diff --git a/research/p104-fiber-tag-orientation-check/README.md b/research/p104-fiber-tag-orientation-check/README.md
new file mode 100644
index 000000000..5ef6500df
--- /dev/null
+++ b/research/p104-fiber-tag-orientation-check/README.md
@@ -0,0 +1,111 @@
+# #4154's P104 fiber-tag seam, checked in its sources' orientation
+
+**Claim type: an independent finite check with a correction. No new κ.** The
+selected conditional κ stays `0.001672660197076918`.
+
+## Object
+
+[PR #4154](https://github.com/CrocSwap/integer-mult-bounds/pull/4154)
+(unpaidmoon, merged, `research/p104-fiber-tag-seam`) compares #3966's row tag
+`z ⊗ q_j` with its column tag `r_i ⊗ z` on the 104 P104 labels, and pins
+#3954's witness. The `p104-side-correction` card lists it as still needing
+independent validation.
+
+**The orientation question.** The sources fix which tag is which:
+
+- **#3966** (`p104-phase-uniform-gauge/PROOF.md` §4) defines the row
+  correction as `z ⊗ q_j` and the column correction as `r_i ⊗ z`, with
+  `q_j = O p_j` for the row label and `r_i = O p_i` for the column label.
+- **#3954**'s witness lines are `386 ⊗ 386` (row port) and `405 ⊗ 386`
+  (donor port), decoded from its `RESULT.json`. Both lie in the row fiber
+  `q = 386`, so the pair is row 386, column 405, on the shared line
+  `r_i ⊗ q_j = 49 ⊗ 386`.
+
+#4154's README states that convention. Its code does not: `tags(row_image,
+column_image)` returns `z ⊗ column_image` as the row tag and
+`row_image ⊗ z` as the column tag. Those are the transposes.
+
+## Results
+
+**Confirmed: everything that does not depend on orientation.**
+
+| Claim | Result |
+| --- | --- |
+| Labels | `labels.json` equals `physical_labels` in #3966's source `RESULT.json` |
+| Gauge | `O` is an orthogonal involution with the weight law, on all of `F₂⁹` |
+| Label classes | 56 of class 1 and 48 of class 3; every image is class 3 |
+| Full address | no agreement in any of the 10,816 ordered pairs |
+| Shared line | 5,440 pairs agree and 5,376 disagree |
+| Uniform tag `a` | never a row tag or a column tag |
+| Minimum Hamming distance | 6, only at labels 4 and 32 |
+| Witness Hamming distance | 20 |
+
+**Corrected: the witness and one census sentence.** These are in #3966's
+orientation, at row 386 and column 405.
+
+| | #4154 | Source orientation |
+| --- | --- | --- |
+| Shared-line flips (row, column) | `(0, 1)` | `(1, 0)` |
+| Does `a` match the row tag on the shared line? | no | **yes** |
+| Separating cell | bit 11 = `(1, 2)` | bit 11 does not separate; bit 19 = `(2, 1)` does |
+| Helper matches the row tag on the shared line exactly when | the column label is class 3 | **the row label** is class 3 |
+
+The count is `104 · 48` ordered pairs either way.
+
+- **The full-address obstruction stands.** `a` is never a row tag, so a
+  helper carrying `a` still fails to match a row owner on the full 81-bit
+  address.
+- **The shared-line conclusion does not.** #4154 says that on the witness's
+  shared line "a helper carrying `a` does not match this row owner". In the
+  source orientation it does match there. That conclusion is the premise of
+  #4154's open question 2.
+
+**Control.** Evaluating #4154's transposed objects reproduces its numbers
+exactly: flips `0/1`, the helper mismatch and bit 11. The census is
+unchanged. The difference is therefore exactly the transpose. The verifier
+also checks the line in #4154's code that builds the row tag from the column
+image.
+
+**Provenance note.** #4154 says #3966's `RESULT.json` "is not a Git object on
+this base". It is on main, as
+`research/p104-phase-uniform-gauge/source/RESULT.json`, with the pinned SHA256
+`ce05a27e…`.
+
+## Method and independence
+
+The verifier imports nothing from #4154. It represents tensors as sets of
+`(α, β)` cells rather than packed integers. It reads the labels from #3966's
+file, not #4154's copy, and it reads the orientation from #3966's text and
+#3954's witness vectors.
+
+## Not covered
+
+- Native routing, owners, dirty return, precision and the inherited
+  44,096 excess, all of which #4154 also leaves open.
+- Which orientation a future P104 caller should use. This package only
+  matches the sources' definitions.
+
+## Reproduce
+
+```sh
+python3.12 -B research/p104-fiber-tag-orientation-check/verify.py
+```
+
+It prints `PASS_P104_FIBER_TAG_ORIENTATION_CHECK` in about 2 seconds, using
+only the standard library.
+
+## Open next questions
+
+1. **Revisit #4154's open question 2 in the source orientation.** At the
+   witness, the uniform tag already agrees with the row tag on the shared
+   line. The remaining question is only full-address agreement, which `a`
+   can never give.
+   - *First step:* restate the question for the full address alone.
+   - *Blocked by:* nothing.
+
+## Credit
+
+The seam comparison is unpaidmoon's (#4154). The labels, the gauge and the
+corrections are #3966's (ToikY); the witness is #3954's (Clear1nsight).
+Prepared by Ryanson Jonathan with substantial Claude Code (Claude Opus 5.5)
+assistance. Apache-2.0.
diff --git a/research/p104-fiber-tag-orientation-check/SOURCES.json b/research/p104-fiber-tag-orientation-check/SOURCES.json
new file mode 100644
index 000000000..6b24ef310
--- /dev/null
+++ b/research/p104-fiber-tag-orientation-check/SOURCES.json
@@ -0,0 +1,54 @@
+{
+  "schema_version": 1,
+  "base_commit": "12fa8c5c844be21bfefb38a19503fbd867a11244",
+  "files": [
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "b1fea0c94436b236cc2b7d41a7be032811b94c81",
+      "sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "64bfc5af6709ed9811171e0a2108f05e1219cb65",
+      "sha256": "352a6184713b8843d96fd2c2071eb7aeaa655906b7c51d45ad444912666c8a96"
+    },
+    {
+      "path": "research/p104-fiber-tag-seam/README.md",
+      "usage": "input",
+      "git_blob": "149821201622a046742baa2df4563296cebe9800",
+      "sha256": "b0b493d9db3b7c99e8497a761678f11c0e331224d37bf8c9d56747517ff9c2a2"
+    },
+    {
+      "path": "research/p104-fiber-tag-seam/verify.py",
+      "usage": "input",
+      "git_blob": "836a547850b9d8740c99bf0f9f9eaa73afcbf2a2",
+      "sha256": "81e24ecc7aad4a504efadddb5afe99b371ca93289d2502efee2621208497777a"
+    },
+    {
+      "path": "research/p104-phase-uniform-gauge/PROOF.md",
+      "usage": "input",
+      "git_blob": "fd701d86feaead86d711fa8d25914c68ea70781c",
+      "sha256": "a0e72e6b48e75edf2948b455ebe59c7d912682204b34d916bfed804cea35c405"
+    },
+    {
+      "path": "research/p104-phase-uniform-gauge/source/RESULT.json",
+      "usage": "input",
+      "git_blob": "c7ed0ea02e8c462694762508b0bdd1dde2e116d0",
+      "sha256": "ce05a27ebc98677bec4febaa488fa03f5b6fc60216aa43b34298444d12f7ab6c"
+    },
+    {
+      "path": "research/p104-two-axis-compatibility/PROOF.md",
+      "usage": "input",
+      "git_blob": "1d84b2a1188eab6a5c38743cc1eed32d66446a3d",
+      "sha256": "bdc7d868e4411541a33179af683f48011e66d6446a45bef363f3f3dd5163f717"
+    },
+    {
+      "path": "research/p104-two-axis-compatibility/RESULT.json",
+      "usage": "input",
+      "git_blob": "c7ed0ea02e8c462694762508b0bdd1dde2e116d0",
+      "sha256": "ce05a27ebc98677bec4febaa488fa03f5b6fc60216aa43b34298444d12f7ab6c"
+    }
+  ]
+}
diff --git a/research/p104-fiber-tag-orientation-check/TASK.json b/research/p104-fiber-tag-orientation-check/TASK.json
new file mode 100644
index 000000000..4ca87bc9b
--- /dev/null
+++ b/research/p104-fiber-tag-orientation-check/TASK.json
@@ -0,0 +1,133 @@
+{
+  "schema_version": 1,
+  "task_id": "p104-side-correction",
+  "priority_revision": 7,
+  "base_commit": "12fa8c5c844be21bfefb38a19503fbd867a11244",
+  "original_card": {
+    "id": "p104-side-correction",
+    "question": "Can a changed P104 producer/consumer contract escape the scoped slot, chart and completion obstructions with a favorable full bill?",
+    "starting_points": [
+      "docs/research/research-family-status.md",
+      "certificates/selected-result.json",
+      "research/p104-literal-slot-defect/README.md",
+      "research/p104-uniform-commutation-gap/README.md",
+      "research/p104-uniform-encoding-gap/README.md",
+      "research/p104-source-chart-shortcut/README.md",
+      "research/p104-fiber-tag-seam/README.md"
+    ],
+    "methods": [
+      "Construct one side-correction word at an explicit chronological cut with the needed frames still available.",
+      "Derive a lower bound on the side correction and hosting costs that decides whether this channel can beat the selected budget."
+    ],
+    "acceptance_gate": "An exact arbitrary-input identity, source/target frame availability and restoration at the chosen cut, plus the complete channel+correction+idle/hosting bill or an explicit favorable bound. The native channel alone is not a supplier. A scoped local result may be retained without a complete supplier. Expanding it into further construction rounds requires the family-level budget and obligation gate in research-family-status.md; an optimistic passing relaxation alone is not acceptance.",
+    "first_step": {
+      "command": [
+        "python3",
+        "-B",
+        "scripts/research_baseline.py"
+      ],
+      "expected": "PASS_RESEARCH_BASELINE_BINDINGS_ONLY",
+      "inspect": "Read the named successor obstructions before extending #2981 static flags. Distinguish literal-slot action, commutation, uniform encoding, source-chart shortcut and fiber-tag seam scopes; none excludes all P104 mechanisms.",
+      "scope": "Source-binding orientation only; no scientific replay, theorem acceptance or supplier admission. Budget any heavy replay separately."
+    },
+    "first_experiment": "Name one hypothesis changed from these exact interfaces. Write a literal producer-to-consumer timeline and an arbitrary-dirty-input positive/control pair, then charge acquisition, coefficients, completion, hosting and restoration. Stop if the same slot/fiber mismatch persists or the favorable full budget already fails. A scoped independent audit remains useful; another static count is not a new completion.",
+    "bottleneck_link": "Completing an already realized changed channel tests a concrete construction opportunity, instead of refining the incumbent packing. A positive local result still needs global and all-size admission.",
+    "allowed_variations": [
+      "One correction factorization, one chronological cut, or one explicit hosting obstruction.",
+      "A changed correction schedule with an updated full bill and native frame contract."
+    ],
+    "scope_change_triggers": [
+      "Recompiling the already certified low-rank channel without closing a new obligation.",
+      "Treating a pending ancestor, local histogram or unpriced side correction as an admitted global supplier.",
+      "Repeating a correction or physical conversion already supplied by a pending follow-up without adding independent coverage or a different unresolved obligation.",
+      "Repeating the 208-flag static matching or calling a static nested cover a chronological correction program. Independent audit of that certificate is distinct and remains useful.",
+      "Repeating a static cover, source-chart shortcut or fiber-tag seam without changing the hypothesis responsible for its obstruction."
+    ],
+    "overlap_terms": [
+      "P104",
+      "I-G/2",
+      "side correction",
+      "native channel",
+      "1972",
+      "1991",
+      "2586",
+      "endpoint transport",
+      "physical dual",
+      "relay composition",
+      "2981",
+      "joint unit flags",
+      "acquisition delivery",
+      "static nesting",
+      "4133",
+      "4136",
+      "4140",
+      "4153",
+      "4154",
+      "p104-literal-slot-defect",
+      "p104-uniform-commutation-gap",
+      "p104-uniform-encoding-gap",
+      "p104-source-chart-shortcut",
+      "p104-fiber-tag-seam"
+    ],
+    "existing_checks": [
+      {
+        "scope": "PR #4133: literal-slot defect",
+        "evidence": "research/p104-literal-slot-defect/README.md",
+        "remaining": "Independently validate the stated scope or change its producer/consumer premise and pay the complete boundary.",
+        "status": "author-checked",
+        "source_commit": "88666f1dee93fbd85b31fd5d6f91a219e111b5f4",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      },
+      {
+        "scope": "PR #4136: uniform commutation gap",
+        "evidence": "research/p104-uniform-commutation-gap/README.md",
+        "remaining": "Independently validate the stated scope or change its producer/consumer premise and pay the complete boundary.",
+        "status": "author-checked",
+        "source_commit": "281ce601a8be9db470de630759179c23e1dfaa88",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      },
+      {
+        "scope": "PR #4140: uniform encoding gap",
+        "evidence": "research/p104-uniform-encoding-gap/README.md",
+        "remaining": "Independently validate the stated scope or change its producer/consumer premise and pay the complete boundary.",
+        "status": "author-checked",
+        "source_commit": "de25c6525aa866962309df5d107b32d55e6e5628",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      },
+      {
+        "scope": "PR #4153: source-chart shortcut exhausts its optimistic allowance",
+        "evidence": "research/p104-source-chart-shortcut/README.md",
+        "remaining": "Independently validate the stated scope or change its producer/consumer premise and pay the complete boundary.",
+        "status": "author-checked",
+        "source_commit": "9123241fea53df77e51ffe30d9aca1625d4bf3a8",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      },
+      {
+        "scope": "PR #4154: fiber-tag seam mismatch",
+        "evidence": "research/p104-fiber-tag-seam/README.md",
+        "remaining": "Independently validate the stated scope or change its producer/consumer premise and pay the complete boundary.",
+        "status": "author-checked",
+        "source_commit": "4ffcdaee5d6255a9cea1fe53a5424435068089d0",
+        "selected_record_sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+      }
+    ]
+  },
+  "scope_status": "refined",
+  "actual_scope": "Independently validate #4154's fiber-tag seam claims from #3966's labels and corrections and #3954's witness lines, in the sources' orientation.",
+  "scope_change_reason": "The card lists #4154 as needing independent validation of its stated scope; this does that and no new channel.",
+  "bottleneck_link": "#4154's helper-mismatch conclusion feeds the P104 side-correction channel; it must hold in the sources' convention before later work builds on it.",
+  "budget_minutes": 30,
+  "stop_rule": "Stop once every #4154 claim is confirmed or corrected in the source orientation.",
+  "outcome": "completed",
+  "evidence": "verify.py: labels equal #3966's source array; census 10,816 pairs, 0 full-address agreements, shared line 5,440/5,376, min Hamming 6 at labels 4 and 32, witness Hamming 20; in #3966's orientation the witness flips are (1,0), the uniform tag matches the row tag on the shared line, cell (2,1)=bit 19 separates; the helper match depends on the row label's class; the transposed control reproduces #4154's 0/1, helper mismatch and bit 11.",
+  "author": "Ryanson Jonathan, with Claude Code (Claude Opus 5.5) assistance",
+  "ai_assistance": "Substantial Claude Code (Claude Opus 5.5) assistance in reading, verifier implementation and writing; no independent human or proof-kernel review.",
+  "open_questions": [
+    {
+      "question": "Restate #4154's open question 2 for the full address alone.",
+      "why_it_matters": "In the source orientation the uniform tag already agrees with the witness row tag on the shared line.",
+      "first_step": "Rewrite the question without the shared-line premise.",
+      "blocked_by": "None."
+    }
+  ]
+}
diff --git a/research/p104-fiber-tag-orientation-check/VERIFY.json b/research/p104-fiber-tag-orientation-check/VERIFY.json
new file mode 100644
index 000000000..16b76ac01
--- /dev/null
+++ b/research/p104-fiber-tag-orientation-check/VERIFY.json
@@ -0,0 +1,25 @@
+{
+  "schema_version": 1,
+  "status": "ready",
+  "python": "3.12",
+  "requirements": [],
+  "inputs": [
+    "research/p104-fiber-tag-seam/README.md",
+    "research/p104-fiber-tag-seam/verify.py",
+    "research/p104-phase-uniform-gauge/PROOF.md",
+    "research/p104-phase-uniform-gauge/source/RESULT.json",
+    "research/p104-two-axis-compatibility/PROOF.md",
+    "research/p104-two-axis-compatibility/RESULT.json"
+  ],
+  "commands": [
+    {
+      "argv": [
+        "{python}",
+        "-B",
+        "{package}/verify.py"
+      ],
+      "timeout_seconds": 120,
+      "checks": "Pinned #3966 source RESULT.json and #4154 labels; #3966 correction text; #3954 witness lines decoded as 386(x)386 and 405(x)386; gauge facts on F2^9; orientation-free census; witness in source orientation; row-label helper rule; transposed control reproducing #4154's numbers; #4154 code line."
+    }
+  ]
+}
diff --git a/research/p104-fiber-tag-orientation-check/verify.py b/research/p104-fiber-tag-orientation-check/verify.py
new file mode 100644
index 000000000..ccc1fc313
--- /dev/null
+++ b/research/p104-fiber-tag-orientation-check/verify.py
@@ -0,0 +1,163 @@
+#!/usr/bin/env python3
+"""Independent check of #4154's P104 fiber-tag seam, in the orientation of its sources.
+
+#3966 (PROOF.md section 4) defines the row correction z (x) q_j and the column correction r_i (x) z,
+with q_j = O p_j for the row label and r_i = O p_i for the column label. #3954's witness lines are
+386 (x) 386 (row port) and 405 (x) 386 (donor port), so its pair is row label 386, column label 405,
+shared line r_i (x) q_j. Tensors here are sets of (alpha, beta) cells, not packed integers; no code
+from #4154. Standard library only.
+Ryanson Jonathan with Claude Code (Claude Opus 5.5) assistance. Apache-2.0.
+"""
+import hashlib
+import json
+from pathlib import Path
+import sys
+
+ROOT = Path(__file__).resolve().parents[2]
+GAUGE = ROOT / "research/p104-phase-uniform-gauge"
+SEAM = ROOT / "research/p104-fiber-tag-seam"
+TWO_AXIS = ROOT / "research/p104-two-axis-compatibility"
+PINS = {
+    GAUGE / "source/RESULT.json": "ce05a27ebc98677bec4febaa488fa03f5b6fc60216aa43b34298444d12f7ab6c",
+    SEAM / "labels.json": "7227a7d6bdb49f36b6e401478034cc542eb887979890c8deda924b8ac0844209",
+}
+Z, ONES = 420, 511
+
+
+def need(condition, message):
+    if not condition:
+        raise AssertionError(message)
+
+
+def bits(x):
+    return [(x >> k) & 1 for k in range(9)]
+
+
+def dot(x, y):
+    return sum(a * b for a, b in zip(bits(x), bits(y))) % 2
+
+
+def gauge(x):
+    """O x = x + (x . z) z"""
+    return x ^ Z if dot(x, Z) else x
+
+
+def cells(u, v):
+    """u (x) v as its set of nonzero cells (alpha, beta), with entry u_alpha v_beta"""
+    return frozenset((a, b) for a, ua in enumerate(bits(u)) for b, vb in enumerate(bits(v)) if ua and vb)
+
+
+def add(*tensors):
+    out = set()
+    for t in tensors:
+        out ^= set(t)
+    return frozenset(out)
+
+
+def pair(s, t):
+    return len(s & t) % 2
+
+
+def tags(row_label, col_label):
+    q, r = gauge(row_label), gauge(col_label)
+    return cells(Z, q), cells(r, Z), cells(r, q)      # row tag, column tag, shared line
+
+
+def census(labels, tag_fn):
+    out = dict(full_equal=0, agree=0, disagree=0, helper_row_match=0, min_ham=99, min_pairs=set())
+    a = add(cells(Z, ONES), cells(ONES, Z))
+    for j in labels:            # row label
+        for i in labels:        # column label
+            row, col, line = tag_fn(j, i)
+            need(pair(line, line) == 1, "shared line is nondegenerate")
+            need(a != row and a != col, "uniform tag is never a fiber tag")
+            out["full_equal"] += row == col
+            same = pair(row, line) == pair(col, line)
+            out["agree" if same else "disagree"] += 1
+            out["helper_row_match"] += pair(a, line) == pair(row, line)
+            ham = len(row ^ col)
+            if ham < out["min_ham"]:
+                out["min_ham"], out["min_pairs"] = ham, set()
+            if ham == out["min_ham"]:
+                out["min_pairs"] |= {j, i}
+    return out
+
+
+def main():
+    for path, sha in PINS.items():
+        need(hashlib.sha256(path.read_bytes()).hexdigest() == sha, f"pinned {path.name}")
+    labels = json.loads((GAUGE / "source/RESULT.json").read_text())["physical_labels"]
+    need(json.loads((SEAM / "labels.json").read_text()) == labels, "#4154's labels.json is #3966's source array")
+    need(len(set(labels)) == 104 and all(0 < x < 512 for x in labels), "104 distinct labels in F2^9")
+    proof = (GAUGE / "PROOF.md").read_text()
+    need("The row correction is `z tensor q_j`; the column correction is `r_i tensor z`." in proof, "#3966 row/column corrections")
+    need("q_j=Op_j" in proof and "r_i=Op_i" in proof, "#3966 images")
+
+    # #3954's witness lines fix the orientation: row fiber q = p_j = 386, donor column p_i = 405
+    w = json.loads((TWO_AXIS / "RESULT.json").read_text())["witness"]
+    def as_cells(n):
+        return frozenset((k // 9, k % 9) for k in range(81) if (n >> k) & 1)
+    need(as_cells(w["u_row"]) == cells(386, 386) and as_cells(w["u_donor"]) == cells(405, 386), "#3954 witness lines are 386(x)386 and 405(x)386")
+    need((w["p_row"], w["p_donor"], w["q"]) == (386, 405, 386), "#3954 witness labels")
+    ROW, COL = w["q"], w["p_donor"]
+
+    # gauge facts, on all of F2^9
+    need(dot(Z, Z) == 0 and bin(Z).count("1") == 4, "z even of weight 4")
+    for x in range(512):
+        need(gauge(gauge(x)) == x and dot(gauge(x), Z) == dot(x, Z), "O involution, z . Ox = z . x")
+        need((bin(gauge(x)).count("1") - bin(x).count("1") - 2 * dot(x, Z)) % 4 == 0, "weight law")
+        for y in range(x, 512):
+            if dot(gauge(x), gauge(y)) != dot(x, y):
+                raise AssertionError("O orthogonal")
+    classes = [bin(x).count("1") % 4 for x in labels]
+    need(classes.count(1) == 56 and classes.count(3) == 48, "56 class-1 and 48 class-3 labels")
+    need(all(bin(gauge(x)).count("1") % 4 == 3 for x in labels), "every image has weight class 3")
+
+    # orientation-free claims of #4154, in the source orientation
+    src = census(labels, tags)
+    need(src["full_equal"] == 0, "no full-address agreement")
+    need((src["agree"], src["disagree"]) == (5440, 5376), "shared-line split 5440 / 5376")
+    need(src["min_ham"] == 6 and src["min_pairs"] == {32, 4}, "minimum Hamming 6, only labels 32 and 4")
+    need(src["helper_row_match"] == 104 * 48, "helper matches the row tag on 104 * 48 ordered pairs")
+    # ... and it is the ROW label's class that decides it: a.line - row.line = z . q_j
+    a = add(cells(Z, ONES), cells(ONES, Z))
+    for j in labels:
+        for i in labels:
+            row, _, line = tags(j, i)
+            need((pair(a, line) == pair(row, line)) == (bin(j).count("1") % 4 == 3), "helper/row match iff row label class 3")
+
+    # the witness, in the source orientation
+    row, col, line = tags(ROW, COL)
+    need((gauge(ROW), gauge(COL)) == (386, 49), "witness images 386 and 49")
+    flips = (pair(row, line), pair(col, line))
+    need(flips == (1, 0), "source orientation: witness shared-line flips are (row 1, column 0)")
+    need(pair(a, line) == flips[0], "source orientation: the uniform tag agrees with the witness row tag on the shared line")
+    need(len(row ^ col) == 20, "witness Hamming distance 20")
+    need((1, 2) not in row ^ col and (2, 1) in row ^ col, "cell (1,2) = bit 11 does not separate; cell (2,1) = bit 19 does")
+    need(a != row, "the uniform tag still differs from the row tag on the full address")
+
+    # control: #4154's code builds (z (x) O p_col, O p_row (x) z, O p_row (x) O p_col), the transposes
+    def transposed(j, i):
+        q, r = gauge(j), gauge(i)
+        return cells(Z, r), cells(q, Z), cells(q, r)
+    t = census(labels, transposed)
+    need({k: t[k] for k in ("full_equal", "agree", "disagree", "min_ham")} == {k: src[k] for k in ("full_equal", "agree", "disagree", "min_ham")}, "control: the census is orientation-free")
+    row_t, col_t, line_t = transposed(ROW, COL)
+    need((pair(row_t, line_t), pair(col_t, line_t)) == (0, 1) and pair(a, line_t) == 1, "control: transposed orientation reproduces #4154's flips 0/1 and its helper mismatch")
+    need((1, 2) in row_t ^ col_t, "control: transposed orientation reproduces #4154's bit 11")
+    seam_code = (SEAM / "verify.py").read_text()
+    need("return tensor(Z, column_image), tensor(row_image, Z)" in seam_code, "#4154's row tag is z (x) column image")
+
+    print(json.dumps({
+        "status": "PASS_P104_FIBER_TAG_ORIENTATION_CHECK",
+        "confirmed": {"pairs": 104 * 104, "full_equal": 0, "shared_agree": 5440, "shared_disagree": 5376,
+                      "min_hamming": 6, "min_labels": [4, 32], "witness_hamming": 20, "helper_row_matches": 104 * 48},
+        "corrected_at_witness_row386_col405": {"shared_flips_row_col": list(flips), "uniform_tag_matches_row_on_shared_line": True,
+                                                "separating_cell": [2, 1], "bit": 19},
+        "corrected_census_wording": "the uniform tag matches the row tag on the shared line iff the ROW label is class 3",
+        "new_kappa_claimed": False,
+    }, indent=1))
+
+
+if __name__ == "__main__":
+    sys.exit(main())