always
+        if sometimes:
+            attrs["params_sometimes"] = sometimes
+        if r["body_fields"]:
+            attrs["body_fields"] = [
+                {"name": f, "type": r["field_types"].get(f, "str"),
+                 "always": r["body_fields"][f] == n}
+                for f in sorted(r["body_fields"])
+            ]
+        if r["id_formats"]:
+            # The templating is a guess; keep what was actually seen beside it.
+            attrs["id_formats"] = sorted(r["id_formats"])
+            attrs["observed_paths"] = [p for p, _ in r["paths"].most_common(5)]
+        g.node(key, r["kind"], key, attrs=attrs)
+
+    # The sequence. Consecutive is not caused-by, so the weight is the signal.
+    pairs = Counter(zip(calls, calls[1:]))
+    for (a, b), weight in pairs.items():
+        if a == b:
+            continue                      # a repeated call is polling, not a step
+        g.edge(a, b, "follows", attrs={"weight": weight})
+
+    return g
diff --git a/soleprint/atlas2/docgen/extractors/usage_main.py b/soleprint/atlas2/docgen/extractors/usage_main.py
new file mode 100644
index 0000000..f6aedd4
--- /dev/null
+++ b/soleprint/atlas2/docgen/extractors/usage_main.py
@@ -0,0 +1,32 @@
+"""    python3 -m docgen.extractors usage --har session.har [-o ir.json]"""
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+
+def main(argv=None):
+    p = argparse.ArgumentParser(prog="python3 -m docgen.extractors usage")
+    p.add_argument("--har", "-s", required=True, type=Path,
+                   help="A HAR recording, as devtools/mitmproxy/Charles export.")
+    p.add_argument("--output", "-o", type=Path)
+    args = p.parse_args(argv)
+    from .usage import extract
+
+    try:
+        ir = extract(args.har)
+    except (OSError, json.JSONDecodeError, KeyError) as e:
+        print(f"Error: could not read {args.har}: {e}", file=sys.stderr)
+        return 1
+    text = json.dumps(ir.to_dict(), indent=2) + "\n"
+    if args.output:
+        args.output.parent.mkdir(parents=True, exist_ok=True)
+        args.output.write_text(text)
+        eps = sum(1 for n in ir.nodes if n.kind == "endpoint")
+        ops = sum(1 for n in ir.nodes if n.kind == "operation")
+        print(f"{len(ir.nodes)} nodes ({eps} endpoints, {ops} graphql), "
+              f"{len(ir.edges)} sequence edges -> {args.output}")
+    else:
+        sys.stdout.write(text)
+    return 0
diff --git a/soleprint/atlas2/docgen/notebook/spec.py b/soleprint/atlas2/docgen/notebook/spec.py
index 04cf7e4..cb8c7e6 100644
--- a/soleprint/atlas2/docgen/notebook/spec.py
+++ b/soleprint/atlas2/docgen/notebook/spec.py
@@ -45,6 +45,7 @@ document today, and it keeps working unchanged when a usage extractor exists.
 """
 
 import json
+import re
 from pathlib import Path
 
 SPEC_VERSION = "1"
@@ -70,10 +71,18 @@ def from_ir(ir: dict, base_url: str = "https://api.example.invalid",
         if n["kind"] == "column" and n.get("parent"):
             fields_of.setdefault(n["parent"], []).append(n)
 
+    # Observed order wins over alphabetical. A usage recording knows which call
+    # comes first, and that sequence is most of what a newcomer needs — sorting
+    # it away would throw out the one thing a spec could not have told us.
+    def _order(n):
+        a = n.get("attrs") or {}
+        if "first_seen" in a:
+            return (0, a["first_seen"], "")
+        return (1, 0, f'{a.get("path", "")} {a.get("method", "")}')
+
     endpoints = sorted(
-        (n for n in ir["nodes"] if n["kind"] == "endpoint"),
-        key=lambda n: ((n.get("attrs") or {}).get("path", ""),
-                       (n.get("attrs") or {}).get("method", "")),
+        (n for n in ir["nodes"] if n["kind"] in ("endpoint", "operation")),
+        key=_order,
     )
 
     steps = [
@@ -90,23 +99,51 @@ def from_ir(ir: dict, base_url: str = "https://api.example.invalid",
 
     for n in endpoints:
         a = n.get("attrs") or {}
+
+        # A usage recording carries facts a spec cannot: which parameters are
+        # *always* sent, which status codes really happen, how often it is
+        # called. Where they exist they are better than anything generated from
+        # a schema, so they win.
+        body_fields = None
+        if a.get("body_fields"):
+            raw = a["body_fields"]
+            if raw and isinstance(raw[0], dict) and "always" in raw[0]:
+                body_fields = [f for f in raw if f.get("always")] or raw
+            else:
+                body_fields = raw
+        elif a.get("request_model"):
+            body_fields = [
+                {"name": f.get("label") or f["id"].rsplit(".", 1)[-1],
+                 **(f.get("attrs") or {})}
+                for f in fields_of.get(a["request_model"], [])
+            ] or None
+
         steps.append(
             _step(
                 n["id"], "call",
-                title=f'{a.get("method", "GET")} {a.get("path", "/")}',
+                title=n["id"] if n["kind"] == "operation"
+                else f'{a.get("method", "GET")} {a.get("path", "/")}',
                 summary=a.get("summary"),
-                method=a.get("method", "GET"),
-                path=a.get("path", "/"),
+                method=a.get("method", "POST" if n["kind"] == "operation" else "GET"),
+                path=a.get("path", "/graphql" if n["kind"] == "operation" else "/"),
                 status=a.get("status"),
                 request_model=a.get("request_model"),
                 response_model=a.get("response_model"),
                 returns_list=a.get("returns_list"),
-                path_params=a.get("path_params"),
-                body_fields=[
-                    {"name": f.get("label") or f["id"].rsplit(".", 1)[-1],
-                     **(f.get("attrs") or {})}
-                    for f in fields_of.get(a.get("request_model") or "", [])
-                ] or None,
+                # A usage-derived path already carries its placeholders; a
+                # spec-derived one lists them separately. Either way the cell
+                # needs a variable, not a literal `{id}` that would 404.
+                path_params=a.get("path_params")
+                or re.findall(r"\{(\w+)\}", a.get("path", "")) or None,
+                body_fields=body_fields,
+                # usage-only
+                params_always=a.get("params_always"),
+                params_sometimes=a.get("params_sometimes"),
+                statuses=a.get("statuses"),
+                calls=a.get("calls"),
+                id_formats=a.get("id_formats"),
+                observed_paths=a.get("observed_paths"),
+                graphql=(n["kind"] == "operation") or None,
             )
         )
 
diff --git a/soleprint/atlas2/docgen/selftest.py b/soleprint/atlas2/docgen/selftest.py
index 0764898..0725431 100644
--- a/soleprint/atlas2/docgen/selftest.py
+++ b/soleprint/atlas2/docgen/selftest.py
@@ -48,6 +48,7 @@ erd_mod = __import__(f"{PKG}.emitters.erd", fromlist=["*"])
 nb_mod = __import__(f"{PKG}.emitters.notebook", fromlist=["*"])
 spec_mod = __import__(f"{PKG}.notebook", fromlist=["*"])
 db_ex = __import__(f"{PKG}.extractors.db", fromlist=["*"])
+usage_ex = __import__(f"{PKG}.extractors.usage", fromlist=["*"])
 
 check_ir = ir_mod.check
 Style, StyleError = style_mod.Style, style_mod.StyleError
@@ -814,6 +815,114 @@ else:
     )
 
 
+# --------------------------------------------------------------------------
+print("\n8b. usage — what a spec cannot say")
+
+# A spec is a set; usage is a sequence. Everything asserted here is something no
+# amount of reading the OpenAPI document harder would recover.
+def _entry(t, method, url, status, body=None):
+    req = {"method": method, "url": url,
+           "headers": [{"name": "Authorization", "value": "Bearer LEAK-ME-IF-BROKEN"}]}
+    if body is not None:
+        req["postData"] = {"mimeType": "application/json", "text": json.dumps(body)}
+    return {"startedDateTime": f"2026-01-01T00:00:{t:02d}Z",
+            "request": req, "response": {"status": status, "content": {}}}
+
+
+HAR = {"log": {"version": "1.2", "entries": [
+    _entry(1, "POST", "https://x.test/auth", 200, {"user": "a", "secret": "b"}),
+    _entry(2, "GET", "https://x.test/pets?status=available&limit=20", 200),
+    _entry(3, "GET", "https://x.test/pets/1042", 200),
+    _entry(4, "GET", "https://x.test/pets/7f3e9b2a-1c4d-4a5b-9e8f-0a1b2c3d4e5f", 200),
+    _entry(5, "POST", "https://x.test/graphql", 200,
+           {"operationName": "PetWithTags", "query": "query PetWithTags($id: ID!) { pet }",
+            "variables": {"id": "1042"}}),
+    _entry(6, "POST", "https://x.test/pets", 422, {"name": "Rex"}),
+    _entry(7, "POST", "https://x.test/pets", 201, {"name": "Rex", "status": "available"}),
+    _entry(8, "GET", "https://x.test/pets?status=available&limit=20", 200),
+]}}
+har_path = ROOT.parent / "session.har"
+har_path.write_text(json.dumps(HAR))
+use = usage_ex.extract(har_path).to_dict()
+by = {n["id"]: n for n in use["nodes"]}
+raw_json = json.dumps(use)
+
+check("the usage IR validates", check_ir(use) == [], str(check_ir(use)[:2]))
+check(
+    "**no credential reaches the IR**",
+    "LEAK-ME-IF-BROKEN" not in raw_json and "Authorization" not in raw_json,
+    "a HAR is full of live tokens and a document gets committed",
+)
+check(
+    "paths are templated into one route",
+    "GET /pets/{id}" in by and "GET /pets/1042" not in by,
+)
+check(
+    "...and both id formats stay on that one route",
+    by["GET /pets/{id}"]["attrs"]["id_formats"] == ["numeric", "uuid"]
+    and by["GET /pets/{id}"]["attrs"]["calls"] == 2,
+    "splitting by format invents an endpoint that does not exist",
+)
+check(
+    "what is always sent is distinguished from what is sometimes",
+    by["GET /pets"]["attrs"]["params_always"] == ["limit", "status"],
+    "a spec calls these optional; traffic says otherwise",
+)
+check(
+    "the statuses that really happen are recorded",
+    by["POST /pets"]["attrs"]["statuses"] == [201, 422],
+    "the 422 everyone hits is not in the spec",
+)
+check(
+    "a body field seen once out of twice is not marked always",
+    [f["name"] for f in by["POST /pets"]["attrs"]["body_fields"] if f["always"]] == ["name"],
+)
+check(
+    "a GraphQL operation is found and named",
+    "query PetWithTags" in by and by["query PetWithTags"]["kind"] == "operation",
+    "one endpoint carries many operations; the path alone says nothing",
+)
+check(
+    "only the shape of a payload is kept, never a value",
+    '"a"' not in raw_json and '"b"' not in raw_json,
+)
+seq = {(e["source"], e["target"]) for e in use["edges"]}
+check(
+    "the sequence is recorded as edges",
+    ("POST /auth", "GET /pets") in seq and all(e["kind"] == "follows" for e in use["edges"]),
+    "this is the half a spec structurally cannot contain",
+)
+check(
+    "a repeated call is not a step in the sequence",
+    not [e for e in use["edges"] if e["source"] == e["target"]],
+)
+check(
+    "usage is a pipeline, and drawn as one",
+    ops_mod.classify(use)["kind"] == "pipeline",
+)
+
+# The payoff: a walkthrough in the order people actually call things.
+u_spec, _ = spec_mod.merge(spec_mod.from_ir(use), None)
+ids = [st["id"] for st in u_spec["steps"] if st["kind"] == "call"]
+check(
+    "the notebook keeps the observed order",
+    ids[0] == "POST /auth" and ids.index("GET /pets") < ids.index("POST /pets"),
+    f"got {ids}",
+)
+u_body = "".join("".join(c["source"]) for c in nb_mod.build(u_spec)["cells"])
+check(
+    "it sends the parameters that are always sent",
+    "PARAMS = " in u_body and "'status'" in u_body,
+)
+check(
+    "a templated path becomes a variable, not a literal",
+    'ID = "<id>"' in u_body and '"/pets/{id}"' not in u_body,
+    "a literal {id} in the URL would 404",
+)
+check("the GraphQL operation gets a query cell", "QUERY = " in u_body)
+har_path.unlink()
+
+
 # --------------------------------------------------------------------------
 print("\n9b. generated base + hand-written overlay")
 
diff --git a/soleprint/atlas2/docgen/style/lucid.json b/soleprint/atlas2/docgen/style/lucid.json
index e40bc11..69e7daa 100644
--- a/soleprint/atlas2/docgen/style/lucid.json
+++ b/soleprint/atlas2/docgen/style/lucid.json
@@ -172,6 +172,24 @@
       "text": "text",
       "bold": true,
       "note": "A task that waits on something outside the pipeline."
+    },
+    "endpoint": {
+      "shape": "box",
+      "rounded": true,
+      "fill": "surface-2",
+      "border": "station",
+      "text": "text",
+      "bold": true,
+      "note": "An HTTP route. From a spec, or from what was actually called."
+    },
+    "operation": {
+      "shape": "box",
+      "rounded": true,
+      "fill": "surface-2",
+      "border": "accent",
+      "text": "text",
+      "bold": true,
+      "note": "A GraphQL operation \u2014 usually one endpoint carrying many, which is why it is its own kind rather than a path."
     }
   },
   "groups": {
@@ -306,6 +324,25 @@
       "arrowhead": "normal",
       "arrowsize": 0.8,
       "text": "accent"
+    },
+    "follows": {
+      "color": "accent",
+      "arrowhead": "normal",
+      "arrowsize": 0.7,
+      "text": "accent",
+      "note": "Observed order. Consecutive, not caused-by; the weight is the signal."
+    },
+    "accepts": {
+      "color": "border-strong",
+      "arrowhead": "normal",
+      "arrowsize": 0.6,
+      "text": "text-dim"
+    },
+    "returns": {
+      "color": "station",
+      "arrowhead": "normal",
+      "arrowsize": 0.6,
+      "text": "text-dim"
     }
   },
   "domain_slots": {