Expose stale catalogue selection context without changing frozen pickers (no κ)

AGENTS.md

@@ -7,9 +7,10 @@ before using the task picker; its coverage fields identify prior checks and
 remaining obligations. Explicit user tasks take precedence over that default
 assignment; do not reroll an already assigned task.
 
-For a fresh assignment, run `python3 -B scripts/choose_current_research_task.py`.
-This entry point resolves generic baseline commands and retains each frozen
-card beside the current command. The historical picker and root guide stay
+For a fresh assignment, run `python3 -B scripts/choose_contextual_research_task.py`.
+This entry point reports whether the catalogue's assessment context matches the
+selected record, then retains the existing current-command routes and frozen
+cards. The historical pickers and root guide stay
 unchanged because research packages pin their bytes. The current command
 examples are in [the standing procedure](docs/research/agent-procedure.md).
 

docs/research/agent-procedure.md

@@ -102,15 +102,15 @@ Run this standard-library helper:
 
 ```sh
 # Python 3.10 or later; no third-party dependencies.
-python3 scripts/choose_current_research_task.py
+python3 scripts/choose_contextual_research_task.py
 # To reproduce a selection, pass the emitted 32-hex-digit nonce:
-python3 scripts/choose_current_research_task.py --seed YOUR_NONCE
+python3 scripts/choose_contextual_research_task.py --seed YOUR_NONCE
 # Expertise or an explicit user preference may override the allocation:
-python3 scripts/choose_current_research_task.py --track complex-mechanisms
+python3 scripts/choose_contextual_research_task.py --track complex-mechanisms
 # Retrieve a specific card after assignment:
-python3 scripts/choose_current_research_task.py --task connector-alphabet-change
+python3 scripts/choose_contextual_research_task.py --task connector-alphabet-change
 # Optional: surface structured handoffs/claims from an existing overlap cache.
-python3 scripts/choose_current_research_task.py --coordination-cache /tmp/overlap.json
+python3 scripts/choose_contextual_research_task.py --coordination-cache /tmp/overlap.json
 ```
 
 It reads [the task catalogue](agent-task-catalog.json), draws a
@@ -121,6 +121,14 @@ that a task is open. A deterministic seed is a reproducibility aid, not novelty.
 Each card supplies a first-step command, expected evidence, a next experiment,
 its connection to the bottleneck, allowed variations and scope-change triggers.
 
+The contextual entry point delegates selection and command routing to the
+unchanged pickers, then adds `catalog_selection`: the original catalogue context
+and its κ/source/review comparison with `selected`. A mismatch means the
+assessment predates or differs from the selection; missing fields are unknown.
+This does not assess whether a task remains useful or authenticate the scientific
+closure. Use `selected` for the current review and replay. See the
+[source-pinned discrepancy and controls](priority-selection-context.md).
+
 **Resolve the generic orientation command before running a historical card.**
 If its `first_step.command` is `python3 -B scripts/research_baseline.py`, use the
 current orientation command in [current priorities](current-priorities.md).

docs/research/priority-selection-context.md

@@ -0,0 +1,134 @@
+# Catalogue assessment versus the selected result
+
+The new contextual task CLI exposes the catalogue's original `selection_context`
+as `catalog_selection.context` and compares its κ, source commit and review
+pointer with the selected record. It delegates to the unchanged current picker.
+This resolves an onboarding discrepancy, not a mathematical claim or a new
+research-priority assessment.
+
+At base **`7c49fe32c55de02c2c873655c804b16e4c3fef0a`**, revision 8 of the
+catalogue and [priority guide](current-priorities.md) name PR5599. The
+[selected record](../../certificates/selected-result.json) and its
+[application review](pr-intake-evidence/pr6030-audit/PROOF-REVIEW.md) select
+PR6030. The old current CLI emitted the latter under `selected` but omitted the
+catalogue's assessment context altogether, while linking the former guide.
+
+| Field | Catalogue assessment | Selected record |
+| --- | --- | --- |
+| κ | `62009/20000000` = .00310045 | `781451/250000000` = .003125804 |
+| Source commit | `09dd0965711ca7eeb169057a420abab1c9052137` | `772dd10a5859a1d15f4851feb371519789213ad8` |
+| Review directory | `pr5599-completion` | `pr6030-audit` |
+
+The source comparison uses the original `audited_source_head`, rather than the
+separate portability-repair head. It does not infer a primary source from the
+order of `source_prs`, nor from a larger number in an open PR title.
+
+## Contract and reproduction
+
+```sh
+python3 -B scripts/choose_contextual_research_task.py --task scope-navigation --seed 00000000000000000000000000000063
+```
+
+On the pinned inputs, expect `catalog_selection.matches_selected: false`, with
+exact old/new values under `differences.kappa`, `differences.source_commit` and
+`differences.review`, and no `uncompared_fields`. `selected.reproduce` remains
+`make frontier6030-verify`. The 32-digit picker seed is 99 in hexadecimal; it
+only makes this explicitly chosen card's output reproducible.
+
+Equal rational representations compare equal. A changed review or source at
+the same κ still differs. Missing comparison fields produce `null` (unknown)
+when no known difference exists; a known difference still produces `false`.
+The exact historical context is retained, including its own dated scope text.
+
+Before comparison, both reread JSON byte strings must match the hashes in the
+child picker's receipt. A change between the child and wrapper reads fails
+without emitting a successful JSON receipt. This detects mixed snapshots at
+that boundary; it is not an atomic filesystem snapshot or authentication of
+the full selected construction. A stable checkout remains required.
+
+Both existing pickers, the catalogue, priority guide and selected record retain their
+bytes. Existing task choices, methods, scope keys, coverage flags and command
+routes are preserved. The contextual CLI adds only `catalog_selection`; existing
+consumers can keep their original entry point and output contract. Matching
+the three metadata fields establishes neither task novelty nor current coverage.
+
+## Verification and source pins
+
+The following command checks the actual discrepancy fixture and controls for
+same-κ provenance changes, rational equivalence, missing context, malformed
+metadata, changed-between-reads inputs and the real child/wrapper CLI:
+
+```sh
+python3 -B -m unittest discover -s tests -p 'test_research_priority_context.py' -v
+```
+
+Expected: **10 tests, OK**. The adjacent command is
+`python3 -B -m unittest discover -s tests -p 'test_research_current_assignment*.py' -v`
+(15 tests). Both pass normally and with `-O -W error` on Linux, CPython 3.12.14.
+The real original/contextual receipt for the command above is identical after removing
+the added field. The existing controls preserve all live cards, frozen generic
+and historical routes, help/error propagation and sparse-input distinctions.
+The 27 onboarding and nine coordination tests also pass. The final replay
+dependency plan selects no scientific packages; the new code has no scientific
+runtime dependency.
+
+The unchanged #6179 conformance replay also passes with 22 live cards, two
+historical route fixtures and seven rejection controls:
+`python3 -I -W error -B research/current-picker-lazy-validation/verify.py`.
+That replay reconstructs the current picker's exact predecessor source, which
+is why this contribution uses a separate entry point. The historical timing
+receipt is checked for conformance, not rebenchmarked; no speed claim is made
+for the contextual CLI's extra subprocess.
+
+Immutable SHA-256 pins at the exact base above:
+
+| Repository path | SHA-256 |
+| --- | --- |
+| `docs/research/agent-task-catalog.json` | `ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81` |
+| `docs/research/current-priorities.md` | `70f61c8d140ae6e892f2b8f065bba4eae3e76f7a8f55f632fd4e2f9f34039bdc` |
+| `certificates/selected-result.json` | `9b8d0eea95b3bf0dfcb7938a732a6439b87a2685dec2a5ab9e525b3b34debbfa` |
+| `scripts/choose_research_task.py` | `484f553c488c6f427ca81f19f3ee2720399dcc28ce61a3886d50bdf28ca72d12` |
+| `scripts/choose_current_research_task.py` (unchanged) | `04e194b14a1f4157be5be83e8c13995d69c3efe8996773d918acba6bfcabb471` |
+
+Only Python's standard library is used. The selected full cold replay, native
+maps, compiler, precision/tape and analytic recovery are outside this metadata
+repair. The five selected assumptions and κ remain unchanged.
+
+## Assignment, overlap and open next question
+
+OS randomness (`secrets.randbelow(100)`) yielded assignment seed **99**:
+research consolidation. Three candidates were considered: this catalogue/record
+discrepancy; source-certificate pointer coverage (current inputs agree, with
+the broader predecessor #1864); and ledger/status navigation (nearby work
+#3695, #4138 and #6050). The chosen bounded experiment was to reproduce and
+expose this exact discrepancy while preserving task selection. Stop after the
+three-field comparison and its failure controls; no global survey or search for
+a stronger bound is part of this result.
+
+The local claim was recorded at 2026-11-09 17:14:30 UTC, with an expected finish
+by 17:47:48 UTC. GitHub reported no pinned issue. The start-here discussion,
+last 20 handoff comments, 25 recently active discussions and relevant open and
+closed PR searches were inspected before implementation. No matching repair
+was found in that bounded coverage; this is not a proof of novelty.
+
+This extends the exact-output boundary in Ali Romman's #5378 and Mihir
+Srivastava's #5520, preserving Nick Petalas's #6179 source-bound conformance
+and benchmark. The omission of `selection_context` was already observed in
+#4906's startup audit; that work repaired the generic command route. This
+change adds source/claim reconciliation and unknown-context handling rather
+than claiming discovery of the earlier omission. No pending construction is a
+dependency.
+
+**Open next question:** should the next maintainer priority assessment revise
+the catalogue's tasks as well as its selected context? The comparison now makes
+the old assessment visible, but cannot establish whether its advice remains
+applicable. First inspect each affected card's current source-specific coverage
+and live handoffs; a substantive strategic reassessment and dependent-input
+review are required before rewriting frozen catalogue bytes.
+
+Prepared by **Sricharan Vagvala (@theon199)** with substantial **OpenAI Codex**
+assistance. Credit Douglas Colkitt, Ali Romman, Mihir Srivastava, Nick Petalas,
+ToikY and the
+retained picker/test contributors; DreamingOfClouds and the selected source's
+[contributor map](community-p1046030-contributors.md) retain all construction
+credit. Apache-2.0. No independent human or proof-kernel review is claimed.

scripts/choose_contextual_research_task.py

@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+"""Show catalogue assessment drift alongside the current task-picker receipt.
+
+The frozen current and legacy pickers retain their source and behavior.
+Sricharan Vagvala, with substantial OpenAI Codex assistance. Apache-2.0.
+"""
+from fractions import Fraction
+from hashlib import sha256
+import json
+from pathlib import Path
+import subprocess
+import sys
+
+ROOT = Path(__file__).resolve().parents[1]
+CURRENT = ROOT / "scripts/choose_current_research_task.py"
+
+
+def catalog_selection_context(result, root):
+    """Compare declared context, without interpreting it as scientific coverage."""
+    documents = []
+    for name, digest_key in (
+        ("docs/research/agent-task-catalog.json", "catalog_sha256"),
+        ("certificates/selected-result.json", "selected_record_sha256"),
+    ):
+        raw = (root / name).read_bytes()
+        if sha256(raw).hexdigest() != result[digest_key]:
+            raise ValueError("input changed since task selection: " + name)
+        documents.append(json.loads(raw))
+    catalog, selected = documents
+    context = catalog.get("selection_context")
+    if context is not None and not isinstance(context, dict):
+        raise ValueError("catalogue selection_context must be an object")
+    differences, missing = {}, []
+    for field, selected_field in (
+        ("kappa", "kappa"), ("source_commit", "audited_source_head"),
+        ("review", "review"),
+    ):
+        before = (context or {}).get(field)
+        after = selected.get(selected_field)
+        if before is None or after is None:
+            missing.append(field)
+            continue
+        if not all(isinstance(value, str) and value.strip() for value in (before, after)):
+            raise ValueError("invalid selection context field: " + field)
+        if field == "kappa":
+            try:
+                equal = Fraction(before) == Fraction(after)
+            except (ValueError, ZeroDivisionError) as exc:
+                raise ValueError("invalid selection context kappa") from exc
+        else:
+            equal = before == after
+        if not equal:
+            differences[field] = {"catalog": before, "selected": after}
+    return {
+        "context": context,
+        "matches_selected": False if differences else (None if missing else True),
+        "differences": differences,
+        "uncompared_fields": missing,
+        "scope": (
+            "Compare only kappa, source_commit/audited_source_head and review. "
+            "Missing fields are unknown. The catalogue context is historical; "
+            "use selected for the current result and replay. Matching metadata "
+            "does not establish current task coverage or scientific validation. "
+            "Input hashes bind this comparison to the picker receipt, not the "
+            "complete selected source closure."
+        ),
+    }
+
+
+def main():
+    child = subprocess.run([sys.executable, "-B", str(CURRENT), *sys.argv[1:]],
+                           capture_output=True, text=True, encoding="utf-8")
+    if child.returncode or any(arg in ("-h", "--help") for arg in sys.argv[1:]):
+        sys.stdout.write(child.stdout)
+        sys.stderr.write(child.stderr)
+        return child.returncode
+    sys.stderr.write(child.stderr)
+    try:
+        result = json.loads(child.stdout)
+        result["catalog_selection"] = catalog_selection_context(result, ROOT)
+    except (OSError, ValueError, KeyError, TypeError) as exc:
+        sys.stderr.write("catalogue selection comparison failed: " + str(exc) + "\n")
+        return 2
+    print(json.dumps(result, indent=2))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

tests/test_research_coordination.py

@@ -121,7 +121,7 @@ def test_root_entry_and_priorities_links(self):
         self.assertIn('docs/research/current-priorities.md',root)
         self.assertIn('docs/research/agent-procedure.md',root)
         procedure=(ROOT/'docs/research/agent-procedure.md').read_text()
-        self.assertLess(procedure.index('**Before drawing a card**'),procedure.index('python3 scripts/choose_current_research_task.py'))
+        self.assertLess(procedure.index('**Before drawing a card**'),procedure.index('python3 scripts/choose_contextual_research_task.py'))
 
 
 if __name__=='__main__':unittest.main()