Warn that Limited Data Set does not scrub an 835, and fix the suite counts

README.md

@@ -132,7 +132,7 @@ newlines parses as readily as `*` and `~`.
 | Path | What it is |
 | --- | --- |
 | `web/EDIWorkbench.html` | **The app.** Self-contained: markup, styles and engine in one file. |
-| `web/tests/` | Eight correctness suites, plus generators for the committed sample files. |
+| `web/tests/` | Nine correctness suites, plus generators for the committed sample files. |
 | `web/bench/` | Benchmark over synthetic interchanges up to 150 MB. |
 | `web/PERFORMANCE.md` | Measured limits and the ordered backlog for improving them. |
 | `docs/pacdr-validation.md` | What the post-adjudicated checks do, what they deliberately don't, and why. |
@@ -146,7 +146,7 @@ it byte for byte, which is what stops the port drifting from the original.
 ## Running the tests
 
 ```sh
-node web/tests/all.mjs     # browser app: 8 suites
+node web/tests/all.mjs     # browser app: 9 suites
 python -m unittest         # Python prototype
 ```
 

docs/835-validation.md

@@ -153,9 +153,17 @@ code change, audited the same way: the rule's identifier-bearing loops are
 opens no loop at all under this feature — see "the governing rule" above —
 so `scrubLoop` never reaches one of these NM1s, and `PERSON_LOOPS` itself
 never names an 835 loop id to begin with. An 835 handed through Limited Data
-Set comes back byte-identical to what went in, which is worth stating loudly
-in the app's own UI copy: silence here would hand a user a file they believe
-is scrubbed.
+Set comes back byte-identical to what went in, and silence about that would
+hand a user a file they believe is scrubbed.
+
+So the app says it twice, at the point of use rather than only in the guide:
+a caution inside the Limited Data Set card (`#deidRemit`, shown by
+`syncRunButton` whenever `has835` is true of the loaded file), and a message
+naming the reason after a run that changes nothing. The button is deliberately
+left **enabled** for an 835 — disabling it would read as "not yet", where the
+truth is "never, by design". `render.mjs` section [5] pins all of it, including
+that the caution clears again when an 837 is loaded over an 835, and that the
+output really is byte-identical rather than merely reported as such.
 
 ## Sources, and what could not be used
 

web/EDIWorkbench.html

@@ -1831,6 +1831,9 @@ <h1 class="wordmark">EDI<span class="sep">*</span>WORKBENCH</h1>
         <p class="hint">Dates are shifted per patient rather than removed, and city, state and ZIP are kept &mdash; that is what a <b>HIPAA Limited Data Set</b> permits, and it is what keeps the file usable as test data. Providers are left real: an NPI is not patient information.</p>
         <p class="hint caution"><b>The result is still PHI.</b> A Limited Data Set is not de-identified data under HIPAA &mdash; it stays protected, and sharing it still needs a data use agreement. Only Safe Harbor or an expert determination produces data that leaves HIPAA behind, and neither is what this does.</p>
         <p class="hint">The mapping is random for every run and is never saved, so this is one-way &mdash; the output cannot be turned back into the original.</p>
+        <!-- Shown only for an 835. The rule is a no-op on one, and a silent
+             no-op hands back a file the user believes is scrubbed. -->
+        <p class="hint caution" id="deidRemit" hidden><b>This file is a remittance advice, and nothing here will be stripped from it.</b> An 835's patient identifiers sit directly inside loop <code>2100</code> (<code>NM1*QC</code>, <code>NM1*IL</code>), outside the loops this rule targets, so the file comes back byte-identical &mdash; unchanged, not de-identified. Run this on the 837 the remittance answers instead.</p>
         <button class="btn deid" id="deidBtn" type="button" disabled>
           <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round" aria-hidden="true">
             <path d="M12 3 4 6v6c0 4.5 3.2 8.4 8 9 4.8-.6 8-4.5 8-9V6l-8-3Z" />
@@ -5938,6 +5941,10 @@ <h3>Using it with real data</h3>
   // De-identification takes no configuration, so a loaded file is the only
   // thing it needs. The rule switches above deliberately don't gate it.
   $("deidBtn").disabled = !state.doc;
+  // The button stays enabled for an 835 -- running it is harmless, and
+  // disabling it would say "not yet" where the truth is "never, by design".
+  // Say so instead, before the click rather than after it.
+  $("deidRemit").hidden = !state.doc || !has835(state.doc.segments);
 }
 
 ["shiftOn", "swapOn"].forEach((id) => $(id).addEventListener("change", syncCards));
@@ -6036,7 +6043,12 @@ <h3>Using it with real data</h3>
   // way back from here, and the file on screen is now the only copy of this
   // particular mapping.
   if (!result.changes.length) {
-    toast("No patient loops found — nothing to strip");
+    // An 835 is the one file that reaches zero changes by design rather than
+    // by accident, and "nothing to strip" reads as "already clean" on it. Name
+    // the reason, so the identical download isn't mistaken for a scrubbed one.
+    toast(has835(state.doc.segments)
+      ? "Remittance advice — nothing stripped, the file is unchanged"
+      : "No patient loops found — nothing to strip");
   }
 }
 

web/PERFORMANCE.md

@@ -142,7 +142,9 @@ move the ceiling far enough that this stays unnecessary.
 
 ## Test harnesses
 
-All committed. `node web/tests/all.mjs` runs the four correctness suites
-(engine/Python parity, outline and qualifier decoding, virtual-list windowing math,
-static self-containment lint); `node web/bench/bench.mjs` runs the benchmark that
-produced the numbers above.
+All committed. `node web/tests/all.mjs` runs the nine correctness suites
+(engine/Python parity, outline and qualifier decoding, structure-tree building, the
+claim index and CSV export, control counts and claim balance, de-identification,
+virtual-list windowing math, the load path against a throwaway DOM, and the static
+self-containment lint); `node web/bench/bench.mjs` runs the benchmark that produced
+the numbers above.

web/tests/render.mjs

@@ -101,7 +101,7 @@ globalThis.getComputedStyle = globalThis.window.getComputedStyle;
 globalThis.matchMedia = globalThis.window.matchMedia;
 
 const app = new Function(
-  `${script}\nreturn { loadFile, SAMPLE_837P, SAMPLE_PACDR, SAMPLE_835, state };`
+  `${script}\nreturn { loadFile, deidentify, SAMPLE_837P, SAMPLE_PACDR, SAMPLE_835, state };`
 )();
 
 const paintedRows = () => (byId.get("docWindow").innerHTML.match(/class="row/g) || []).length;
@@ -151,6 +151,28 @@ check("the workspace is hidden", byId.get("workspace").hidden, true);
 check("an error is shown", byId.get("parseError").innerHTML.length > 0, true);
 check("stale readouts are cleared", byId.get("readouts").innerHTML, "");
 
+console.log("\n[5] the Limited Data Set warns on an 835, and stops warning off one");
+// The rule is a no-op on an 835 -- its identifiers sit in loop 2100, which
+// scrubLoop never reaches. A silent no-op hands back a file the user believes
+// is scrubbed, so the caution is the safeguard and this is what pins it.
+// Loading an 837 over an 835 has to clear it again: a caution left standing on
+// a file the rule *does* scrub is the same failure pointed the other way.
+app.loadFile("sample_835.edi", app.SAMPLE_835);
+check("the 835 caution is shown for a remittance", byId.get("deidRemit").hidden, false);
+check("the button stays enabled anyway", byId.get("deidBtn").disabled, false);
+
+app.deidentify();
+check("running it on an 835 changes nothing", app.state.result.changes.length, 0);
+check("the output is byte-identical to the input", app.state.result.output, app.SAMPLE_835);
+check("the message names the reason", byId.get("toast").textContent,
+  "Remittance advice — nothing stripped, the file is unchanged");
+
+app.loadFile("sample_837p.edi", app.SAMPLE_837P);
+check("the caution clears for an 837", byId.get("deidRemit").hidden, true);
+
+app.loadFile("broken.edi", "this is not an interchange");
+check("the caution clears when nothing is loaded", byId.get("deidRemit").hidden, true);
+
 console.log(`\n${pass} passed, ${fail} failed`);
 if (fail) process.exit(1);
 console.log("clean");