consequa-x 2.1.0__tar.gz → 2.1.2__tar.gz

{consequa_x-2.1.0 → consequa_x-2.1.2}/CHANGELOG.md

@@ -1,12 +1,45 @@
 # Changes
 
+## 2.1.2 - 20 October 2026
+
+A maintenance release; the API, exported records and trace schema are
+unchanged.
+
+- `ActionExplanation.diagnosis`, the one-line display label, is always the
+  diagnosis codes joined with ` / `. A timeout after commit no longer shows the
+  1.x label `BELIEF / RESPONSE DIVERGENCE`: `consequa-x demo` and its README now
+  print `ASSESSMENT_DIVERGENCE / ACKNOWLEDGEMENT_EFFECT_DIVERGENCE`, the codes
+  that explanations and reports already carried.
+
+## 2.1.1 - 19 October 2026
+
+A maintenance release; the API and trace schema are unchanged. Published to
+PyPI with registry hashes verified; see the
+[release record](output/release-2.1.1-2026-10-19/README.md). Installed checks
+passed on Windows Python 3.11 to 3.15, on one Alpine Linux QEMU guest with
+Python 3.14 and from a fresh PyPI install. macOS has never been tested.
+
+- The project's name is written ConseqaX: README, documentation, report title
+  and footer, the trace schema's title and command-line text. The package,
+  command and import names stay `consequa-x` and `consequa_x`.
+- Python 3.13, 3.14 and 3.15 are tested and declared (PyPI classifiers and
+  documentation), alongside 3.11 and 3.12.
+- Each framework integration was also run with a real model in a real agent
+  loop (`scripts/check_live_integrations.py`); the recorded attempts matched
+  the tool calls that ran.
+- New example `examples/two_part_outcome.py`: an `all_of` contract over a refund
+  and a ticket closure, read back from two SQLite files.
+
 ## 2.1.0 - 19 October 2026
 
 Additive: the roadmap features left out of 2.0.0 (verifier composition,
 framework integrations and OpenTelemetry). Schema `1.0` and the 2.0 API are
-unchanged, and the core still has no required runtime dependency. Checked
-before publication; see the
-[release record](output/release-2.1.0-2026-10-19/README.md).
+unchanged, and the core still has no required runtime dependency. Published
+to PyPI with registry hashes verified; see the
+[release record](output/release-2.1.0-2026-10-19/README.md). Installed checks
+passed on Windows Python 3.11.9 and 3.12.10, on one Alpine Linux 3.24.2 QEMU
+guest with Python 3.14.8 and from a fresh PyPI install; the integrations passed against the installed wheel
+with each SDK present. macOS has never been tested.
 
 - Verifier composition. `all_of(contract_id, *parts)` and `any_of(...)` build
   one outcome contract from part contracts, each with its own predicate,
@@ -50,6 +83,8 @@ before publication; see the
   `PostToolUseFailure`). Tested offline with openai-agents 0.24.0,
   langchain-core 1.6.12, claude-agent-sdk 0.2.168 and opentelemetry 1.45.1; no
   live model was called.
+- Report tables scroll inside their frame on narrow screens; their cells no
+  longer collapse to one character per line (present since 2.0).
 
 ## 2.0.0 - 19 October 2026 (breaking)
 
@@ -264,7 +299,7 @@ preselected complete traces. The
 [derived findings](output/interaction-sampling-findings-2026-10-15/FINDINGS.md)
 retain coverage, width, detection and refusal denominators. The defined
 framework and protocol/software paper scope is complete on 16 October 2026:
-the [illustrated eight-page IEEE PDF](output/pdf/Consequa-X_IEEE_Paper.pdf) compiled and all
+the [illustrated eight-page IEEE PDF](output/pdf/ConseqaX_IEEE_Paper.pdf) compiled and all
 rendered pages passed visual review. The [QA receipt](output/pdf/qa.json) binds
 it to the frozen [editable source](paper/consequa-x-ieee.tex) and four vector
 figures. Actual venue selection, final human

{consequa_x-2.1.0 → consequa_x-2.1.2}/CONTRIBUTING.md

@@ -1,4 +1,4 @@
-# Contributing to Consequa-X
+# Contributing to ConseqaX
 
 Thanks for helping. This public repository holds the Python package, its tests
 and documentation, the website, and the research records behind the paper. The
@@ -23,7 +23,7 @@ to a pull request. Report vulnerabilities privately as described in
 
 ## Set up a development environment
 
-Fork and clone the repository, then create a branch for your change. Use Python 3.11 or 3.12:
+Fork and clone the repository, then create a branch for your change. Use Python 3.11 or newer:
 
 ```sh
 python -m venv .venv

{consequa_x-2.1.0 → consequa_x-2.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: consequa-x
-Version: 2.1.0
+Version: 2.1.2
 Summary: Evidence-aware explanation of consequential agent actions
 Author: Devrajsinh Jhala
 License-Expression: MIT
@@ -14,6 +14,9 @@ Classifier: Development Status :: 5 - Production/Stable
 Classifier: Programming Language :: Python :: 3 :: Only
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3.15
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -33,7 +36,7 @@ Provides-Extra: otel
 Requires-Dist: opentelemetry-api>=1.45; extra == "otel"
 Dynamic: license-file
 
-# Consequa-X
+# ConseqaX
 
 **Evidence-aware explanation of consequential agent actions.**
 
@@ -42,7 +45,7 @@ database write), the tool's response is only part of the story. A timeout can
 hide a refund that went through. A "success" can hide one that never landed.
 Another actor can produce the very outcome the agent was asked for.
 
-Consequa-X keeps four things apart: the **tool response**, what the **agent
+ConseqaX keeps four things apart: the **tool response**, what the **agent
 reported**, the outcome established by an **independent read-only readback**, and
 **who produced the effect**. Missing, stale or incomplete evidence stays
 `unknown` instead of being guessed.
@@ -166,7 +169,7 @@ receipt = verify(action, observation, verifier=MyVerifier())
 ```
 
 Your application owns source trust, scope, completeness, timing and provenance;
-Consequa-X checks that receipts bind to the supplied action, contract and
+ConseqaX checks that receipts bind to the supplied action, contract and
 observation, and keeps the evidence structured. The
 [artifact-delivery example](https://github.com/Devrajsinh-Jhala/consequa-x/blob/main/examples/artifact_delivery.py)
 implements a verifier with public APIs only, and the
@@ -252,8 +255,8 @@ and the [paper notes](https://github.com/Devrajsinh-Jhala/consequa-x/blob/main/p
 ## Platforms
 
 The package is pure Python. Released wheels have been install-tested on Windows
-with Python 3.11 and 3.12 and on one Alpine Linux virtual machine with Python
-3.14. macOS has not been tested yet. See the
+with Python 3.11, 3.12, 3.13, 3.14 and 3.15, and on one Alpine Linux virtual
+machine with Python 3.14. macOS has not been tested yet. See the
 [release notes](https://consequa-x.onrender.com/changelog.html) for exact
 environments.
 

{consequa_x-2.1.0 → consequa_x-2.1.2}/README.md

@@ -1,4 +1,4 @@
-# Consequa-X
+# ConseqaX
 
 **Evidence-aware explanation of consequential agent actions.**
 
@@ -7,7 +7,7 @@ database write), the tool's response is only part of the story. A timeout can
 hide a refund that went through. A "success" can hide one that never landed.
 Another actor can produce the very outcome the agent was asked for.
 
-Consequa-X keeps four things apart: the **tool response**, what the **agent
+ConseqaX keeps four things apart: the **tool response**, what the **agent
 reported**, the outcome established by an **independent read-only readback**, and
 **who produced the effect**. Missing, stale or incomplete evidence stays
 `unknown` instead of being guessed.
@@ -131,7 +131,7 @@ receipt = verify(action, observation, verifier=MyVerifier())
 ```
 
 Your application owns source trust, scope, completeness, timing and provenance;
-Consequa-X checks that receipts bind to the supplied action, contract and
+ConseqaX checks that receipts bind to the supplied action, contract and
 observation, and keeps the evidence structured. The
 [artifact-delivery example](https://github.com/Devrajsinh-Jhala/consequa-x/blob/main/examples/artifact_delivery.py)
 implements a verifier with public APIs only, and the
@@ -217,8 +217,8 @@ and the [paper notes](https://github.com/Devrajsinh-Jhala/consequa-x/blob/main/p
 ## Platforms
 
 The package is pure Python. Released wheels have been install-tested on Windows
-with Python 3.11 and 3.12 and on one Alpine Linux virtual machine with Python
-3.14. macOS has not been tested yet. See the
+with Python 3.11, 3.12, 3.13, 3.14 and 3.15, and on one Alpine Linux virtual
+machine with Python 3.14. macOS has not been tested yet. See the
 [release notes](https://consequa-x.onrender.com/changelog.html) for exact
 environments.
 

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/application-integration.md

@@ -1,6 +1,6 @@
 # Integrating an application
 
-Consequa-X explains captured application actions without using the refund
+ConseqaX explains captured application actions without using the refund
 simulator, calling a model, or executing a write. Register each action's target
 before running it, keep its responses and the agent's reported assessments, then
 obtain a separate readback observation. A response is evidence about the request;

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/design-2.0.md

@@ -1,4 +1,4 @@
-# Consequa-X 2.0 design
+# ConseqaX 2.0 design
 
 Status: released as 2.0.0 on 19 October 2026 (steps 1 to 6 below, except
 verifier composition). Verifier composition and step 7 were released in 2.1.0. Decided 19 October 2026, after

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/framework-integration.md

@@ -121,7 +121,14 @@ refund happened; readback does.
 Tested with openai-agents 0.24.0, langchain-core 1.6.12 and claude-agent-sdk
 0.2.168. The tests drive each SDK offline: the Agents SDK's `ScriptedModel` in a
 real `Runner` loop, real LangChain tools with callbacks, and the Claude Agent
-SDK's hook protocol through a scripted transport. No live model was called.
+SDK's hook protocol through a scripted transport. Each integration was also run
+once with a real model in a real agent loop
+([`scripts/check_live_integrations.py`](../scripts/check_live_integrations.py)):
+the OpenAI Agents SDK and LangChain with a local 4B Qwen model served by Ollama,
+the Claude Agent SDK with Claude Haiku. In each run the refund tool timed out
+after committing, the recorded attempts matched the calls that ran, and readback
+showed the refund the agent had reported as failed. One run per framework shows
+the wiring works; it does not cover every model or SDK release.
 
 ## Doing it by hand
 

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/migrating-to-2.0.md

@@ -1,4 +1,4 @@
-# Migrating to Consequa-X 2.0
+# Migrating to ConseqaX 2.0
 
 2.0 makes the breaking changes planned in the [2.0 design](design-2.0.md) once.
 Evidence discipline is unchanged: tool response, the agent's report, verified

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/offline-reports.md

@@ -1,6 +1,6 @@
 # Offline visual reports
 
-Consequa-X renders a dependency-free HTML report for a captured, validated
+ConseqaX renders a dependency-free HTML report for a captured, validated
 trace:
 
 ```console

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/opentelemetry.md

@@ -1,6 +1,6 @@
 # OpenTelemetry export and import
 
-`consequa_x.otel` (added in 2.1) moves Consequa-X traces through an
+`consequa_x.otel` (added in 2.1) moves ConseqaX traces through an
 OpenTelemetry pipeline, next to the rest of an agent's telemetry, and reads
 tool spans from other instrumentation back into a recorder. The OTLP/JSON
 functions need no dependency; `emit` needs `opentelemetry-api`
@@ -9,7 +9,7 @@ functions need no dependency; `emit` needs `opentelemetry-api`
 Tool spans follow the OpenTelemetry GenAI semantic conventions for
 `execute_tool` spans. Those conventions are still in *development* status
 upstream, so attribute names may change in later OpenTelemetry releases. Every
-Consequa-X finding uses its own `consequa_x.*` attributes.
+ConseqaX finding uses its own `consequa_x.*` attributes.
 
 ## What an export contains
 
@@ -77,7 +77,7 @@ command line:
 consequa-x otel-import spans.json --output-dir restored
 ```
 
-Spans written by other instrumentation hold no Consequa-X trace, but their tool
+Spans written by other instrumentation hold no ConseqaX trace, but their tool
 spans can still start one. `record_tool_spans` reads `execute_tool` spans into
 a `TraceRecorder` (which must use the default `unix_ms` time base). You decide
 which calls are consequential and what their outcome contract is:

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/product-roadmap.md

@@ -1,7 +1,7 @@
 # Product roadmap
 
 Agreed 18 October 2026, after publishing 1.1.1. This page tracks package
-features for users of Consequa-X; the research agenda stays in the
+features for users of ConseqaX; the research agenda stays in the
 [research roadmap](research-roadmap.md). Update the status column as work lands.
 
 ## Decision
@@ -113,7 +113,9 @@ or new research interfaces in the core package.
 during the 2.0 build, contrary to the maintainer's instruction to include every
 feature in the breaking release; the maintainer then asked for them to be built
 and released. They are additive, so they ship as 2.1.0 without another breaking
-change. With them, every feature in this roadmap is released.
+change. Released as 2.1.0 on 19 October 2026; see the
+[release record](../output/release-2.1.0-2026-10-19/README.md). With them, every
+feature in this roadmap is released.
 
 | ID | Feature | Status |
 | --- | --- | --- |

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/quickstart.md

@@ -1,10 +1,10 @@
 # Five-minute offline quickstart
 
-Use Python 3.11 or 3.12. No model, API key, provider account or runtime dependency
-is required. Version 2.1.0 is public on PyPI. Traces use schema 1.0.
+Use Python 3.11 or newer. No model, API key, provider account or runtime dependency
+is required. Version 2.1.2 is public on PyPI. Traces use schema 1.0.
 
 ```console
-python -m pip install consequa-x==2.1.0
+python -m pip install consequa-x==2.1.2
 consequa-x --version
 ```
 
@@ -89,7 +89,7 @@ The schema validates structure and identity linkage. Source authentication,
 transaction guarantees and the truth of adapter-provided metadata remain your
 application's responsibility. Unknown evidence must stay unknown.
 
-Traces use schema 1.0. Convert traces saved by Consequa-X 1.x with
+Traces use schema 1.0. Convert traces saved by ConseqaX 1.x with
 consequa-x migrate OLD.json --output NEW.json. Research replay
 interfaces have separate assumptions; no baseline superiority or live-model
 accuracy improvement is claimed. Platform validation is listed in release notes

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/release-notes.md

@@ -1,5 +1,31 @@
 # Release notes
 
+## 2.1.2 — 20 October 2026
+
+- The demo's timeout-after-commit case now prints its diagnosis codes,
+  ASSESSMENT_DIVERGENCE and ACKNOWLEDGEMENT_EFFECT_DIVERGENCE, instead of an
+  old label from version 1. Reports and saved explanations already showed these
+  codes.
+- Nothing else changed: saved traces, explanations and code written for 2.1
+  work as before.
+
+## 2.1.1 — 19 October 2026
+
+- The project is now written ConseqaX. You still install it with
+  pip install consequa-x and import consequa_x.
+- Tested on Python 3.13, 3.14 and 3.15 as well as 3.11 and 3.12.
+- Each framework integration was run once with a real model: the OpenAI Agents
+  SDK and LangChain with a local model through Ollama, the Claude Agent SDK
+  with Claude Haiku. The recorded tool calls matched what actually ran.
+- A new example, two_part_outcome.py, checks a refund and a ticket closure
+  as one outcome.
+
+Both files are published on PyPI with their registry hashes and sizes verified,
+and a fresh install from PyPI passed the installed checks. Installed checks of
+the same wheel passed on Windows with Python 3.11.9, 3.12.10, 3.13.16, 3.14.8
+and 3.15.0, and on Alpine Linux 3.24.2 with Python 3.14.8 in a local QEMU guest.
+The Linux result covers one emulated guest; macOS has not been tested.
+
 ## 2.1.0 — 19 October 2026
 
 An additive release with the features 2.0.0 left out. Saved traces and code
@@ -18,7 +44,17 @@ written for 2.0 work unchanged.
 
 The integrations were tested offline against openai-agents 0.24.0,
 langchain-core 1.6.12, claude-agent-sdk 0.2.168 and opentelemetry 1.45.1,
-without a live model or tracing backend.
+without a live model or tracing backend. Report tables now scroll on narrow
+screens instead of squeezing their cells.
+
+Both files are published on PyPI with their registry hashes and sizes verified,
+and a fresh install from PyPI passed the installed checks. The source suite
+passed 1,921 tests plus 241 subtests. Installed checks of the same wheel passed on Windows with embedded Python 3.11.9 and Python 3.12.10, and
+on Alpine Linux 3.24.2 with Python 3.14.8 in a local QEMU guest. The integration,
+OpenTelemetry and composite tests passed against the installed wheel with each
+SDK present. The source distribution installed on Windows Python 3.12.10 and
+ran the OpenTelemetry commands and a composite verify. The Linux result covers
+one emulated guest; macOS has not been tested.
 
 ## 2.0.0 — 19 October 2026
 

{consequa_x-2.1.0 → consequa_x-2.1.2}/docs/stable-api.md

@@ -38,7 +38,7 @@ Everything exported from `consequa_x` is stable through 2.x:
   (`ToolCallCapture`, `ToolCall`, `PendingCall`, `classify_exception`, and the
   `openai_agents`, `langchain` and `claude_agent_sdk` modules). The integration
   modules follow their frameworks' public APIs; a framework release that changes
-  those APIs may need a Consequa-X update, and each module states the version it
+  those APIs may need a ConseqaX update, and each module states the version it
   was tested with. The OpenTelemetry GenAI conventions are in development
   upstream, so exported `gen_ai.*` attribute names follow them as they change;
   `consequa_x.*` attributes are stable through 2.x.

consequa_x-2.1.2/docs/v2-migration-conformance.md

@@ -0,0 +1,220 @@
+# Offline 2.0 migration conformance on the retained matched traces
+
+This is a new **post-hoc compatibility study**, using the 503 retained factual
+traces from the closed 2026-10-18 matched local study. Its original denominator
+remains 504 planned slots: 503 completed, one failed (`episode_00109`, invalid
+policy JSON), and 896 already completed local model calls. The failed slot is
+inventoried and reported; it is never regenerated or given a substitute trace.
+The original outputs, labels, analysis and historical package version remain
+unchanged. There are zero new model calls, action executions or service requests.
+
+The measured corpus has **not** been processed during runner development. The
+protocol is frozen only after the runner, its tests and this design are committed.
+This is a prospective freeze of a post-hoc conformance analysis, not a claim that
+the original experiment or the analysis hypothesis was externally preregistered.
+
+## Scope and feasibility
+
+The public 2.0 `migrate_trace` accepts the saved schema `0.1` JSON and returns a
+schema `1.0` trace. Its one action and assessment become one-item lists; responses
+and verification receipts gain the action ID; the assessment gains an ID. The
+legacy schema never recorded a clock origin, so migration declares `relative_ms`.
+No epoch is inferred, and this clock-origin loss is disclosed for every trace.
+
+The retained study uses the built-in `refund_processed` predicate. Consequently
+the public 2.0 `verify(action, recorded_observation)` can independently recompute
+its receipts, and `explain(trace).for_action(action_id)` supplies the per-action
+assessment alignment and diagnosis codes. Target matching moved from the 1.1
+explanation into the 2.0 verifier. Migration adds target IDs and findings only
+when the current verifier reproduces the recorded **status, origin, ordered
+effect IDs and ordered attributable effect IDs**. That is the migration
+implementation's four-field condition, not a promise of byte-identical receipts.
+
+Migration preserves old recorded conclusions by construction. Therefore the
+runner treats structural preservation, migration explanation and **fresh 2.0
+verification** as separate results. A successful migration alone cannot establish
+verifier parity. Fresh receipts replace receipts only in a separate in-memory
+trace for explanation; the historical trace remains untouched.
+
+This study cannot establish new empirical performance for 2.0 policies, general
+multi-action correctness, arbitrary custom-verifier compatibility, a recovered
+absolute time origin, causal responsibility, or performance on a new population.
+It evaluates compatibility with this retained single-action evidence only.
+
+## Frozen inputs and code
+
+`scripts/run_v2_migration_conformance.py freeze` inventories bytes and identities;
+it does not execute migration, verification, explanation, a collector or a
+benchmark on any corpus trace. Import inspection is limited to source version,
+module path and installed distribution metadata.
+
+The protocol records:
+
+- The explicit full current Git commit, complete `src/consequa_x` file inventory,
+  source package version `2.0.0`, project metadata, migration documentation,
+  existing migration tests and their development fixtures.
+- Runner, dedicated tests and this document hashes. Pinned files must be tracked
+  and clean at freeze time; unrelated paper/research edits are allowed.
+- The frozen legacy commit `940ebfcc0059c992977dd8e9f112361ff5b31edd`, its complete
+  package inventory and source version `1.1.0`. Every legacy runtime hash must
+  also match the **original study's** recorded runtime inventory.
+- The complete historical study and analysis directory inventories, the 503
+  individual trace/explanation/receipt hashes and slot identities, the failed
+  slot's receipt/error, and each old analysis `consequa_state_origin` prediction.
+  The original finalization receipt and closed-study inventory are checked.
+  Evaluator truth is hashed as historical provenance, never used for a prediction
+  or new outcome label.
+- Both exact Python executables, executable hashes and interpreter versions,
+  imported module paths/source versions, and ambient installed distribution
+  versions/locations. Imported pinned source defines behavior. A wheel's installed
+  metadata does not silently replace source provenance.
+- The canonical mappings, planned metrics, denominators, missing-data rule,
+  diagnosis policy, limitations and development-fixture disclosure in `PLAN`.
+
+Before constructing slot dictionaries, the runner requires exactly 504 unique
+schedule IDs and analysis-episode IDs with equal sets, and exactly 503 unique
+baseline-row IDs matching the completed episodes. Duplicate or missing IDs cannot
+be hidden by dictionary overwrites. Source cleanliness checks include entire
+runtime/fixture directory pathspecs, so a tracked module deleted before freeze
+cannot silently disappear from the pinned inventory.
+
+The old arm runs under a separate `-I -B` Python subprocess with only the frozen
+legacy `src` explicitly inserted. It never imports the current old collector or
+changes either environment. The new arm requires `PYTHONPATH` to the pinned
+paper-writing `src` and checks the actual imported path and code version.
+
+## Comparisons and missing evidence
+
+For every recorded receipt, fresh verification uses its **one exact evidence
+observation ID** to select the saved observation. It does not select the latest
+readback as a fallback. Multiple evidence items, missing identities, unsupported
+predicates and runtime errors are retained as processing failures/unavailable
+comparisons. The latest explanation uses the last recorded receipt and assessment
+for that action, preserving recorded order rather than sorting by timestamp.
+
+Canonical state is `satisfied → true`, `violated → false`, `unknown → null`.
+Origin retains the exact `confirmed`, `other_actor` and `unknown` values. Reported
+assessment retains `success`, `failure`, `partial` and `unknown`. Alignment
+retains `consistent`, `contradicted` and `unassessable`. Action-linked outcome uses
+each pinned version's `benchmark.targets.action_linked_consequence`; it is not
+`bool(target_effect_ids)` because count, absence, external-origin and missing
+provenance semantics differ.
+
+Primary trace-level checks cover migration success; exact legacy structural
+projection; schema round-trip; preservation of old receipts and reported
+assessment; old re-verification versus recorded receipt core; fresh old/new core,
+state, origin, action linkage and receipt identities; migrated/fresh assessment
+alignment; frozen-old agreement with the saved explanation alignment and old
+analysis prediction. Receipt identity includes contract ID, observed time,
+ordered effect/attributable IDs, and every evidence observation ID/source/watermark.
+
+Secondary checks retain full legacy receipt equality (including details/reason),
+exact ordered diagnosis-code equality, additions/removals, migration reproduction
+counts, unavailable/empty/nonempty target metadata and newly unassessable
+alignment. An empty target tuple and unavailable (`null`) target matching are
+separate. Existing known unassessability and newly lost assessability are separate.
+2.0 can add `UNASSESSABLE_ASSESSMENT` for known refund outcomes whose report or
+target linkage cannot be judged; those changes are reported, not normalized away.
+Messages, display labels and graph serialization are not primary invariants.
+
+Every trace metric reports success/equal, failure/different and unavailable counts
+over **all 503 retained traces**, plus successes divided by the original 504
+planned slots. The failed historical slot is listed, not scored as semantic
+disagreement. Receipt-level counts have a separate receipt denominator. Equality
+of two null states, or two unassessable alignments, is agreement rather than
+known-state/assessable coverage; state and alignment distributions are reported.
+No failures are excluded, no retries or replacement samples are allowed, and no
+inferential p-values or population-level claims are planned.
+
+## Reproducible commands
+
+Use the isolated 2.0 environment for freezing and the measured new arm, and the
+isolated 1.1 environment for the old worker. The published 2.0 wheel installed in
+the new environment supplies metadata; `PYTHONPATH` deliberately selects the
+committed, separately hashed 2.0 source. These commands do not enable Actions.
+
+After committing the three research files, create a detached source worktree at
+that exact commit, named `D:/ConsequaX-Research/paper-frozen-v2-conformance` in
+this execution plan. Run the committed script there while explicitly pointing to
+the historical input directories in the writing worktree. This allows the frozen
+protocol to be committed in the writing worktree before measurement without
+moving the execution source's pinned HEAD. Freeze to a **new** directory:
+
+```powershell
+$env:PYTHONPATH = 'D:/ConsequaX-Research/paper-frozen-v2-conformance/src'
+$env:PYTHONDONTWRITEBYTECODE = '1'
+$currentRevision = (& git -C 'D:/ConsequaX-Research/paper-frozen-v2-conformance' rev-parse HEAD).Trim()
+& 'D:/ConsequaX-Research/migration-2.0-venv/Scripts/python.exe' `
+  'D:/ConsequaX-Research/paper-frozen-v2-conformance/scripts/run_v2_migration_conformance.py' freeze `
+  --expected-current-commit $currentRevision `
+  --study-dir 'D:/ConsequaX-Research/paper-writing/output/matched-local-study-2026-10-18' `
+  --analysis-dir 'D:/ConsequaX-Research/paper-writing/output/matched-local-analysis-2026-10-18' `
+  --legacy-root 'D:/ConsequaX-Research/paper-frozen-940ebfc' `
+  --legacy-python 'D:/ConsequaX-Research/analysis-1.1.0-venv/Scripts/python.exe' `
+  --output-dir 'D:/ConsequaX-Research/paper-writing/output/v2-migration-protocol-2026-10-19'
+```
+
+Review the frozen protocol and record/commit its hash **before** the measured run.
+Then run once, supplying the already frozen hash and a different new directory:
+
+```powershell
+$frozenHash = (Get-Content `
+  'D:/ConsequaX-Research/paper-writing/output/v2-migration-protocol-2026-10-19/protocol.sha256.json' `
+  -Raw | ConvertFrom-Json).sha256
+& 'D:/ConsequaX-Research/migration-2.0-venv/Scripts/python.exe' `
+  'D:/ConsequaX-Research/paper-frozen-v2-conformance/scripts/run_v2_migration_conformance.py' run `
+  --protocol 'D:/ConsequaX-Research/paper-writing/output/v2-migration-protocol-2026-10-19/protocol.json' `
+  --expected-protocol-sha256 $frozenHash `
+  --output-dir 'D:/ConsequaX-Research/paper-writing/output/v2-migration-conformance-2026-10-19'
+```
+
+The run checks frozen hashes before computation and again before its final
+receipt. It creates an exclusive `run-claim.json` beside the protocol. That claim
+persists after interruption or failure and prevents retry even into another
+output directory. Outputs are always created exclusively; there is no overwrite,
+resume, repair, automatic retry or sample regeneration. A source or input change
+requires a separately disclosed amendment, never silent reuse of this protocol.
+
+Outputs include per-slot paired results, migrated traces, fresh receipts,
+runtime metadata, an aggregate `summary.json`, an artifact inventory and a
+completion receipt. Processing errors remain in the results and yield exit code
+2; semantic disagreements remain ordinary measured results. A fatal error leaves
+its claim and `fatal-error.json`. No completion receipt claims input integrity
+unless the final hash recheck succeeds.
+
+Migration, migrated explanation, fresh verification and fresh explanation have
+separate failure accounting. A later verifier/explanation failure preserves
+successful migration, its available metrics and its migrated artifact; affected
+fresh comparisons remain unavailable. A legacy re-verification failure likewise
+does not erase an available recorded-alignment comparison.
+
+## Development checks and disclosure
+
+The dedicated tests use the four existing schema-0.1 migration examples only:
+refund timeout after commit, stale readback, another actor and record timeout
+after commit. Only the timeout refund example has a saved explanation fixture;
+other expected alignments/codes are explicit synthetic assertions. Additional
+mutations cover receipt non-reproduction, unknown assessment, conservative target
+matching, observation identity, evidence/reason versus receipt-core differences,
+hash tampering, exclusive claims, denominator preservation and protocol gating.
+Synthetic schedule mutations cover duplicate/missing IDs and failed-slot baseline
+substitution; a mocked Git deletion checks pre-freeze source completeness.
+An optional integration check invokes frozen 1.1 on the **three refund fixtures**.
+No retained study trace is opened by these tests.
+
+The existing primary development environment has pytest; the two minimal isolated
+version environments need no new installed test dependencies:
+
+```powershell
+$env:PYTHONPATH = 'D:/ConsequaX-Research/paper-writing/src'
+$env:PYTHONDONTWRITEBYTECODE = '1'
+$env:CONSEQUA_V2_LEGACY_ROOT = 'D:/ConsequaX-Research/paper-frozen-940ebfc'
+$env:CONSEQUA_V2_LEGACY_PYTHON = 'D:/ConsequaX-Research/analysis-1.1.0-venv/Scripts/python.exe'
+& 'C:/Users/DELL/OneDrive/Desktop/ConseqaX/.venv/Scripts/python.exe' -m pytest `
+  tests/test_v2_migration_conformance.py tests/test_migrate.py -q -p no:cacheprovider
+```
+
+Development results are software checks, excluded from measured counts. The
+manuscript must retain the original experiment's frozen 1.1.0 provenance and
+describe any eventual 2.0 result separately as this offline post-hoc conformance
+study.

{consequa_x-2.1.0 → consequa_x-2.1.2}/examples/agent_loop_sqlite.py

@@ -1,8 +1,8 @@
-"""A tool-calling agent loop, captured with Consequa-X and checked against SQLite.
+"""A tool-calling agent loop, captured with ConseqaX and checked against SQLite.
 
 Standard library only, offline and deterministic. A scripted support agent reads
 a customer message, decides to call an ``issue_refund`` tool, and reports what it
-believes happened. The tool writes to a real SQLite database. Consequa-X
+believes happened. The tool writes to a real SQLite database. ConseqaX
 records the loop as it runs, then reads the outcome back through a separate
 read-only connection and verifies it with the built-in ``RecordVerifier``.
 
@@ -16,7 +16,7 @@ outcome:
 - ``stale_readback``: the tool succeeds, but readback comes from a replica that
   lags behind the contract horizon, so the outcome stays unknown.
 
-Run from this folder with Consequa-X 2.0 or newer installed::
+Run from this folder with ConseqaX 2.0 or newer installed::
 
     python agent_loop_sqlite.py --mode timeout_after_commit --output agent-run