D-sudoasd · D-sudoasd · Aug 27, 2026 · Aug 27, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,19 @@
 
 ## [2.0.0] - Unreleased
 
+### Release-readiness hardening (2026-08-27)
+
+- Harden the Workbench preflight boundary with content-aware input identities,
+  language-only display refreshes, localized empty-queue/completion messages,
+  and stale-progress reset on failed Tab 2/3 runs. Update the release-readiness
+  snapshot and operator documentation; this remains an unreleased source
+  candidate.
+- Harden structured canSAS/NXcanSAS Q-axis handling to fail closed on missing
+  or unknown units, and canonicalize `1/m` to Å⁻¹. Detect NaN/Inf in headerless
+  profiles, publish canSAS XML atomically, handle Pydidas numeric tokens and
+  inline comments, and reject empty arrays in the scientific core. These are
+  engineering safeguards, not measured scientific acceptance.
+
 ### Fluorescence subtraction (1D)
 
 - Add an opt-in absolute-scale fluorescence kernel

diff --git a/README.md b/README.md
@@ -20,11 +20,15 @@
 reusable data writers, and provenance checks. The result and the processing
 record remain reviewable together.
 
-Reviewers should use the unreleased 2.0.0 tree on
-[`main`](https://github.com/D-sudoasd/SASAbs). GitHub Release
-[v1.1.1](https://github.com/D-sudoasd/SASAbs/releases/tag/v1.1.1) is an earlier
-archive and is not this candidate. Do not treat the Zenodo concept DOI as a
-version DOI for 2.0.0.
+Version status:
+
+- Current source candidate: branch [`main`](https://github.com/D-sudoasd/SASAbs),
+  version `2.0.0`, unreleased.
+- Stable archive: GitHub Release
+  [`v1.1.1`](https://github.com/D-sudoasd/SASAbs/releases/tag/v1.1.1) release assets.
+- The `2.0.0` candidate is source-only: no PyPI installation is documented, and
+  no version tag, GitHub Release, or Zenodo version archive has been created.
+  The DOI above is the project concept DOI.
 
 <p align="center">
   <a href="#quick-start"><strong>Quick start</strong></a> ·
@@ -55,8 +59,11 @@ python -m pip install -e ".[gui]"
 saxsabs-workbench --lang en
 ```
 
-The core package requires Python 3.10+, NumPy, pandas, and xraydb. The project
-does not currently document a PyPI installation.
+On Windows, `py -m pip install -e ".[gui]"` and `py saxsabs_workbench.py --lang en`
+are equivalent Python-launcher forms.
+
+The core package requires Python 3.10+, NumPy, pandas, and xraydb. The commands
+above install from the source tree.
 
 <details>
 <summary><strong>Optional dependency groups</strong></summary>
@@ -189,9 +196,10 @@ python scripts/check_submission_readiness.py \
 ```
 
 Run the strict command from the exact branch and commit that will be submitted.
-PR #1 is already on `main`. A PASS recorded on an earlier revision does not
-cover a later commit; update `submitted_branch` and `submitted_commit` and rerun
-the gate on the revision sent to JOSS.
+The submitted branch and 40-character SHA must identify the same revision as the
+public README and paper blobs and the successful CI run. A PASS recorded for an
+earlier revision does not cover a later commit; update `submitted_branch` and
+`submitted_commit` and rerun both gates.
 
 After the strict local gate passes, verify the same commit, branch, visible
 README and paper, repository identity, and successful CI run against GitHub:

diff --git a/SASAbs.py b/SASAbs.py
diff --git a/SUBMISSION_READINESS.md b/SUBMISSION_READINESS.md
@@ -1,29 +1,32 @@
 # Submission readiness snapshot
 
-Updated: 16 August 2026 (Asia/Shanghai)
+Updated: 27 August 2026 (Asia/Shanghai)
 
-Review the unreleased 2.0.0 tree on `main`, not GitHub Release v1.1.1. Do not
-create `v2.0.0`, a GitHub Release, or a Zenodo version archive during review.
+Review the current unreleased 2.0.0 source candidate on `main`; the stable
+archive is GitHub Release v1.1.1 and its release assets. `v2.0.0` remains
+unreleased: do not create its tag, GitHub Release, or Zenodo version archive
+during this review.
 
 ## Locally verified
 
-- Full source suite: PASS in a fully provisioned Python 3.13 environment; exact
-  count and duration are retained in the dated external validation record.
-- Ruff: root modules, package, tests, paper scripts, and submission gate pass.
+- Full source suite: PASS under Python 3.11 and 3.12, with
+  `py -3.11 -B -m pytest -q -p no:cacheprovider --tb=short -W error` and the
+  equivalent Python 3.12 command each reporting `1155 passed`.
+- Full repository Ruff check: PASS.
+- Current `git diff --check`: PASS.
+- Distribution smoke from a clean temporary clone outside the checkout: sdist
+  and wheel builds, fresh-venv installation of `wheel[gui,hdf5]`, CLI/import/
+  `pip check`, and the `minimal_2d` synthetic smoke all PASS. The temporary
+  clone path is intentionally omitted because it is not durable evidence.
+- Python 3.10 and 3.13 remain pending the remote CI matrix; the local full-suite
+  evidence above covers only Python 3.11 and 3.12.
 - README: 5 local images and all local links resolve; SVG/image audit passes.
 - Minimal 2D example: 9×9 homemade radial average (not pyFAI) recovers planted
   K and sample maximum relative errors of `0.001933697...`; CSV, TSV, XML, and
-  HDF5 outputs are written with unknown uncertainty.
-- Fresh-copy distribution build: wheel and sdist PASS from a source tree
-  outside every Git checkout. The exact archive inventory is retained in the
-  dated external validation record; the sdist includes README assets,
-  workflows, docs, examples, tests, and paper sources.
-- Installed-wheel smoke: CLI reports `saxsabs 2.0.0`; `SASAbs`,
-  `saxs_mpl_style`, and `saxsabs` import from the temporary environment; the
-  copied minimal example passes outside the checkout. A fresh Python 3.13
-  environment resolves the declared GUI/HDF5 extras with no broken
-  requirements.
-- Paper: 1100-word body by the documented Pandoc method; 16 references; current
+  HDF5 outputs are written with unknown uncertainty. This synthetic smoke is
+  an engineering/reproducibility check, not BL19B2 measured scientific
+  acceptance.
+- Paper: 1228-word body by the documented Pandoc method; 16 references; current
   Inara TeX and well-formed JATS resolve both figures.
 - Review PDF: the official CI paper job produces a five-page draft whose pages,
   bounds, figures, citations, and embedded fonts have been visually checked.
@@ -56,6 +59,9 @@ create `v2.0.0`, a GitHub Release, or a Zenodo version archive during review.
 6. Before submission, verify that the public GitHub description, homepage
    concept DOI, visible README, submitted branch, and green CI all identify the
    exact candidate revision.
+7. Complete measured beamline/scientific acceptance with archived raw inputs,
+   repeatability, and an independent comparison; synthetic validation and
+   engineering tests do not satisfy this gate.
 Run the strict decision gate with Pandoc available:
 
 ```bash
@@ -64,10 +70,10 @@ python scripts/check_submission_readiness.py \
   --manual-confirmations path/to/submission-confirmations.json
 ```
 
-The gate must run on the exact branch and commit submitted to JOSS. PR #1 is
-already on `main`; rerun the gate on the clean `main` commit that will be
-submitted and record `submitted_branch` and `submitted_commit` accordingly.
-Evidence from an earlier revision is not evidence for a later commit.
+The gate must run on the exact branch and commit submitted to JOSS. Record the
+submitted branch and 40-character SHA, and require that the local/public
+README, paper blobs, and successful CI run all resolve to that same revision.
+Evidence from an earlier commit does not cover a later commit.
 
 After that local PASS, run:
 
@@ -83,7 +89,8 @@ editorialbot branch command when the paper is not on `main`.
 
 The current strict result is intentionally **FAIL** because the paper still has
 four author-input placeholders, no confirmed corresponding author, and no paper
-email. The mechanical preflight passes when
+email. No research-use evidence or measured scientific acceptance is recorded
+as complete. The mechanical preflight passes when
 `--allow-author-placeholders --as-of 2026-08-26` is used; this override is not a
 submission authorization.
 

diff --git a/codemeta.json b/codemeta.json
@@ -40,7 +40,7 @@
     "SRM 3600"
   ],
   "dateCreated": "2026-02-25",
-  "dateModified": "2026-08-16",
+  "dateModified": "2026-08-27",
   "developmentStatus": "active",
   "softwareRequirements": [
     "numpy >= 1.24",

diff --git a/docs/architecture.md b/docs/architecture.md
@@ -68,11 +68,18 @@
   control and both Tab 2/Tab 3 existence-only resume controls are UI-disabled;
   forced values make Dry Check BLOCKED and are rejected again at Run. K and μ
   are read-only in Tab 2, and K is read-only in Tab 3.
-- Workbench file identities currently bind resolved path, size, and mtime, not a
-  content SHA-256 for every selected source. `CAUTION` currently permits Run
-  without a separately persisted acknowledgement; these are deliberate open
-  boundaries, not properties of the strict runners. BG/Dark reference-library
-  mutations explicitly invalidate the in-memory Tab 2 approval.
+- Workbench identities bind resolved path, size, mtime, and a streaming SHA-256
+  for selected queue files and currently active configured inputs. A disabled
+  optional field may retain a stale raw path, but its identity is normalized as
+  disabled: it is not hashed and does not block Run. When that input is enabled,
+  a missing, unreadable, or changing file is unverified and blocks Run; replacing
+  content at the same path with the same size/mtime changes the preflight
+  fingerprint. Changing a switch or configuration still invalidates approval;
+  language refresh only redraws display text and retains current Tab 2/Tab 3
+  approvals. `CAUTION` currently permits Run without a separately persisted
+  acknowledgement; these are deliberate open boundaries, not properties of the
+  strict runners. BG/Dark reference-library mutations explicitly invalidate the
+  in-memory Tab 2 approval.
 - Tab 3 raw correction is disabled. Formal K/Kd accepts only an explicitly
   reduced `relative` profile; `raw_counts`, `absolute_cm^-1`, and `ambiguous`
   states fail closed. K/d requires `d > 0`; K-only applies K without repeating
@@ -149,7 +156,8 @@
 - **Implemented**: normalization, header parsing, external 1D parsing, robust K
   estimation, NIST 30 keV material core, Elam diagnostic calculator, 1D
   intensity ledger, signed-in-memory Workbench preflight, fixed-thickness
-  enforcement, disabled legacy/resume controls, exact K-only/Kd/buffer gates,
+  enforcement, content-aware source identities, disabled legacy/resume controls,
+  exact K-only/Kd/buffer gates,
   absolute-buffer validation, optional absolute 1D fluorescence subtraction,
   provenance-aware scrollable μ UI, disabled Tab 3 raw mode, screen-aware
   startup, strict BL19B2 workflows, standard writers, bilingual GUI, CLI, CI,
@@ -164,7 +172,9 @@
   is disabled rather than treated as safe.
 - **Desktop operation**: long GUI jobs run on the Tk event thread and are not
   cancellable. Users should prefer headless workflows for unattended or large
-  campaigns. `CAUTION` remains visible but is not separately persisted as an
+  campaigns. Tab 2/Tab 3 progress resets at job entry and is marked failed on
+  an outer run error, but whole-job atomic publication is not implemented.
+  `CAUTION` remains visible but is not separately persisted as an
   acknowledgement.
 - **Input resources**: Workbench and headless detector readers share
   `saxsabs.io.detector_images`. Reviewers should still use the documented

diff --git a/docs/bl19b2_abs2d_batch_runbook.md b/docs/bl19b2_abs2d_batch_runbook.md
@@ -211,10 +211,32 @@ scientific outputs silently.
 
 ## Reuse Command
 
-For a new beamtime, copy `examples/bl19b2_abs2d_template/processing_config.example.yml`,
-edit the paths, run a dry scan first, then run the full export.
+For a new beamtime, copy `examples/bl19b2_abs2d_template/processing_config.example.yml`
+and edit the paths. Always use the following two-step sequence: inspect the
+dry-run result first, then repeat the same command without `--dry-run` only
+after the inputs and planned output root have been confirmed.
 
-Template command:
+Step 1 — preflight only (no formal output publication):
+
+```powershell
+$env:PYTHONPATH='src'
+python -m saxsabs.cli bl19b2-abs2d `
+  --input-root '<BL19B2 DATA>\datXXX' `
+  --pydidas-cali-yaml '<BL19B2 DATA>\datXXX\reference_saxs\Cali.yaml' `
+  --output-root '<BL19B2 DATA>\datXXX_absolute_corrected_2D' `
+  --monitor-mode rate `
+  --mu 20.2 `
+  --standard-key SRM3600 `
+  --correct-solid-angle-for-k `
+  --no-polarization-correction `
+  --dry-run
+```
+
+Review the returned status, discovered sample inventory, geometry/mask
+selection, thickness and uncertainty gates, and output paths. Correct any
+failure or unexpected classification before continuing.
+
+Step 2 — formal run after explicit confirmation:
 
 ```powershell
 $env:PYTHONPATH='src'

diff --git a/docs/joss-submission-checklist.md b/docs/joss-submission-checklist.md
@@ -1,7 +1,9 @@
 # JOSS submission checklist
 
 This checklist follows the current JOSS author and reviewer documentation,
-accessed 16 August 2026:
+accessed 27 August 2026. It is a release-readiness snapshot for the unreleased
+2.0.0 source candidate on `main`; the stable archived assets remain Release
+v1.1.1.
 
 - [Submission requirements](https://joss.readthedocs.io/en/latest/submitting.html)
 - [Paper format](https://joss.readthedocs.io/en/latest/paper.html)
@@ -11,9 +13,10 @@ accessed 16 August 2026:
 ## Pre-review screening gates
 
 - [ ] **More than six months of public development.** GitHub reports that this
-      repository was created on 25 February 2026. The date gate is therefore not
-      satisfied on 16 August 2026; 26 August 2026 is the first conservative
-      submission date, provided public development remains active.
+      repository was created on 25 February 2026. The first conservative
+      eligibility date is 26 August 2026; the author must still recheck the
+      public history on the actual submission date and keep this gate open until
+      that evidence is confirmed.
 - [ ] **Demonstrated research use.** The repository contains a concrete BL19B2
       workflow and reproducible synthetic validation material, but the author
       must supply evidence that the software has been used in research. Claims
@@ -43,6 +46,8 @@ accessed 16 August 2026:
 - [ ] Immediately before submission, record green push and Draft-PR runs for
       the exact submitted HEAD in the dated external validation record. Do not
       embed a self-referential commit hash in this tracked checklist.
+- [ ] Confirm that the submitted branch and 40-character SHA identify the same
+      revision as the visible public README, paper blobs, and successful CI run.
 - [ ] Verify that the public repository description, homepage concept DOI,
       visible README, and submitted branch identify the same candidate.
 - [ ] Run `scripts/check_public_candidate.py` against the completed confirmation
@@ -75,6 +80,9 @@ accessed 16 August 2026:
       acknowledgements, funding, conflicts of interest, and contribution roles.
 - [ ] The author confirms the complete AI disclosure and human review statement.
 - [ ] The author supplies research-use evidence suitable for the impact section.
+- [ ] Measured beamline/scientific acceptance is archived with raw inputs,
+      repeatability, and an independent comparison; synthetic validation and
+      engineering tests do not satisfy this gate.
 - [x] The current official Inara workflow converts the paper to TeX and
       well-formed JATS with citations and figures resolved.
 - [x] The current candidate PDF was built from Inara-generated TeX with

diff --git a/examples/manual-verification.md b/examples/manual-verification.md
@@ -14,7 +14,7 @@ cannot be fully public.
 1. Install package:
 
    ```bash
-   pip install -e .[dev,hdf5]
+   pip install -e ".[dev,hdf5]"
    ```
 
 2. Run tests:
@@ -104,12 +104,23 @@ front end to the strict BL19B2 campaign runner.
    one tracked scientific parameter or source path and confirm Run immediately
    disables. Add BG/Dark files, add either library recursively, and clear the
    libraries; every mutation must immediately invalidate Tab 2 approval. Run
-   must also reject a stale approval if a file's recorded size or modification
-   time changes after Dry Check.
+   must also reject a stale approval if a selected queue file or currently active
+   configured input's content, size, or modification time changes after Dry
+   Check. Replace a fixture with different bytes while preserving its size and
+   mtime to verify the streaming SHA-256 catches the replacement. Missing,
+   unreadable, or changing-while-read active files must fail closed rather than
+   become valid identities. Leave a stale missing path in a disabled optional
+   buffer/fluorescence field and confirm it does not block; then enable that
+   option without changing the path and confirm the active identity blocks Run.
+   Changing a switch or configuration must invalidate approval. Toggle the UI
+   language and confirm the approval remains current; this display-only action
+   must not invalidate Tab 2/Tab 3 preflight.
 5. Repeat with BLOCKED fixtures and confirm Run remains disabled. Include Tab 3
    K/d with blank, non-finite, zero, and negative thickness. Record that a
    CAUTION result currently permits Run without a separately persisted
-   acknowledgement; this remains an open release gate.
+   acknowledgement; this remains an open release gate. Set a prior progress bar
+   to 100%, trigger a preflight failure at each Run entry, and confirm the bar
+   returns to 0 with a visible failed-job status.
 6. In the material calculator, select the NIST 30 keV source and verify:
 
    - Ti-24Nb-4Zr-8Sn: `74.550355 cm^-1`
@@ -202,13 +213,18 @@ These capabilities are outside the current Workbench support contract. They are
 not claimed by the README or paper; use the strict headless workflow where
 applicable:
 
-- formal multi-folder/per-sample fixed-thickness campaigns have a Workbench
-  owner equivalent to the strict CLI/batch campaign;
-- Workbench and strict BL19B2 runner use one shared scientific kernel;
-- Workbench output root has an owner manifest, atomic campaign publication, and
-  content-signature resume (existence-only resume must remain disabled);
-- Workbench preflight binds critical file content hashes and persists explicit
-  CAUTION acceptance;
+- formal multi-folder/per-sample fixed-thickness campaigns do not have a
+  Workbench owner equivalent to the strict CLI/batch campaign;
+- Workbench and strict BL19B2 runner do not use one shared campaign-level
+  scientific kernel;
+- Workbench output root does not have an owner manifest, atomic campaign
+  publication, or content-signature resume (existence-only resume must remain
+  disabled);
+- Workbench preflight binds SHA-256 content identities for selected queue files
+  and currently active configured inputs. Disabled optional fields may retain a
+  stale path without hashing or blocking; enabling them makes missing,
+  unreadable, or changing inputs fail closed. Configuration changes still
+  invalidate approval, and explicit CAUTION acceptance is not persisted;
 - all FabIO readers pass OS-level handle audits on every Windows workstation
   (unit tests now cover the shared copy-and-close helper; a full desktop
   handle audit remains a local check);

diff --git a/examples/minimal_2d/README.md b/examples/minimal_2d/README.md
@@ -24,7 +24,7 @@ Expected key result:
 - `summary.json` with `k_relative_error < 0.005` and
   `sample_max_relative_error < 0.01`
 - `absolute_profile.csv`, `absolute_profile.tsv`, `absolute_profile.xml`
-- `absolute_profile.h5` if `h5py` is installed (`pip install -e .[hdf5]`)
+- `absolute_profile.h5` if `h5py` is installed (`pip install -e ".[hdf5]"`)
 
 The example recovers a planted synthetic $K$ and sample curve on a 9×9 array
 using a homemade integer-bin radial average (not pyFAI), writes labeled

diff --git a/src/saxsabs/cli.py b/src/saxsabs/cli.py
@@ -471,6 +471,9 @@ def _normalize_q_profile(
     if source_unit == "nm^-1":
         x = x / 10.0
         conversion = "nm^-1_to_A^-1"
+    elif source_unit == "m^-1":
+        x = x * 1.0e-10
+        conversion = "m^-1_to_A^-1"
 
     updated = dict(profile)
     updated["x"] = x.copy()

diff --git a/src/saxsabs/core/buffer_subtraction.py b/src/saxsabs/core/buffer_subtraction.py
@@ -66,6 +66,8 @@ def _as_1d_float_array(name: str, values: np.ndarray | None, *, require_finite:
     arr = np.asarray(values, dtype=np.float64)
     if arr.ndim != 1:
         raise ValueError(f"{name} must be a 1-D array")
+    if arr.size == 0:
+        raise ValueError(f"{name} must not be empty")
     if require_finite and not np.all(np.isfinite(arr)):
         raise ValueError(f"{name} contains non-finite values")
     return arr