diff --git a/NEWS.md b/NEWS.md
index 975720a..ea54c05 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -29,6 +29,21 @@ No computed value changes.
   `stability = "percent_change"`. The 1975 printing's pages could not be
   checked.
 
+## The handoff contract, stated for readers
+
+Asked for by nomologR before handoff schema 1 becomes a 1.0 promise. The
+schema and every handoff are unchanged.
+
+* `?content_handoff` has a new section, "Reading the decisions": take each
+  item's decision from `carried` and `status`, never re-derive it by
+  comparing `value` with `criterion` or by branching on the producer version.
+  The example is the nine-expert correction in 0.8.0, where an item's I-CVI
+  stayed at .778 while its decision changed. The section also restates that
+  `keying` of `NA` means unknown, never forward-worded.
+* `?content_handoff` says an item sits in at most one scale, the one its
+  `item_evidence$scale` names. A test holds the sort, rating, and congruence
+  workflows to it, and checks that each refuses an item with two targets.
+
 # contentvalidR 0.9.0
 
 Ninth public release. v0.9.0 is about how the package presents itself before
diff --git a/R/content_handoff.R b/R/content_handoff.R
index 42b852a..798f986 100644
--- a/R/content_handoff.R
+++ b/R/content_handoff.R
@@ -498,7 +498,10 @@
 #'   \item{`scales`}{named list mapping each construct to its carried items, or
 #'     `NULL` when the design has no construct mapping. Expert relevance and
 #'     essentiality rate a single item set with no construct column, so they
-#'     produce `NULL`. Membership is one to one.}
+#'     produce `NULL`, as does a Delphi study. An item belongs to at most one
+#'     scale, the one its `item_evidence$scale` names: every workflow that maps
+#'     items to constructs requires exactly one target per item, and stops
+#'     otherwise.}
 #'   \item{`item_evidence`}{data frame with one row per reviewed item: `item`,
 #'     `scale` (`NA` without a construct mapping), `carried`, `status`,
 #'     `recommendation`, `n_judges`, `rule`, and `round`. For a Delphi handoff
@@ -530,6 +533,29 @@
 #' Fields and columns added within schema version 1 are optional for a reader,
 #' which should check that they are present rather than assume it.
 #'
+#' @section Reading the decisions:
+#' Take each item's decision from the handoff, not from its statistics.
+#' `carried` says whether the item travels forward, and `status` says why: it
+#' is one of the four values `keep` accepts. `recommendation` words the same
+#' decision for a person, and like `rule` it is prose.
+#'
+#' Don't re-derive a decision by comparing `value` with `criterion`, and don't
+#' branch on `provenance$package_version`. A corrected rule can change a
+#' decision between releases while the statistics stay the same, and the
+#' handoff records the decision its producer made. contentvalidR 0.8.0 is the
+#' example. For an item that 7 of 9 experts rated relevant, the I-CVI is .778
+#' in both releases. 0.7.0 compared it with a rounded .78 and held the item
+#' back. 0.8.0 applies Lynn's (1986) 7 of 9, stores 7/9 as the criterion, and
+#' carries the item. There the value equals the criterion, so a recomputed
+#' `value >= criterion` would hang on floating-point rounding, where the
+#' package itself compares counts of experts. Read `carried`, which holds what
+#' each release decided.
+#'
+#' Keying is read the same way, from the field and not from a default:
+#' `keying` is `1` for a forward-worded item, `-1` for a reverse-worded one,
+#' and `NA` when nobody said, as described under "Instrument metadata". Treat
+#' `NA` as unknown, never as forward-worded.
+#'
 #' @section What version 1 freezes:
 #' Schema version 1 is frozen as of contentvalidR 0.7.0. Code that reads a
 #' handoff can rely on all of the following, in every release that reports
@@ -726,6 +752,11 @@
 #' empirical analysis tests, which is why the item set travels with its
 #' evidence rather than as a bare list of names.
 #'
+#' @references
+#' Lynn, M. R. (1986). Determination and quantification of content validity.
+#' *Nursing Research, 35*(6), 382–385.
+#' \doi{10.1097/00006199-198611000-00017}
+#'
 #' @param fit A fitted `contentvalid_sort`, `contentvalid_rating`,
 #'   `contentvalid_expert`, or `contentvalid_delphi` object.
 #' @param keep Statuses that travel forward, defaulting to `"Supported"`. Any of
diff --git a/ROADMAP.md b/ROADMAP.md
index 879d8c4..06cfc4d 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -810,8 +810,8 @@ reads no handoffs and takes no part in the joint 1.0.
         0.7.0.
   - [x] schema version 1 is frozen and becomes a compatibility promise (0.7.0).
   - [ ] both packages are on CRAN. Waiting on CRAN's review of 0.4.0, which
-        has been in the `newbies` queue since 2026-09-18, and on nomologR's
-        first submission.
+        has been in the `newbies` queue since 2026-09-18, and of nomologR
+        0.3.0, submitted on 2026-09-26.
   - [x] both public APIs are stable, under a written deprecation policy, as far
         as this package goes (0.7.0). `anova_content()`'s `posthoc_pass`
         column was removed in 0.8.0. `agreement_summary()`, the one function
@@ -820,6 +820,32 @@ reads no handoffs and takes no part in the joint 1.0.
   - [x] there is a joint walkthrough from content review to empirical
         validation (0.7.0).
   - [ ] the two releases go out on the same day, each linking the other.
+- The release-candidate checklist, proposed by nomologR and approved by the
+  maintainer on 2026-09-27. It adds to #53's criteria rather than replacing
+  them:
+  - [ ] a final fixture set from each release candidate, with a manifest of
+        md5 sums, that nomologR's tests pass before either package tags.
+  - [ ] the compatibility table in both READMEs: contentvalidR 1.x writes
+        handoff schema 1, and nomologR 1.x reads it. It goes in at the release
+        candidate, so neither README describes a 1.x before one exists.
+  - [ ] one handoff example shown in both READMEs, regenerated from
+        contentvalidR's release candidate.
+  - [ ] the joint walkthrough lives in one package, and the other links to
+        it. Recommended, and acceptable to nomologR: this package's
+        `vignette("one-item-set-both-stages")`, linking to nomologR's
+        guided-workflow article for the empirical side. The maintainer
+        decides.
+  - [ ] the release notes link each other.
+- Three handoff clarifications nomologR asked for before schema 1 becomes a
+  1.0 promise, approved on 2026-09-27. None changes the schema:
+  - [x] a fixture that records keying as checked with no item reversed
+        (`reverse_keyed = character(0)`, so `keying` is `1` throughout),
+        generated from v0.9.0.
+  - [x] `?content_handoff` says an item sits in at most one scale, and a test
+        holds every workflow to it.
+  - [x] `?content_handoff` says a reader takes each decision from `carried`
+        and `status`, never re-deriving it from the statistics or the
+        producer version.
 
 ***
 
diff --git a/man/content_handoff.Rd b/man/content_handoff.Rd
index 59e08bb..47fa7c2 100644
--- a/man/content_handoff.Rd
+++ b/man/content_handoff.Rd
@@ -68,7 +68,10 @@ consumer matches on \code{"cv_handoff"} and reads these fields:
 \item{\code{scales}}{named list mapping each construct to its carried items, or
 \code{NULL} when the design has no construct mapping. Expert relevance and
 essentiality rate a single item set with no construct column, so they
-produce \code{NULL}. Membership is one to one.}
+produce \code{NULL}, as does a Delphi study. An item belongs to at most one
+scale, the one its \code{item_evidence$scale} names: every workflow that maps
+items to constructs requires exactly one target per item, and stops
+otherwise.}
 \item{\code{item_evidence}}{data frame with one row per reviewed item: \code{item},
 \code{scale} (\code{NA} without a construct mapping), \code{carried}, \code{status},
 \code{recommendation}, \code{n_judges}, \code{rule}, and \code{round}. For a Delphi handoff
@@ -101,6 +104,31 @@ Fields and columns added within schema version 1 are optional for a reader,
 which should check that they are present rather than assume it.
 }
 
+\section{Reading the decisions}{
+
+Take each item's decision from the handoff, not from its statistics.
+\code{carried} says whether the item travels forward, and \code{status} says why: it
+is one of the four values \code{keep} accepts. \code{recommendation} words the same
+decision for a person, and like \code{rule} it is prose.
+
+Don't re-derive a decision by comparing \code{value} with \code{criterion}, and don't
+branch on \code{provenance$package_version}. A corrected rule can change a
+decision between releases while the statistics stay the same, and the
+handoff records the decision its producer made. contentvalidR 0.8.0 is the
+example. For an item that 7 of 9 experts rated relevant, the I-CVI is .778
+in both releases. 0.7.0 compared it with a rounded .78 and held the item
+back. 0.8.0 applies Lynn's (1986) 7 of 9, stores 7/9 as the criterion, and
+carries the item. There the value equals the criterion, so a recomputed
+\code{value >= criterion} would hang on floating-point rounding, where the
+package itself compares counts of experts. Read \code{carried}, which holds what
+each release decided.
+
+Keying is read the same way, from the field and not from a default:
+\code{keying} is \code{1} for a forward-worded item, \code{-1} for a reverse-worded one,
+and \code{NA} when nobody said, as described under "Instrument metadata". Treat
+\code{NA} as unknown, never as forward-worded.
+}
+
 \section{What version 1 freezes}{
 
 Schema version 1 is frozen as of contentvalidR 0.7.0. Code that reads a
@@ -334,6 +362,11 @@ handoff$item_statistics
 # Carry items flagged for review as well, when the study protocol says so.
 content_handoff(fit, keep = c("Supported", "Review"))$items
 }
+\references{
+Lynn, M. R. (1986). Determination and quantification of content validity.
+\emph{Nursing Research, 35}(6), 382–385.
+\doi{10.1097/00006199-198611000-00017}
+}
 \seealso{
 \code{\link[=as.data.frame.contentvalid_workflow]{as.data.frame.contentvalid_workflow()}} for the full results table,
 and \code{\link[=content_report]{content_report()}} for manuscript tables.
diff --git a/tests/testthat/test-content-handoff.R b/tests/testthat/test-content-handoff.R
index 82293be..bab6e79 100644
--- a/tests/testthat/test-content-handoff.R
+++ b/tests/testthat/test-content-handoff.R
@@ -155,6 +155,48 @@ test_that("statistics stack long with their criteria", {
   expect_setequal(unique(st$item), as.character(fit$results$item))
 })
 
+test_that("an item sits in at most one scale, the one its evidence row names", {
+  fits <- list(sort = sort_fit(), rating = rating_fit(),
+               congruence = congruence_fit())
+  for (workflow in names(fits)) {
+    h <- content_handoff(fits[[workflow]], keep = c("Supported", "Review"))
+    members <- unlist(h$scales, use.names = FALSE)
+    expect_equal(anyDuplicated(members), 0L, info = workflow)
+    carried <- h$item_evidence[h$item_evidence$carried, , drop = FALSE]
+    expect_true(nrow(carried) > 0, info = workflow)
+    for (i in seq_len(nrow(carried))) {
+      expect_true(carried$item[i] %in% h$scales[[carried$scale[i]]],
+                  info = paste(workflow, carried$item[i]))
+    }
+  }
+
+  # A handoff cannot place an item in two scales, because every workflow that
+  # maps items to constructs refuses an item with two targets.
+  sort_dat <- data.frame(
+    item = rep(c("A1", "B1"), each = 4), rater = rep(1:4, 2),
+    target_construct = c("A", "A", "A", "B", rep("B", 4)),
+    assigned_construct = c(rep("A", 4), rep("B", 4)),
+    stringsAsFactors = FALSE
+  )
+  expect_error(sort_validity(sort_dat), "exactly one target construct")
+
+  d <- expand.grid(item = c("A1", "B1"), rater = 1:4, construct = c("A", "B"),
+                   stringsAsFactors = FALSE)
+  d$target_construct <- ifelse(d$item == "B1", "B", "A")
+  d$target_construct[d$item == "A1" & d$rater == 1] <- "B"
+  d$rating <- ifelse(d$construct == d$target_construct, 5, 2)
+  expect_error(rating_validity(d, scale_min = 1, scale_max = 5),
+               "exactly one target construct")
+
+  con <- expand.grid(item = c("I1", "I2"), judge = 1:4, objective = c("A", "B"),
+                     KEEP.OUT.ATTRS = FALSE, stringsAsFactors = FALSE)
+  con$score <- 1
+  con$target_objective <- ifelse(con$item == "I1", "A", "B")
+  con$target_objective[con$item == "I1" & con$judge == 1] <- "B"
+  expect_error(expert_validity(con, mode = "congruence"),
+               "exactly one target objective")
+})
+
 test_that("essentiality and congruence modes hand off", {
   ess <- content_handoff(expert_validity(c(10, 8, 6), mode = "essentiality", N = 12),
                          keep = c("Supported", "Review"))
@@ -721,3 +763,24 @@ test_that("instrument metadata covers held-back items too", {
   expect_false(anyNA(ev$keying))
   expect_true(all(ev$response_max == 7L))
 })
+
+test_that("status is always one of the values keep accepts", {
+  valid <- .status_definitions()$status
+  handoffs <- list(
+    content_handoff(expert_fit()),
+    content_handoff(sort_fit()),
+    content_handoff(rating_fit()),
+    content_handoff(congruence_fit()),
+    content_handoff(congruence_fit(target = FALSE), keep = "Descriptive only"),
+    content_handoff(expert_validity(c(10, 8, 6), mode = "essentiality", N = 12)),
+    content_handoff(delphi_fit(B = 0))
+  )
+  for (h in handoffs) {
+    expect_true(all(h$item_evidence$status %in% valid),
+                info = h$provenance$workflow)
+  }
+  # `carried` is the decision; it follows `status` and `keep` exactly.
+  h <- content_handoff(sort_fit(), keep = c("Supported", "Review"))
+  expect_identical(h$item_evidence$carried,
+                   h$item_evidence$status %in% c("Supported", "Review"))
+})