diff --git a/AGENTS.md b/AGENTS.md
index f6c7ca9..7bd10f9 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -46,6 +46,7 @@ In [docs/decisions.md](docs/decisions.md):
 - A change to what exists is a major; a minor only adds
 - Seeded output is promised within one version
 - An error is a contract by what it names, not its bytes
+- A load reports every mistake at once
 - Formats are checked after all data is loaded, so their errors name the category
 - Raising the lowest supported Go is a major
 - The format check runs on the latest Go only
diff --git a/README.md b/README.md
index 0bb4537..80b7546 100644
--- a/README.md
+++ b/README.md
@@ -291,6 +291,11 @@ ok, err := fejkdata.IsTemplate(arg)       // an inline template by its shape, el
 | `WithDataFS(fsys)` | layer an `fs.FS`, such as your own `embed.FS` |
 | `WithoutShippedData()` | load only what you give |
 
+`New` refuses a mistake in the data and `NewTemplate` one in the template, so on a
+loaded generator `Fake` fails only for a path that names nothing or could name two
+things, `FakeStruct` only for a non-struct argument or a type its tags do not describe,
+with the same error every call, and `Template.Fake` cannot fail at all.
+
 A `*Record` carries its columns via `Columns()` — each a `Column` of `Name`,
 `DataType`, rendered `Value` and `Null` — and serializes them with `JSON()`
 (one object), `CSVHeader()`/`CSVLine()`, or `SQLInsert(table)` — the shapes the
@@ -837,10 +842,7 @@ the Development section below, and who ships a register the five above then draw
    Where two writers could expect different outputs, the template is a load error
    naming a spelling for each. One spelling per result; the wrong one is a load error
    naming the right one.
-5. **Every mistake is a load error** — `New` rejects the data and `NewTemplate`
-   the inline template; on a loaded generator `Fake` fails only for a path that
-   names nothing or could name two things, `FakeStruct` only for a non-struct argument or a type its tags do not
-   describe, with the same error every call, and `Template.Fake` cannot fail at all.
+5. **Errors are up front and easy to understand.**
 6. **Zero to a value in one command** — `go install`, then `fejkdata sv_SE.person`:
    no checkout, no flag. Flags are GNU-form (`--seed 42`, `-n 3`) in any position;
    the first custom template needs no escape and no option.
diff --git a/docs/decisions.md b/docs/decisions.md
index 7cdcb04..bc66c4d 100644
--- a/docs/decisions.md
+++ b/docs/decisions.md
@@ -148,8 +148,16 @@ this repo's own are.
 
 2026-09-16, Lilleman auf Larv.
 
-A script branches on the exit code and reads the named path or spelling, so those hold;
-wording improves in a minor.
+Goal 5. A script branches on the exit code and reads the named path or spelling, so
+those hold; wording improves in a minor.
+
+## A load reports every mistake at once
+
+2026-09-30, Lilleman auf Larv. Goal 5. An author learns everything wrong with the data
+in one load, not one mistake per run: `New` and `NewTemplate` report every mistake in one
+error, ordered by the path each names. A mistake following only from another, such as a
+reference into a category that did not compile, is left out, so each one names something
+to fix. Valid while a load checks the whole set before its first render.
 
 ## Formats are checked after all data is loaded, so their errors name the category
 
diff --git a/todo.md b/todo.md
index f748ed9..d73406a 100644
--- a/todo.md
+++ b/todo.md
@@ -208,6 +208,9 @@ by a `data-import/` script, as README goal 12 asks.
 
 ### Library and CLI
 
+- Report every mistake a load finds in one error, as the decision "A load reports every
+  mistake at once" states: `New` and `NewTemplate` stop at the first, so data holding
+  two mistakes takes two runs to fix. Split it into sub-items before starting.
 - Rename `newRand` to `newGeneratorState` and `Generator.rand` to `state`: both name
   half of a `generatorState`, hiding its `{seq()}` counters, and `New`'s local `rng`
   shadows the `rng` interface.
@@ -297,10 +300,6 @@ by a `data-import/` script, as README goal 12 asks.
   together, where the fence counts them alternatives and compares neither. Have the error
   name the path spelling that holds, `{/geo.SE.locality.name}|{/geo.SE.municipality.name}`,
   as the README's rule on a rejected spelling asks.
-- Decide, in "An error is a contract by what it names", which of several mistakes a
-  load names, and link the entry to goal 5, which it serves and does not name. The
-  fence names the lowest by route and path, so a data edit can move which one is named;
-  proposed: fixed within a version, as seeded output is, and free to change in any release.
 - Name the file a link error comes from beside its category path, `sub.x
   (/d/a/sub/x.json): …`: with several `--data-path` layers the author has to work out
   which directory won, while a parse error on the same file names it.