diff --git a/ROADMAP.md b/ROADMAP.md
index d9f56ef..28b2701 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -1,10 +1,7 @@
 # display-report — Roadmap
 
 This layer's roadmap. Cross-repo coordination and the pipeline-wide
-sequencing live in
-[color-wrangler](https://github.com/Fuse-Technical-Group/color-wrangler);
-workstreams there are addressed by slug in backticks and resolve in
-that repository's ROADMAP.md.
+sequencing live in the umbrella project's roadmap.
 
 ## Contract-driven analysis §road:contract-analysis-impl
 
diff --git a/SPEC.md b/SPEC.md
index 2ee83da..9ab3d73 100644
--- a/SPEC.md
+++ b/SPEC.md
@@ -1,10 +1,7 @@
 # display-report — Specification
 
 This layer's specification. The pipeline-wide architecture and the
-artifact chain live in
-[color-wrangler](https://github.com/Fuse-Technical-Group/color-wrangler);
-sections there are addressed by slug in backticks and resolve in that
-repository's SPEC.md.
+artifact chain live in the umbrella project's specification.
 
 ## Scope §spec:scope
 
@@ -15,18 +12,16 @@ serial port and no DeckLink, imports no device driver, and needs no
 rig: a machine that reports needs a file and nothing else.
 
 One tool touches instruments and signal hardware, and it is
-display-measure (`§spec:session-ownership`). This repository consumes
-what that tool emits.
+display-measure. This repository consumes what that tool emits.
 
 **Why the validator owns no measure path.** display-report validates
 either this pipeline or a display-side calibration, and its value is
 that it has no stake in the correction under test. A second measure
 path living inside it would be an ungated one — the session gates
-(`§spec:signal-contract`) live in display-measure, and a path that
-skips them measures a display in an undeclared state and renders a
-well-formed report of it. That is the failure the gates exist to
-prevent, reintroduced in the repository whose independence is the
-reason they matter.
+live in display-measure, and a path that skips them measures a display
+in an undeclared state and renders a well-formed report of it. That is
+the failure the gates exist to prevent, reintroduced in the repository
+whose independence is the reason they matter.
 
 **Reproducibility is a property of the seam, not of a bundled loop.**
 A third party reproduces a report by writing the seam file from
@@ -37,16 +32,16 @@ second device path inside the validator was never what delivered it.
 Not owned here: instrument and signal-generator access
 (display-measure), OCIO semantics and config generation
 (ocio-display-gen), the show manifest and the promotion decision
-(color-wrangler).
+(the umbrella project).
 
 ## Report input §spec:report-input
 
 *Status: complete*
 
-The report's input is one file: the measurement seam file
-(`§spec:measurement-seam`), carrying the measurements, the spectra
-behind them, the protocol that produced them, the declared signal
-contract, the attested panel state, and the hash chain.
+The report's input is one file: the measurement seam file, carrying
+the measurements, the spectra behind them, the protocol that produced
+them, the declared signal contract, the attested panel state, and the
+hash chain.
 
 Analysis is a pure function of that file. Two runs over one file
 produce one report.
@@ -61,7 +56,7 @@ catches it.
 
 **Rows without a measured spectrum are legible as such.** A
 disciplined session reads its dark end with a colorimeter, and those
-rows carry a reconstructed spectrum or none (`§spec:spectral-retention`).
+rows carry a reconstructed spectrum or none.
 An analysis needing a measured spectrum reports which rows it excluded
 and why, rather than treating a scaled estimate as a measurement.
 
@@ -113,8 +108,8 @@ judges from. Plot bounds derived from a hardcoded PQ inverse place the
 measured points wrongly on a gamma session's page — the chart renders,
 and it is wrong.
 
-The page's content — what figures it carries and what they
-discriminate — is specified in `§spec:report-metrics`.
+The umbrella project's specification defines the page's content:
+what figures it carries and what they discriminate.
 
 ## Programmatic surface §spec:report-api
 
@@ -126,9 +121,9 @@ rendered report as bytes. The command-line entry point is a caller
 like any other and holds no logic of its own.
 
 **Why an importable surface.** The operator's surface is a browser
-served from the session host (`§spec:web-ui`), and it generates the
-report from a loaded artifact without the operator leaving the page or
-learning a second tool. A caller reduced to shelling out to a CLI and
+served from the session host, and it generates the report from a
+loaded artifact without the operator leaving the page or learning a
+second tool. A caller reduced to shelling out to a CLI and
 scraping a path cannot report a failure precisely.
 display-report remains the only thing that decides what a report says;
 which surface invokes it is a separate question.
diff --git a/display_report/analysis.py b/display_report/analysis.py
index db636ba..28e98cb 100644
--- a/display_report/analysis.py
+++ b/display_report/analysis.py
@@ -296,9 +296,8 @@ def black(self) -> dict:
         tmp = self._black = {}
         tmp["measurements"] = measurements = self._data.measurements[mask]
 
-        # A disciplined session reads its dark end with a colorimeter
-        # (§spec:spectral-retention), so the black rows are exactly the ones
-        # most likely to carry no spectrum. Black's tristimulus is measured
+        # A disciplined session reads its dark end with a colorimeter, so
+        # the black rows are exactly the ones most likely to carry no spectrum. Black's tristimulus is measured
         # either way; the spectral fields are only available when a black
         # row carried a spectrum, and are None rather than zero when not --
         # a zero here would read as a perfectly black display.
diff --git a/display_report/gamut.py b/display_report/gamut.py
index 12d4ef5..8d08697 100644
--- a/display_report/gamut.py
+++ b/display_report/gamut.py
@@ -1,15 +1,14 @@
-"""Gamut arithmetic for the situational-awareness view (§spec:gamut-visualization).
+"""Gamut arithmetic for the situational-awareness view.
 
 Everything here works in CIE 1976 u'v'. The 1931 xy diagram exaggerates
 green distances and crushes blue ones, which misleads exactly where
-narrow-band LED primaries land; artifacts record xy and views convert
-(§spec:gamut-visualization).
+narrow-band LED primaries land; artifacts record xy and views convert.
 
 **Coverage, never area ratio.** The fraction of a target a display can
 actually reproduce is bounded above by 1.0. The ratio of triangle areas
 is not, and it flatters a display that is large in a direction no
 content uses: the bench display measures 131.6% of Rec.709 by area and
-cannot reach Rec.709 blue (§spec:report-metrics).
+cannot reach Rec.709 blue.
 
 **A scalar hides where the shortfall is.** Coverage weights by
 chromaticity area, so a thin slice of unreachable colour reads as a
@@ -165,8 +164,7 @@ def primary_deficits(display: list[Point], standard: str) -> dict[str, float]:
     Read against the display's own native primaries rather than a
     standard's wherever the question is what a *configuration* costs: no
     display reaches Rec.709 blue, so that shortfall is a property of
-    the emitter and discriminates nothing between calibrations
-    (§spec:report-metrics).
+    the emitter and discriminates nothing between calibrations.
     """
     target = [uv_from_xy(*p) for p in STANDARD_GAMUTS[standard]]
     deficits = {}
diff --git a/display_report/transfer.py b/display_report/transfer.py
index e02d5a9..14fc787 100644
--- a/display_report/transfer.py
+++ b/display_report/transfer.py
@@ -1,4 +1,4 @@
-"""Transfer fidelity and quantization headroom (§spec:gamut-visualization).
+"""Transfer fidelity and quantization headroom.
 
 Two questions about the same ramp. Does the display follow the transfer
 function its contract declares, and is the encoding precise enough that
@@ -17,7 +17,7 @@
 here asks the same question of any encoding — a gamma contract at 12
 bits, a PQ contract at 10 — rather than asking whether a display tracks
 one particular curve. That generality is the point: it is the metric
-that survives changing the contract (§road:lut-transfer-probe).
+that survives changing the contract.
 
 Its threshold is not a hard line. Barten's model takes viewing distance,
 field size and spatial frequency, and the value used here is the peak of
@@ -46,7 +46,7 @@
 Ramp = list[tuple[int, float]] | tuple[tuple[int, float], ...]
 
 # Readings at or under this are instrument noise rather than the display's
-# response; fitting through them fits the noise (§road:instrument-floors).
+# response; fitting through them fits the noise.
 FLOOR = 0.001
 
 # The spatial frequency Barten's sensitivity peaks near, cycles per
@@ -182,8 +182,7 @@ def quantization_headroom(
     measured. The ramp is driven at 12-bit codes, so a step at 10 bits
     is four of them: this is how one contract's quantization gets
     compared against another's on the same measured display, which is what
-    §road:lut-transfer-probe needs to settle whether a PQ-like LUT
-    contract beats the 12-bit gamma one.
+    settles whether a PQ-like LUT contract beats the 12-bit gamma one.
     """
     codes_per_step = 2.0 ** (MEASURED_BIT_DEPTH - bit_depth)
     rows: list[HeadroomRow] = []
diff --git a/display_report/volume.py b/display_report/volume.py
index deb5735..6858f54 100644
--- a/display_report/volume.py
+++ b/display_report/volume.py
@@ -1,4 +1,4 @@
-"""The reproducible set as a solid (§spec:gamut-visualization).
+"""The reproducible set as a solid.
 
 A chromaticity triangle is one slice of a three-dimensional set, and it
 hides what decides whether a colour is usable: the luminance it is
@@ -15,7 +15,7 @@
 rather than drawing a capability nobody measured. This display sums to
 within 0.75% of its measured white, which is what earns the picture.
 
-**Why CIELAB.** §spec:report-metrics asks for a luminance-inclusive
+**Why CIELAB.** The report's metrics call for a luminance-inclusive
 volume in a perceptually uniform space, so that equal distances in the
 picture mean roughly equal perceived differences and the solid's shape
 carries information rather than the space's distortion. L* is relative
diff --git a/tests/fixtures/README.md b/tests/fixtures/README.md
index 35941ae..97afbc6 100644
--- a/tests/fixtures/README.md
+++ b/tests/fixtures/README.md
@@ -12,10 +12,9 @@ display-measure characterize --out hybrid_session.csmf --instrument doubles-hybr
 72 rows from the `color-wrangler/characterize/3` protocol against
 display-measure's deterministic display double, under a GAMMA 2.4 contract.
 71 rows are spectral; the black row is colorimetric and carries no spectrum
-at all, because a disciplined session reads its dark end with a colorimeter
-(§spec:spectral-retention). That row is the point of the fixture — it is the
-shape the analysis has to tolerate, and no hand-built file would have
-produced it.
+at all, because a disciplined session reads its dark end with a colorimeter.
+That row is the point of the fixture — it is the shape the analysis has to
+tolerate, and no hand-built file would have produced it.
 
 Regenerate it from display-measure when the protocol or the seam format
 changes. Do not hand-edit it.
diff --git a/tests/test_gamut.py b/tests/test_gamut.py
index 4a86ac3..e6e0218 100644
--- a/tests/test_gamut.py
+++ b/tests/test_gamut.py
@@ -1,8 +1,7 @@
-"""Gamut arithmetic for the situational-awareness view (§spec:gamut-visualization).
+"""Gamut arithmetic for the situational-awareness view.
 
 The numbers here are the bench display's, measured under three different
-panel configurations, because the arithmetic's job is to tell them apart
-(§spec:report-metrics).
+panel configurations, because the arithmetic's job is to tell them apart.
 """
 
 import pytest
@@ -42,7 +41,7 @@ def test_a_gamut_covers_itself_entirely(self) -> None:
     def test_coverage_never_exceeds_one(self) -> None:
         """The bench display's triangle is larger than Rec.709's by area and
         still cannot reproduce all of it — the distinction area ratios
-        lose (§spec:report-metrics)."""
+        lose."""
         for primaries in (AUG12, AUG29):
             for name in STANDARD_GAMUTS:
                 target = [uv_from_xy(*p) for p in STANDARD_GAMUTS[name]]
diff --git a/tests/test_transfer.py b/tests/test_transfer.py
index 38aacb7..f817c29 100644
--- a/tests/test_transfer.py
+++ b/tests/test_transfer.py
@@ -1,4 +1,4 @@
-"""Transfer fidelity and quantization headroom (§spec:gamut-visualization).
+"""Transfer fidelity and quantization headroom.
 
 The bench display's own gray ramp, dark-room capture. Its declared contract
 is a 2.35 power law and it does not follow one: the measured response
@@ -108,7 +108,7 @@ def test_the_threshold_flag_tracks_the_ratio(self) -> None:
 
 class TestBitDepthComparison:
     """The same measured display, judged against a different encoding — the
-    comparison §road:lut-transfer-probe needs."""
+    comparison that settles which LUT contract to run."""
 
     def test_fewer_bits_make_every_step_coarser(self) -> None:
         twelve = {
diff --git a/tests/test_volume.py b/tests/test_volume.py
index 6e88c6e..8b39bcb 100644
--- a/tests/test_volume.py
+++ b/tests/test_volume.py
@@ -1,4 +1,4 @@
-"""The reproducible set as a solid (§spec:gamut-visualization).
+"""The reproducible set as a solid.
 
 A chromaticity triangle is one slice of a three-dimensional set, and it
 hides the thing that decides whether a colour is usable: the luminance