diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 58f73f86a..c6c9de516 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -11,6 +11,7 @@
 * `rugnux` writes a batch header in the unmerged MTZ for every image the observations span, not only for the images that produced one, so a scaling program reads one run per sweep.
 * `rugnux` leaves an event out of the unmerged MTZ when less of its rocking curve was captured than `--min-captured-fraction`, as the merge does, since the exported rows are declared full.
 * `rugnux` writes `FreeR_flag` with 0 for the test set and 1 for the working set, the CCP4 convention REFMAC5 defaults to; it was the other way round.
+* `rugnux` writes the merged and P1 MTZ with the reserved `HKL_base` dataset ahead of its own, so a CCP4 program reads the wavelength from the file instead of falling back to 1.54187 A.
 * `rugnux` marks the free set in the merged mmCIF as `_refln.status` = `f` instead of a `_refln.status_free` column that is not in the PDBx dictionary and that `cif2mtz` refused to read.
 * `rugnux -S` refuses or re-seats a fixed space group whose symmetry axes the indexed cell does not carry, not only one whose centring differs; `--mode scale` refuses it too.
 * The rugnux results report names the groups the data cannot separate, the enantiomorph state and any refused higher point group as `SPACE_GROUP_ALTERNATIVES`, `SPACE_GROUP_ENANTIOMORPH`, `SPACE_GROUP_REFUSED_POINT_GROUP` and `SPACE_GROUP_REFUSED_REASON`; `REPORT_VERSION` is 5.
@@ -34,6 +35,7 @@
 * `jfjoch_viewer` reads PILATUS miniCBF sweeps natively, and draws grid scan cells in the proportion of the scan steps.
 * The HDF5 and API documentation say how a grid scan records the angle its spindle stood at: send the goniometer axis with a step of 0.
 * The rugnux manual is reorganised into task pages with a run overview, and gains worked phenix / REFMAC5 / POINTLESS-AIMLESS / careless examples.
+* The rugnux manual gains worked Phaser and SHELXC/D/E examples, and says how to hand each of them the space groups the merged intensities could not separate.
 * `jfjoch_viewer` labels the merge-statistics plot over the range the axis is drawn on, so the CC1/2 curve is no longer read against tick labels covering only part of it.
 
 ### 1.0.0-rc.165
diff --git a/docs/RUGNUX_INTEGRATION.md b/docs/RUGNUX_INTEGRATION.md
index 846c6fc2d..3033d38c6 100644
--- a/docs/RUGNUX_INTEGRATION.md
+++ b/docs/RUGNUX_INTEGRATION.md
@@ -49,6 +49,20 @@ expect. Two properties worth knowing before using it:
   untouched, but the absolute scale is not meaningful. This matters only if you intend to compare
   magnitudes with another file; SHELXC and ANODE use ratios alone.
 
+**MTZ** (`<prefix>.mtz`, and `<prefix>_P1.mtz` beside it). The CCP4 anomalous layout, with the
+column types CCP4 programs dispatch on:
+
+```
+H K L IMEAN SIGIMEAN I(+) SIGI(+) I(-) SIGI(-) F SIGF F(+) SIGF(+) F(-) SIGF(-) FreeR_flag
+H H H    J        Q    K       M    K       M  F    Q    G       L    G       L          I
+```
+
+`F` is the French–Wilson amplitude. The header carries the determined space group, the refined cell
+and the wavelength, on a dataset of its own behind the reserved `HKL_base` — so `mtzinfo` and
+`mtzdmp` report the wavelength the data were collected at, and a program that takes *f′*/*f″* from
+the file is given the right edge. The Bijvoet columns are present on any rotation merge, with or
+without `-A`; a stills merge has no Bijvoet split and the file then stops after `F SIGF FreeR_flag`.
+
 ## The unmerged export
 
 `<prefix>_unmerged.mtz` holds every integrated observation, before scaling and merging, in the column
@@ -173,13 +187,188 @@ careless mono --anomalous "BATCH,dHKL,Hobs,Kobs,Lobs,XDET,YDET,BGMEAN,BGVAR,LP,F
     myrun_unmerged.mtz out/myrun
 ```
 
-**Molecular replacement.** When the report names an enantiomorphic pair (`P 41 or P 43`), let
-Phaser try both hands: `SGALTERNATIVE SELECT HAND`. When section 4's candidate table shows a close
-runner-up, `SGALTERNATIVE SELECT ALL` searches the alternatives, and `myrun_P1.mtz` is there for
-re-merging in a subgroup when the call itself is in doubt.
+**Molecular replacement and experimental phasing** each get a section of their own below —
+[Phaser](#molecular-replacement-with-phaser) and [SHELXC/D/E](#experimental-phasing-with-shelx).
+Both are where rugnux stops and the next program starts, and both meet the one thing the merged
+intensities could not decide: which of several space groups the data are in.
 
-**`iotbx.merging_statistics myrun_unmerged.mtz`** and **SHELXC on `myrun.hkl`** need no arguments
-or label choices at all.
+**`iotbx.merging_statistics myrun_unmerged.mtz`** needs no arguments or label choices at all.
+
+## Molecular replacement with Phaser
+
+rugnux does not do molecular replacement, so Phaser is the next program for anyone who has a search
+model. Both CCP4 and phenix ship it — `phaser` and `phenix.phaser`, the same 2.8.3 build in the
+versions this was checked against — and either takes the merged `myrun.mtz` as it is written.
+
+**No `LABIN`, no label choices.** Phaser reads the cell, the space group and the resolution range
+out of the file and picks the intensity columns itself. Where phenix stops on a merged file because
+it cannot choose between two equally usable observation arrays (see above), Phaser simply announces
+what it took:
+
+```
+   Data read from mtz file: myrun.mtz
+   Space-Group Name (Hall Symbol): P 41 21 2 ( P 4abw 2nw)
+   Unit Cell:   78.06   78.06   37.70   90.00   90.00   90.00
+   Column Labels Selected: IMEAN SIGIMEAN
+   Resolution on Mtz file:  0.99 39.03
+```
+
+So the whole run is the model and the cell contents:
+
+```
+phaser <<eof
+MODE MR_AUTO
+HKLIN myrun.mtz
+ENSEMBLE model PDBFILE model.pdb IDENTITY 1.0
+COMPOSITION PROTEIN MW 14300 NUMBER 1
+SEARCH ENSEMBLE model NUMBER 1
+ROOT myrun_mr
+eof
+```
+
+On a 1.0 Å dataset in a tetragonal point group that run placed one copy at TFZ 11.1, refining to
+TFZ== 80.3 and LLG 10247, in 54 s of wall clock, with no warnings about the file. The one trap in
+that script has nothing to do with rugnux: `COMPOSITION PROTEIN SEQUENCE` wants a **file name**, and
+given a chain identifier instead it fails with `FILE OPENING ERROR: X` before it reads anything. Use
+`MW` unless you have the sequence file to hand.
+
+**The space group is the interesting part.** `SPACE_GROUP_NAME` in the results report is a scalar
+and reads like a determination, but it is one of the groups the absences allow, chosen by
+convention — section 4 says which others it could not separate, as `SPACE_GROUP_ALTERNATIVES`,
+and whether the hand is open, as `SPACE_GROUP_ENANTIOMORPH= UNDETERMINED`. Merged intensities never
+name a hand: an enantiomorphic pair has the same absences and the same Laue class. Phaser is one of
+the few programs that can settle it, because a wrong hand simply fails to place the model.
+
+It does this **without being asked**. `MODE MR_AUTO` defaults to `SGALTERNATIVE SELECT HAND`, so the
+run above listed
+
+```
+   Space Group(s) to be tested:
+     P 43 21 2
+     P 41 21 2
+```
+
+and returned a single solution in `P 43 21 2` — the hand opposite the one in the MTZ header.
+Nothing in the command asked for that. The space group of the solution is the answer, whichever hand
+the file happened to carry, and it is on the `SOLU SPAC` line of the `.sol` file and in the `CRYST1`
+of the placed model.
+
+**When the alternative is not the hand**, name it. `SPACE_GROUP_ALTERNATIVES` also carries screw
+variants that share a point group — `I 2 3` and `I 21 3` on a body-centred cubic lattice is the
+common one — and `SGALTERNATIVE SELECT ALL` searches every group Phaser derives from the input one
+by translation symmetry. On a `P 41 21 2` input that is all eight of `P 4 2 2` … `P 43 21 2`, and it
+took the run above from 54 s to 65 s; on an `I 2 3` input it is `I 2 3`, `I 21 3` and an
+origin-shifted `I 2 3`. To see the list a given file would produce without searching it,
+`MODE CCA` prints it and stops:
+
+```
+phaser <<eof
+MODE CCA
+HKLIN myrun.mtz
+COMPOSITION PROTEIN MW 14300 NUMBER 1
+ROOT myrun_cca
+eof
+```
+
+**What Phaser cannot repair from this file** is a wrong *point* group. `SGALTERNATIVE` moves within
+one, so a run whose report carries a non-`NONE` `SPACE_GROUP_REFUSED_POINT_GROUP`, or a point group
+you suspect is too high, has to be merged again rather than searched again — `myrun_P1.mtz` is
+written for exactly that, and `myrun_unmerged.mtz` will do it through pointless.
+
+**mmCIF is not a route into Phaser.** `HKLIN myrun.cif` stops at `FILE OPENING ERROR: myrun.cif`,
+in both the CCP4 and the phenix build — 2.8.3 reads MTZ only. Convert rather than look for a
+keyword:
+
+```
+gemmi cif2mtz myrun.cif fromcif.mtz
+```
+
+That file gives the same solution — same space group, same placement to a hundredth of a degree,
+LLG 10248 against 10247. Its amplitude columns come out as `FP`/`SIGFP` where rugnux's own MTZ
+writes `F`/`SIGF`, which matters only if you were naming columns by hand; the automatic choice is `IMEAN`/`SIGIMEAN` either way. Since rugnux writes
+the MTZ and the mmCIF in the same run, the conversion is only worth knowing about for a file that
+arrived without its `.mtz`.
+
+## Experimental phasing with SHELX
+
+`shelxc`, `shelxd` and `shelxe` come with CCP4 (phenix does not ship them). The input is
+**`myrun.hkl`**, and it is the only one of the three reflection files that works: SHELXC 2016/1
+reads XDS and SHELX formats, not MTZ, and `SAD myrun.mtz` gets `** Cannot open file myrun.mtz **`
+— after which SHELXC exits **0** and writes nothing, so a script has to check for the `_fa.hkl` it
+should have produced rather than trust the exit status.
+
+**Nothing has to be switched on to get the anomalous signal.** A default rotation merge keeps the
+Bijvoet split, whether or not `-A` was given: `myrun.mtz` carries `I(+)`/`I(-)` and `F(+)`/`F(-)`
+beside the means, and `myrun.hkl` writes each mate as its own record, `I(+)` at `+hkl` and `I(-)` at
+`-hkl`. `-A` changes what the merging statistics are counted over, not whether the signal is in the
+file. The one case with no anomalous columns at all is a **stills** run, which computes no Bijvoet
+split; there `myrun.hkl` holds means only and there is nothing for SHELXC to work with. Unmerged
+data are not wanted anywhere in this chain either, so a run with `--no-export-unmerged` is not
+missing a file SHELX needs.
+
+**HKLF 4 carries no metadata**, so the cell and the space group have to be repeated on the SHELXC
+command — take them from `UNIT_CELL_CONSTANTS` and `SPACE_GROUP_NAME` in sections 3 and 4 of the
+report, with the spaces taken out of the group's name. (SHELXC also puts a wavelength in the `CELL`
+line of the `.ins` files it writes; that is its own 0.98 Å default, not anything read from the data,
+and neither SHELXD nor SHELXE uses it.) The whole chain, for a sulfur substructure — the cell and
+group here are tetragonal lysozyme's, so substitute your own report's:
+
+```
+shelxc sad <<eof
+SAD myrun.hkl
+CELL 79.0 79.0 38.0 90 90 90
+SPAG P41212
+FIND 10
+SFAC S
+MAXM 2
+eof
+shelxd sad_fa
+```
+
+SHELXC's own table is the first honest look at whether this is worth continuing — `<d"/σ>` should be
+about 0.80 where there is no anomalous signal. Two sweeps are quoted below, both collected at 5 keV
+for the sulfur signal: a cubic one that went all the way, and a tetragonal one that did not. The
+cubic one, 2.5 Å at 95 % completeness and multiplicity 30, reads:
+
+```
+ Resl.   Inf. 13.02  8.01  6.03  4.93  4.22  3.71  3.33  3.04  2.80  2.60  2.43
+ <I/sig>   108.8  91.4  63.0  64.7  70.6  61.8  45.5  34.7  23.7  12.4   5.0
+ %Complete  96.2 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0  99.0  72.7
+ <d"/sig>   2.58  5.06  3.97  2.92  2.36  1.68  1.50  1.33  1.48  1.38  1.79
+```
+
+**SHELXD will separate space groups the merged intensities could not.** That sweep's report named a
+body-centred cubic pair as indistinguishable, so SHELXC and SHELXD were run once per candidate —
+same reflections, same `FIND`, only `SPAG` different. One gave `CC 37.93 / CC(weak) 14.05 /
+CFOM 51.98` and the other `CC 46.76 / CC(weak) 22.61 / CFOM 69.37`. The substructure is where the
+screw axis shows itself, and the second group is the right one. This is the same handover as
+Phaser's arrived at from the other side, and it is worth doing whenever `SPACE_GROUP_ALTERNATIVES`
+is not `NONE` — SHELXD takes seconds, and the pair of runs costs less than reprocessing anything.
+
+**SHELXE decides the hand**, and says so. Run it twice, `-i` inverting the substructure. `-s` is the
+solvent fraction, `-h` says the substructure atoms belong to the native structure, as sulfur does,
+and `-a` turns on autotracing, which is what actually makes the two hands separate. The two runs
+write `sad.pdb` and `sad_i.pdb`, so they can share a directory:
+
+```
+shelxe sad sad_fa -h -s0.62 -m20 -a15 -q
+shelxe sad sad_fa -h -s0.62 -m20 -a15 -q -i
+```
+
+At 63 % solvent the two hands came out at 42.93 % and 15.28 % for the autotrace CC against the
+native data — pseudo-free CC 66.49 against 37.12, map contrast 0.87 against 0.44, 215 traced atoms
+— which is a solved structure, from `myrun.hkl` and nothing else. Where the group is one of the 22
+that come in enantiomorphic pairs, SHELXE makes the group change itself: the inverted run prints
+`** Space group converted to enantiomorph **` and writes the changed group into the `CRYST1` of its
+traced model, so the answer is readable off the output file the same way it is off Phaser's.
+
+**A negative result, for calibration.** A tetragonal dataset at the same wavelength with the same
+kind of substructure, but 87 % complete at multiplicity 20 rather than 95 % at 30, gave a plausible
+SHELXD `CFOM 47.62` and then failed at the hand: 15.33 % against 15.60 % autotrace CC, map contrast
+0.33 either way. That is not a discrimination and it is not a solution. Nothing about the file was
+the limit — the anomalous signal SHELXC measured on it was real, `<d"/σ>` reaching 4.2 — so the
+reading is that sulfur phasing wants the completeness and the multiplicity, and a `.hkl` from a
+sweep that does not have them will get this far and no further.
 
 ## Comparing the geometry with XDS
 
diff --git a/image_analysis/WriteReflections.cpp b/image_analysis/WriteReflections.cpp
index c1de19809..8332288e4 100644
--- a/image_analysis/WriteReflections.cpp
+++ b/image_analysis/WriteReflections.cpp
@@ -451,7 +451,11 @@ void WriteMtzReflections(const std::vector<MergedReflection> &reflections,
     mtz.title = "Rugnux merged reflections";
     mtz.history.push_back("From Rugnux " + jfjoch_version() + ", data reduction");
 
-    // Add dataset
+    // Add dataset. HKL_base first, so the data dataset is id 1: a dataset written as id 0 is the
+    // reserved base dataset as far as CCP4's mtzlib is concerned, and the wavelength on it is
+    // dropped - mtzinfo then reports the Cu K-alpha default 1.54187 A, which is what a program that
+    // takes f'/f'' from the file would phase on.
+    mtz.add_base();                 // the HKL_base dataset and the H K L columns
     gemmi::Mtz::Dataset& ds = mtz.add_dataset("native");
     ds.crystal_name = experiment.GetSampleName();
     ds.wavelength = experiment.GetWavelength_A();
@@ -463,9 +467,6 @@ void WriteMtzReflections(const std::vector<MergedReflection> &reflections,
     bool has_anom = true;
     const std::vector<MergedOutRow> out_rows = BuildMergedRows(reflections, experiment, has_anom);
 
-    mtz.add_column("H", 'H', dataset_id, -1, false);
-    mtz.add_column("K", 'H', dataset_id, -1, false);
-    mtz.add_column("L", 'H', dataset_id, -1, false);
     mtz.add_column("IMEAN", 'J', dataset_id, -1, false);
     mtz.add_column("SIGIMEAN", 'Q', dataset_id, -1, false);
     if (has_anom) {