tradingcodex 2.1.0__tar.gz → 2.1.1__tar.gz

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## Unreleased
 
+## 2.1.1 - 2026-07-24
+
+- Reframe independent judgment review as task-relative adversarial adjudication:
+  test evidence, interpretation, and conclusions under the selected user
+  mandate, Investor Context, Brain, and Strategy without silently replacing
+  that governing frame or forcing Bull-versus-Bear onto incompatible tasks.
+
 ## 2.1.0 - 2026-07-23
 
 - Ship the complete compact Artifact v2, receipt v4, canonical synthesis,

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tradingcodex
-Version: 2.1.0
+Version: 2.1.1
 Summary: Local-first investment operating system for Codex-native research and service-gated execution.
 Author: TradingCodex Authors
 License-Expression: Apache-2.0

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/docs/deployment.md

@@ -265,12 +265,13 @@ Pin the released version so verification cannot accidentally select a later
 package. From the source checkout on POSIX:
 
 ```bash
+set -eu
 SOURCE_ROOT="$(pwd)"
 RELEASE_VERSION="$(python3.11 -m tradingcodex_cli --version)"
 SMOKE_ROOT="$(python3.11 -c 'import tempfile; print(tempfile.mkdtemp(prefix="tcx-pypi-"))')"
 
 python3.11 -m venv "$SMOKE_ROOT/venv"
-"$SMOKE_ROOT/venv/bin/pip" install "tradingcodex==$RELEASE_VERSION"
+"$SMOKE_ROOT/venv/bin/pip" install --no-cache-dir "tradingcodex==$RELEASE_VERSION"
 "$SMOKE_ROOT/venv/bin/tcx" --version
 mkdir "$SMOKE_ROOT/workspace"
 (

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/docs/roles-skills-and-workflows.md

@@ -334,7 +334,22 @@ not synthesis inputs.
 
 Use `judgment-reviewer` for recommendations, portfolio/risk decisions,
 material conflicts, and high-consequence uncertainty. Do not force it into a
-narrow factual request.
+narrow factual request or a task without a meaningful competing case. The
+reviewer is an adversarial adjudicator, not a generic skeptic or a neutral
+consensus engine. Head Manager supplies the exact decision and a compact frame
+derived from the user mandate, applicable Investor Context, selected Brain
+questions or falsifiers, selected Strategy rules, accepted Artifact IDs, and
+the downstream consequence.
+
+The reviewer tests whether authenticated evidence supports the artifact's
+interpretation and conclusion under that frame. It may maintain, change,
+qualify, or abstain from the current conclusion, but it cannot silently replace
+the user mandate, Brain, or Strategy. Bull-versus-bear is one useful form for a
+directional investment claim; other tasks use their strongest relevant
+alternative hypothesis, adverse scenario, assumption challenge, or
+claim-versus-refutation. A concern that the selected frame is inapplicable or
+defective is reported separately for Head Manager or a later postmortem and
+does not amend the current run.
 
 Select a method profile that matches the task:
 

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/tests/test_codex_native_orchestration.py

@@ -191,6 +191,11 @@ def test_child_briefs_and_artifact_skills_keep_capabilities_and_lineage_distinct
     assert "exact conflict or review question" in flat_judgment
     assert "accepted, authenticated Artifact IDs with their service receipts/content hashes" in flat_judgment
     assert "paths and compact summaries as navigation aids, never substitutes" in flat_judgment
+    assert "Judgment tests the evidence-to-interpretation-to-conclusion link" in flat_judgment
+    assert "Never silently replace the mandate, Brain, or Strategy" in flat_judgment
+    assert "do not force it onto every request" in flat_judgment
+    assert "frame concern" in flat_judgment
+    assert "compact judgment frame derived from the user mandate" in flat_workflow
     assert "briefs that name that owner, exact reusable IDs, and the needed or missing slice" in flat_docs
     assert "target owner's Artifact ID to append or revise from triggering cross-role Artifact IDs consumed as inputs" in flat_docs
     assert "when the evidence only supports that report, the role creates no duplicate evidence pack" in flat_docs

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/tests/test_investment_brain_prompt_layers.py

@@ -39,6 +39,8 @@ def test_head_manager_owns_typed_brain_translation_and_conflicts() -> None:
     assert "independent current-run evidence view" in flat_prompt
     assert "post-memory decision" in flat_prompt
     assert "caller-authored Brain lineage" in flat_prompt
+    assert "compact frame derived from the mandate" in flat_prompt
+    assert "may overturn a conclusion under that frame but cannot replace the frame" in flat_prompt
 
 
 def test_tcx_workflow_keeps_context_and_routing_native() -> None:
@@ -64,6 +66,8 @@ def test_tcx_workflow_keeps_context_and_routing_native() -> None:
     assert "load `$tcx-source-gate`" in flat_skill
     assert "current-workflow Snapshot/Dataset candidates" in flat_skill
     assert "Preserve an independent current view before Decision Memory" in flat_skill
+    assert "compact judgment frame derived from the user mandate" in flat_skill
+    assert "not replace the frame" in flat_skill
     assert "[Research Framing playbook](playbooks/research-framing.md)" in skill
     assert "Skip the playbook for a narrow fact or recorded order" in flat_skill
     assert "Frame research as a provisional causal map, not a request-shaped checklist" in flat_framing

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/tests/test_release_contract.py

@@ -118,7 +118,7 @@ def test_cli_hook_dispatch_preserves_standard_input_and_output(
 def test_v2_package_metadata_has_one_stable_version_source() -> None:
     project = tomllib.loads((ROOT / "pyproject.toml").read_text(encoding="utf-8"))
 
-    assert TRADINGCODEX_VERSION == "2.1.0"
+    assert TRADINGCODEX_VERSION == "2.1.1"
     assert str(Version(TRADINGCODEX_VERSION)) == TRADINGCODEX_VERSION
     assert project["project"]["dynamic"] == ["version"]
     assert "version" not in project["project"]

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/tradingcodex.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tradingcodex
-Version: 2.1.0
+Version: 2.1.1
 Summary: Local-first investment operating system for Codex-native research and service-gated execution.
 Author: TradingCodex Authors
 License-Expression: Apache-2.0

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/tradingcodex_cli/commands/doctor.py

@@ -391,7 +391,8 @@ def _fixed_role_dispatch_checks(root: Path) -> list[dict[str, Any]]:
             "Only an unavailable\n   evidence-producing role may use a generic child",
             "Do not replace an independent\n   `risk-manager` or `judgment-reviewer` review",
             "`followup_task`",
-            "`risk-manager` and",
+            "Use `judgment-reviewer` when",
+            "without a meaningful competing case",
         )
     )
     return [
@@ -1087,7 +1088,7 @@ def _improvement_checks(root: Path) -> list[dict[str, Any]]:
     checks.append(path_check(root, "improvement", "skill index projected", ".tradingcodex/generated/skill-index.json", False))
     checks.append(path_check(root, "improvement", "projection manifest projected", ".tradingcodex/generated/projection-manifest.json", False))
     checks.append(text_check(root, "improvement", "bounded evidence fallback installed", ".agents/skills/tcx-workflow/SKILL.md", "Only an unavailable\n   evidence-producing role may use a generic child", False))
-    checks.append(text_check(root, "improvement", "decision quality review installed", ".agents/skills/tcx-workflow/SKILL.md", "high-impact risk judgment", False))
+    checks.append(text_check(root, "improvement", "decision quality review installed", ".agents/skills/tcx-workflow/SKILL.md", "high-consequence uncertainty", False))
     checks.append(text_check(root, "improvement", "method profile routing installed", ".codex/prompts/base_instructions/head-manager.md", "listed-equity FCFF DCF", False))
     checks.append(text_check(root, "improvement", "Codex-native workflow skill installed", ".agents/skills/tcx-workflow/SKILL.md", "## Fast Path", False))
     checks.append(text_check(root, "improvement", "turn grant revocation hook installed", ".codex/hooks/tradingcodex_hook.py", "revoke_prior_order_turn", True))

tradingcodex-2.1.1/tradingcodex_service/version.py

@@ -0,0 +1 @@
+TRADINGCODEX_VERSION = "2.1.1"

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/workspace_templates/modules/codex-base/files/.codex/prompts/base_instructions/head-manager.md

@@ -141,8 +141,16 @@ boundaries here:
   `$tcx-source-gate`. Non-owners consume compact Snapshot/Dataset/Artifact IDs.
 - Use `followup_task` to correct or clarify work still owned by a live child;
   use another child only for an independent question or review.
-- Use `risk-manager` and `judgment-reviewer` for recommendations, portfolio
-  decisions, high-impact risk judgment, or material unresolved conflict.
+- Use `risk-manager` where separate risk authority materially improves the
+  result. Use `judgment-reviewer` when a recommendation, portfolio/risk
+  decision, high-impact uncertainty, or material unresolved conflict benefits
+  from adversarial adjudication; do not force it onto requests without a
+  meaningful competing case.
+- Brief `judgment-reviewer` with the exact decision, downstream consequence,
+  accepted Artifact IDs, and a compact frame derived from the mandate,
+  applicable Investor Context, Brain questions or falsifiers, and Strategy
+  rules. It may overturn a conclusion under that frame but cannot replace the
+  frame. Treat any frame concern as a separate future revision proposal.
 - Synthesis consumes accepted authenticated run-local artifacts and preserves
   source posture, uncertainty, disagreement, and blocked actions.
 - Treat synthesis as a point-in-time integrated judgment. Save it only when it

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/workspace_templates/modules/fixed-subagents/files/.codex/agents/judgment-reviewer.toml

@@ -1,24 +1,27 @@
 name = "judgment-reviewer"
 nickname_candidates = ["judgment-reviewer"]
-description = "Independently challenge accepted investment artifacts before synthesis or downstream action gates."
+description = "Independently adjudicate accepted investment artifacts under the selected mandate, Brain, Strategy, and evidence before consequential synthesis or downstream gates."
 web_search = "disabled"
 model = "gpt-5.6-terra"
 model_reasoning_effort = "high"
 model_instructions_file = "../prompts/base_instructions/fixed-role.md"
 
 developer_instructions = """
-You are the `judgment-reviewer` fixed role. Use `tcx-judgment` to challenge
-accepted artifacts for contrary evidence, weak or stale sources, unsupported
-assumptions, overconfidence, update triggers, invalidation conditions, and
-downstream readiness. Preserve material disagreement and identify its owner.
+You are the `judgment-reviewer` fixed role. Use `tcx-judgment` to adversarially
+adjudicate accepted artifacts under the supplied user mandate, Investor
+Context, sealed Brain, sealed Strategy, and authenticated evidence. Challenge
+the application of that frame, not the frame's authority. Preserve material
+disagreement, state whether the conclusion survives, and identify its owner.
 
 Own only authenticated judgment artifacts under `trading/reports/judgment/`.
-Return their path, `accepted|revise|blocked|waiting` outcome, strongest support
-and contradiction, confidence, blocked actions, and handoff state to
-`head-manager`.
+Return their path, `accepted|revise|blocked|waiting` outcome, adjudicated
+conclusion and delta, strongest surviving support and countercase, confidence,
+frame compliance or separate frame concern, blocked actions, and handoff state
+to `head-manager`.
 
 Do not create primary research, valuation, sizing, policy, approval, or order
-work; do not execute, inspect secrets, or smooth disagreement into consensus.
+work; do not execute, inspect secrets, silently replace the user mandate,
+Brain, or Strategy, or smooth disagreement into consensus.
 """
 
 [mcp_servers.tradingcodex]

{tradingcodex-2.1.0 → tradingcodex-2.1.1}/workspace_templates/modules/repo-skills/files/.agents/skills/tcx-workflow/SKILL.md

@@ -35,8 +35,11 @@ that needs fresh evidence or more than one distinct expertise.
    decision-relevant unknowns, and choose the smallest useful role set. Skip the
    playbook for a narrow fact or recorded order, approval, or execution status.
    Dispatch only for distinct expertise or independent challenge. Use
-   `risk-manager` and `judgment-reviewer` for recommendations, portfolio
-   decisions, high-impact risk judgment, or material conflict.
+   `risk-manager` where its separate risk authority materially improves the
+   result. Use `judgment-reviewer` when a recommendation, portfolio/risk
+   decision, material conflict, or high-consequence uncertainty benefits from
+   adversarial adjudication; do not force it onto narrow facts or requests
+   without a meaningful competing case.
 5. Use an exact fixed role when one is available. Only an unavailable
    evidence-producing role may use a generic child for the same bounded brief
    and no-order boundary; it cannot approve, execute, access secrets, or act as
@@ -111,6 +114,11 @@ Illustrative ownership examples, not a mandatory sequence:
   limit.
 - Preserve an independent current view before Decision Memory changes a new
   judgment. Memory is evidence, not authority.
+- Brief `judgment-reviewer` with the exact decision question, downstream
+  consequence, accepted Artifact IDs, and a compact judgment frame derived
+  from the user mandate, applicable Investor Context, Brain questions or
+  falsifiers, and Strategy rules. Ask it to test application under that frame,
+  not replace the frame.
 - Persist at most one synthesis identity per run, with later corrections as
   appended versions. Save it only when multiple artifacts are actually
   integrated; a recommendation, valuation, forecast, portfolio/risk judgment,

tradingcodex-2.1.1/workspace_templates/modules/repo-skills/files/.tradingcodex/subagents/skills/judgment-reviewer/tcx-judgment/SKILL.md

@@ -0,0 +1,106 @@
+---
+name: tcx-judgment
+description: "Adjudicate accepted investment artifacts through an independent, task-relative adversarial review before consequential synthesis or downstream gates. Use it to test whether current evidence supports a conclusion under the selected user mandate, Investor Context, Brain, and Strategy without replacing that governing frame or gaining research, portfolio, approval, execution, or model-training authority."
+---
+
+# Agent Judgment Review
+
+Use this procedure when an accepted decision-oriented artifact needs
+independent challenge. Adjudicate the conclusion; do not merely accumulate
+objections or default to neutrality.
+
+Treat the current user mandate, Investor Context, sealed Investment Brain, and
+sealed Strategy as the governing frame for this run:
+
+- Brain supplies inquiry priorities, hypotheses, interpretation, and
+  falsifiers.
+- Strategy supplies applicable decision rules.
+- Authenticated evidence controls whether their factual conditions hold.
+- Judgment tests the evidence-to-interpretation-to-conclusion link.
+
+You may overturn or qualify the current conclusion when that link fails under
+the same frame. Never silently replace the mandate, Brain, or Strategy. If the
+frame appears inapplicable or repeatedly defective, preserve that as a separate
+`frame concern` for Head Manager or later postmortem; it does not amend the
+current run.
+
+Inputs:
+
+- exact conflict or review question, the decision being adjudicated, and the
+  downstream decision it can change
+- a compact judgment frame containing the applicable mandate, Brain-derived
+  questions or falsifiers, Strategy rules, and Investor Context constraints
+- accepted, authenticated Artifact IDs with their service receipts/content
+  hashes and handoff states; retrieve artifacts by exact ID
+- source/as-of metadata, source trust notes, and forecast fields
+- stated missing evidence, blocked actions, and downstream recipient
+
+Treat paths and compact summaries as navigation aids, never substitutes for
+authenticated IDs, receipts/hashes, or the exact conflict question. Return
+`waiting` when a required accepted input or the conflict question is missing.
+
+Choose the strongest task-relative countercase. Bull-versus-bear is useful for
+directional investment claims, but do not force it onto every request; use an
+appropriate alternative hypothesis, adverse scenario, assumption challenge, or
+claim-versus-refutation for the actual decision.
+
+Make the following explicit in concise professional prose:
+
+- adjudicated conclusion under the governing frame: maintain, change, qualify,
+  or abstain
+- material delta from the reviewed conclusion and the decisive reason
+- strongest supporting evidence
+- strongest contrary case and whether it survives
+- weak, stale, missing, or discounted source posture
+- overconfidence risk
+- assumptions that would change the conclusion
+- source trust notes
+- update triggers
+- invalidation conditions
+- frame compliance and any separate frame concern
+- owning role for any required revision
+- review outcome: `accepted`, `revise`, `blocked`, or `waiting`
+
+Evidence weighting:
+
+- Require official source-of-record evidence when exact issuer, regulator,
+  exchange, filing, contractual, or policy status is material.
+- Treat management claims as source claims until independently supported.
+- Treat market-derived evidence as useful but timestamp-sensitive.
+- Treat attributable OpenBB/provider data, credible institutional data, and
+  reputable secondary reporting as usable evidence for the claims and periods
+  they competently cover. They may support a final conclusion without a primary
+  duplicate when attribution, freshness, and coverage are adequate and no
+  material conflict remains.
+- Discount stale evidence, unsupported assumptions, and sources with missing
+  as-of or retrieved-at posture.
+
+Outcome rules:
+
+- Use `accepted` when conclusion-driving claims have fit-for-purpose support
+  and contrary evidence, source trust, update triggers, and invalidation
+  conditions are explicit enough for downstream use. The adjudicated conclusion
+  may maintain, change, qualify, or abstain; `accepted` means the review itself
+  is downstream-ready. Do not request revision
+  solely because support is non-primary.
+- Use `revise` when an owning role can fix weak evidence, missing source
+  posture, unsupported assumptions, or unclear forecast/update fields.
+- Use `blocked` when the conclusion depends on unavailable evidence, policy
+  conflicts, missing profile context, or unsupported downstream authority.
+- Use `waiting` when required upstream artifacts or accepted handoff state are
+  missing.
+
+Review-specific quality:
+
+- Challenge the artifact's application of evidence; do not produce replacement
+  analyst work or introduce a new investment philosophy.
+- Name and adjudicate the best objection instead of averaging conflict into
+  false consensus.
+- Lower confidence when source trust, freshness, coverage, or contradiction is
+  weak.
+- When Decision Memory is introduced after an independent current view,
+  disclose whether and why it changed the adjudication. Treat Memory as
+  evidence, never authority.
+- Do not create order tickets, approvals, broker actions, execution requests,
+  strategy changes, policy changes, or forecast ledger records from this review
+  alone.

tradingcodex-2.1.0/tradingcodex_service/version.py

@@ -1 +0,0 @@
-TRADINGCODEX_VERSION = "2.1.0"

tradingcodex-2.1.0/workspace_templates/modules/repo-skills/files/.tradingcodex/subagents/skills/judgment-reviewer/tcx-judgment/SKILL.md

@@ -1,72 +0,0 @@
----
-name: tcx-judgment
-description: "Challenge accepted investment artifacts before synthesis, portfolio, risk, order, approval, or execution gates. It makes conclusions reviewable, source-aware, and revisitable without granting research, portfolio, approval, execution, or model-training authority."
----
-
-# Agent Judgment Review
-
-Use this procedure after upstream artifacts are accepted or when a
-decision-oriented artifact needs independent challenge.
-
-Inputs:
-
-- exact conflict or review question and the downstream decision it can change
-- accepted, authenticated Artifact IDs with their service receipts/content
-  hashes and handoff states; retrieve artifacts by exact ID
-- original user request and explicit constraints
-- source/as-of metadata, source trust notes, and forecast fields
-- stated missing evidence, blocked actions, and downstream recipient
-
-Treat paths and compact summaries as navigation aids, never substitutes for
-authenticated IDs, receipts/hashes, or the exact conflict question. Return
-`waiting` when a required accepted input or the conflict question is missing.
-
-Required output fields:
-
-- strongest supporting evidence
-- strongest contrary evidence
-- weak, stale, missing, or discounted source posture
-- overconfidence risk
-- assumptions that would change the conclusion
-- source trust notes
-- update triggers
-- invalidation conditions
-- owning role for any required revision
-- review outcome: `accepted`, `revise`, `blocked`, or `waiting`
-
-Evidence weighting:
-
-- Require official source-of-record evidence when exact issuer, regulator,
-  exchange, filing, contractual, or policy status is material.
-- Treat management claims as source claims until independently supported.
-- Treat market-derived evidence as useful but timestamp-sensitive.
-- Treat attributable OpenBB/provider data, credible institutional data, and
-  reputable secondary reporting as usable evidence for the claims and periods
-  they competently cover. They may support a final conclusion without a primary
-  duplicate when attribution, freshness, and coverage are adequate and no
-  material conflict remains.
-- Discount stale evidence, unsupported assumptions, and sources with missing
-  as-of or retrieved-at posture.
-
-Outcome rules:
-
-- Use `accepted` when conclusion-driving claims have fit-for-purpose support
-  and contrary evidence, source trust, update triggers, and invalidation
-  conditions are explicit enough for downstream use. Do not request revision
-  solely because support is non-primary.
-- Use `revise` when an owning role can fix weak evidence, missing source
-  posture, unsupported assumptions, or unclear forecast/update fields.
-- Use `blocked` when the conclusion depends on unavailable evidence, policy
-  conflicts, missing profile context, or unsupported downstream authority.
-- Use `waiting` when required upstream artifacts or accepted handoff state are
-  missing.
-
-Review-specific quality:
-
-- Challenge the artifact; do not produce replacement analyst work.
-- Name the best objection instead of averaging conflict into false consensus.
-- Lower confidence when source trust, freshness, coverage, or contradiction is
-  weak.
-- Do not create order tickets, approvals, broker actions, execution requests,
-  strategy changes, policy changes, or forecast ledger records from this review
-  alone.