bantamkit 0.27.0__py3-none-any.whl

bantamkit/assets/evals/devteam/tasks/dt-error-contract.yaml

@@ -0,0 +1,186 @@
+# GENERATED by tools/devteam/build_tasks.py from assets/evals/devteam/manifest.yaml.
+# Do not hand-edit: `build_tasks.py check` fails if you do. The task's selection
+# rationale, reference walk and exclusion record live in the manifest.
+name: dt-error-contract
+family: dev-repo-code
+tools:
+- read_file
+- list_files
+workspace:
+  HISTORY.md: '# Release history
+
+
+    `git log --oneline --date=short --pretty=''%h %ad %s''`, newest first. Patch files
+
+    for selected commits are under `patches/`.
+
+
+    ```
+
+    9f2c1ab 2026-07-29 fix(retry): read retry_max_attempts from settings, not the constant
+
+    7d4e88c 2026-07-22 feat(settle): batch settlement with per-entry retry
+
+    6b1a903 2026-07-15 refactor(config): move every tunable into CONFIG_KEYS
+
+    5c0de41 2026-07-08 fix(validate): reject zero-amount entries
+
+    4a9bb27 2026-07-01 feat(report): render a per-handler summary
+
+    3e88f10 2026-06-24 feat(posting): post_entry writes one entry
+
+    2d5a6c9 2026-06-17 feat(errors): ValidationError and SettlementError
+
+    1c4f0b8 2026-06-10 chore: initial layout
+
+    ```
+
+
+    Branch `main` is at `9f2c1ab`. Tags: `v0.9.0` on `9f2c1ab`, `v0.8.0` on
+
+    `5c0de41`.
+
+    '
+  README.md: '# svc-ledger
+
+
+    Double-entry ledger service. Three handlers, one settings module, no globals.
+
+
+    - Module map and the call order between modules: `docs/architecture.md`
+
+    - Operating procedure, escalation, and where the tunables live: `docs/runbook.md`
+
+    - Release history, newest first, with commit ids: `HISTORY.md`
+
+    - Patch files for selected commits: `patches/`
+
+    - Open incidents, including captured tracebacks: `issues/`
+
+
+    Handlers are resolved at import time from `src/ledger/registry.py`. Nothing else
+
+    in the service knows the handler names.
+
+    '
+  docs/architecture.md: "# Architecture\n\nThe three entry points are the handlers listed in `src/ledger/registry.py`.\n\
+    \n- `settle.py` drives a batch. It wraps `posting.post_entry` in\n  `retry.with_retry`, and raises\
+    \ its own error when a batch is too large.\n- `posting.py` writes one entry. It validates first, via\n\
+    \  `validate.validate_entry`, and defines no exception of its own.\n- `report.py` renders a summary.\
+    \ It reads the registry and imports no handler.\n- `validate.py` is the only module that decides an\
+    \ entry is malformed.\n- `errors.py` defines every exception class in the service. No other module\n\
+    \  declares one.\n\nEvery tunable is a key in `src/ledger/config.py`; that module is the only one\n\
+    that touches `os.environ`. Modules ask `load_settings()` for values, so an\nattribute read on a `Settings`\
+    \ object is how you tell that a module consumes a\nconfig key.\n"
+  docs/runbook.md: '# Runbook
+
+
+    Attempt counts, backoff intervals and batch sizes are **not written on this
+
+    page** and never will be. Every one of them is a key with its default in
+
+    `src/ledger/config.py` — read that file for the current value. Do not answer an
+
+    operational question about a number from this page; it will be stale.
+
+
+    Escalation: three consecutive settlement failures page the on-call. Foreign
+
+    currency entries are converted, not rejected. A malformed entry is a caller bug
+
+    and is never retried past the wrapper''s own attempt budget.
+
+    '
+  issues/142-settlement-timeout.md: "# 142 — settlement aborts on a zero-amount entry\n\nReported by ops\
+    \ on 2026-08-02 against `main` (`9f2c1ab`). One entry in a batch\nof 40 carried `amount: 0`; the whole\
+    \ batch aborted after the wrapper exhausted\nits attempts. Captured traceback, verbatim from the worker\
+    \ log:\n\n```\nTraceback (most recent call last):\n  File \"src/ledger/settle.py\", line 15, in settle_batch\n\
+    \    posted.append(with_retry(post_entry, entry))\n  File \"src/ledger/retry.py\", line 17, in with_retry\n\
+    \    raise last\n  File \"src/ledger/posting.py\", line 9, in post_entry\n    validate_entry(entry)\n\
+    \  File \"src/ledger/validate.py\", line 13, in validate_entry\n    raise ValidationError(f\"amount\
+    \ must be non-zero: {entry!r}\")\nledger.errors.ValidationError: amount must be non-zero: {'amount':\
+    \ 0, 'ccy': 'EUR'}\n```\n\nOps question: which change put this check in, and should a caller bug really\n\
+    burn the whole attempt budget? Runbook says a malformed entry is never retried\npast the wrapper's\
+    \ budget, so behaviour matches the runbook; the argument is\nabout whether the budget should apply\
+    \ at all.\n"
+  patches/0009-retry-budget.patch: "From 9f2c1ab5e2d84c1f0a7b6935cc41d0e2a8f37b41 Mon Sep 17 00:00:00\
+    \ 2001\nDate: Wed, 29 Jul 2026 09:14:02 +0000\nSubject: [PATCH] fix(retry): read retry_max_attempts\
+    \ from settings, not the\n constant\n\nThe wrapper had its own hard-coded attempt count, so raising\
+    \ the deployed\nbudget changed nothing. It now asks load_settings() like every other module.\n---\n\
+    \ src/ledger/retry.py | 8 +++++---\n 1 file changed, 5 insertions(+), 3 deletions(-)\n\ndiff --git\
+    \ a/src/ledger/retry.py b/src/ledger/retry.py\n--- a/src/ledger/retry.py\n+++ b/src/ledger/retry.py\n\
+    @@ -1,14 +1,15 @@\n-\"\"\"Retry wrapper. Attempts are fixed at the module constant below.\"\"\"\n\
+    +\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\"\"\n\n import\
+    \ time\n\n-_MAX_ATTEMPTS = 3\n+from ledger.config import load_settings\n\n\n def with_retry(fn, *args):\n\
+    +    settings = load_settings()\n     last = None\n-    for _attempt in range(_MAX_ATTEMPTS):\n+ \
+    \   for _attempt in range(settings.retry_max_attempts):\n         try:\n             return fn(*args)\n\
+    \         except Exception as exc:\n             last = exc\n-            time.sleep(0.25)\n+    \
+    \        time.sleep(settings.retry_backoff_ms / 1000)\n     raise last\n--\n2.45.2\n"
+  src/ledger/__init__.py: '"""svc-ledger: a double-entry ledger service."""
+
+
+    __version__ = "0.9.0"
+
+    '
+  src/ledger/config.py: "\"\"\"The only module in svc-ledger that reads os.environ.\n\nEvery tunable is\
+    \ a key in CONFIG_KEYS with its default beside it. docs/runbook.md\ndeliberately does not repeat these\
+    \ values: one place, one default.\n\"\"\"\n\nimport os\n\nCONFIG_KEYS = {\n    \"retry_max_attempts\"\
+    : 5,\n    \"retry_backoff_ms\": 250,\n    \"settle_batch_size\": 200,\n    \"posting_currency\": \"\
+    EUR\",\n    \"report_top_n\": 10,\n}\n\n\nclass Settings:\n    \"\"\"Attribute access over CONFIG_KEYS,\
+    \ environment first.\"\"\"\n\n    def __init__(self, values):\n        self.values = values\n\n  \
+    \  def __getattr__(self, name):\n        if name not in self.values:\n            raise AttributeError(f\"\
+    unknown config key: {name}\")\n        return self.values[name]\n\n\ndef load_settings():\n    values\
+    \ = {}\n    for key, default in CONFIG_KEYS.items():\n        raw = os.environ.get(\"LEDGER_\" + key.upper())\n\
+    \        values[key] = type(default)(raw) if raw is not None else default\n    return Settings(values)\n"
+  src/ledger/errors.py: "\"\"\"Exception hierarchy for svc-ledger. Every module raises from here.\"\"\"\
+    \n\n\nclass LedgerError(Exception):\n    \"\"\"Base class. Nothing raises this directly.\"\"\"\n\n\
+    \nclass ValidationError(LedgerError):\n    \"\"\"An entry failed validate_entry.\"\"\"\n\n\nclass\
+    \ SettlementError(LedgerError):\n    \"\"\"A batch could not be settled.\"\"\"\n"
+  src/ledger/posting.py: "\"\"\"Post one entry to the ledger. Validates before writing; raises nothing\
+    \ itself.\"\"\"\n\nfrom ledger.config import load_settings\nfrom ledger.validate import validate_entry\n\
+    \n\ndef post_entry(entry):\n    settings = load_settings()\n    validate_entry(entry)\n    if entry[\"\
+    ccy\"] != settings.posting_currency:\n        return {\"status\": \"converted\", \"ccy\": settings.posting_currency}\n\
+    \    return {\"status\": \"posted\", \"ccy\": entry[\"ccy\"]}\n"
+  src/ledger/registry.py: "\"\"\"Handler registry. Resolved once at import time by the service entry point.\"\
+    \"\"\n\nHANDLERS = {\n    \"settle\": \"ledger.settle:settle_batch\",\n    \"post\": \"ledger.posting:post_entry\"\
+    ,\n    \"report\": \"ledger.report:build_report\",\n}\n"
+  src/ledger/report.py: "\"\"\"Per-handler summary. Reads the registry; imports no handler and no settings.\"\
+    \"\"\n\nfrom ledger.registry import HANDLERS\n\n\ndef build_report(rows):\n    ranked = sorted(rows,\
+    \ key=lambda r: r[\"amount\"], reverse=True)\n    # Left over from 6b1a903: this 10 was never moved\
+    \ into CONFIG_KEYS.\n    return {\"handlers\": sorted(HANDLERS), \"top\": ranked[:10]}\n"
+  src/ledger/retry.py: "\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\
+    \"\"\n\nimport time\n\nfrom ledger.config import load_settings\n\n\ndef with_retry(fn, *args):\n \
+    \   settings = load_settings()\n    last = None\n    for _attempt in range(settings.retry_max_attempts):\n\
+    \        try:\n            return fn(*args)\n        except Exception as exc:\n            last =\
+    \ exc\n            time.sleep(settings.retry_backoff_ms / 1000)\n    raise last\n"
+  src/ledger/settle.py: "\"\"\"Batch settlement. Drives posting through the retry wrapper.\"\"\"\n\nfrom\
+    \ ledger.config import load_settings\nfrom ledger.errors import SettlementError\nfrom ledger.posting\
+    \ import post_entry\nfrom ledger.retry import with_retry\n\n\ndef settle_batch(entries):\n    settings\
+    \ = load_settings()\n    if len(entries) > settings.settle_batch_size:\n        raise SettlementError(f\"\
+    batch of {len(entries)} exceeds settle_batch_size\")\n    posted = []\n    for entry in entries:\n\
+    \        posted.append(with_retry(post_entry, entry))\n    return posted\n"
+  src/ledger/validate.py: "\"\"\"Entry validation. The only module that decides an entry is malformed.\"\
+    \"\"\n\nfrom ledger.errors import ValidationError\n\nREQUIRED_FIELDS = (\"amount\", \"ccy\")\n\n\n\
+    def validate_entry(entry):\n    for field in REQUIRED_FIELDS:\n        if field not in entry:\n  \
+    \          raise ValidationError(f\"missing field {field!r}: {entry!r}\")\n    if entry[\"amount\"\
+    ] == 0:\n        raise ValidationError(f\"amount must be non-zero: {entry!r}\")\n    return entry\n"
+  tests/test_posting.py: "import pytest\n\nfrom ledger.errors import ValidationError\nfrom ledger.posting\
+    \ import post_entry\n\n\ndef test_post_entry_rejects_zero_amount():\n    with pytest.raises(ValidationError):\n\
+    \        post_entry({\"amount\": 0, \"ccy\": \"EUR\"})\n\n\ndef test_post_entry_converts_foreign_currency():\n\
+    \    assert post_entry({\"amount\": 10, \"ccy\": \"USD\"})[\"status\"] == \"converted\"\n"
+  tests/test_settle.py: "import pytest\n\nfrom ledger.errors import SettlementError\nfrom ledger.settle\
+    \ import settle_batch\n\n\ndef test_settle_batch_rejects_oversized_batch():\n    with pytest.raises(SettlementError):\n\
+    \        settle_batch([{\"amount\": 1, \"ccy\": \"EUR\"}] * 201)\n"
+prompt: 'Two questions about svc-ledger''s exceptions. (a) An entry with amount 0 reaches posting.post_entry
+  — name the exception class that leaves post_entry, and the file that DEFINES it. (b) Name the exception
+  class settle.settle_batch raises when a batch is too large, and the file that DEFINES it. For each,
+  read the module that raises and then the module that defines the class; do not answer from memory. Answer
+  with ONLY this JSON, nothing else: {"a_class": "<name>", "a_path": "<path>", "b_class": "<name>", "b_path":
+  "<path>"}'
+scoring:
+  kind: json_equal
+  expected:
+    a_class: ValidationError
+    a_path: src/ledger/errors.py
+    b_class: SettlementError
+    b_path: src/ledger/errors.py

bantamkit/assets/evals/devteam/tasks/dt-handler-map.yaml

@@ -0,0 +1,183 @@
+# GENERATED by tools/devteam/build_tasks.py from assets/evals/devteam/manifest.yaml.
+# Do not hand-edit: `build_tasks.py check` fails if you do. The task's selection
+# rationale, reference walk and exclusion record live in the manifest.
+name: dt-handler-map
+family: dev-repo-code
+tools:
+- read_file
+- list_files
+workspace:
+  HISTORY.md: '# Release history
+
+
+    `git log --oneline --date=short --pretty=''%h %ad %s''`, newest first. Patch files
+
+    for selected commits are under `patches/`.
+
+
+    ```
+
+    9f2c1ab 2026-07-29 fix(retry): read retry_max_attempts from settings, not the constant
+
+    7d4e88c 2026-07-22 feat(settle): batch settlement with per-entry retry
+
+    6b1a903 2026-07-15 refactor(config): move every tunable into CONFIG_KEYS
+
+    5c0de41 2026-07-08 fix(validate): reject zero-amount entries
+
+    4a9bb27 2026-07-01 feat(report): render a per-handler summary
+
+    3e88f10 2026-06-24 feat(posting): post_entry writes one entry
+
+    2d5a6c9 2026-06-17 feat(errors): ValidationError and SettlementError
+
+    1c4f0b8 2026-06-10 chore: initial layout
+
+    ```
+
+
+    Branch `main` is at `9f2c1ab`. Tags: `v0.9.0` on `9f2c1ab`, `v0.8.0` on
+
+    `5c0de41`.
+
+    '
+  README.md: '# svc-ledger
+
+
+    Double-entry ledger service. Three handlers, one settings module, no globals.
+
+
+    - Module map and the call order between modules: `docs/architecture.md`
+
+    - Operating procedure, escalation, and where the tunables live: `docs/runbook.md`
+
+    - Release history, newest first, with commit ids: `HISTORY.md`
+
+    - Patch files for selected commits: `patches/`
+
+    - Open incidents, including captured tracebacks: `issues/`
+
+
+    Handlers are resolved at import time from `src/ledger/registry.py`. Nothing else
+
+    in the service knows the handler names.
+
+    '
+  docs/architecture.md: "# Architecture\n\nThe three entry points are the handlers listed in `src/ledger/registry.py`.\n\
+    \n- `settle.py` drives a batch. It wraps `posting.post_entry` in\n  `retry.with_retry`, and raises\
+    \ its own error when a batch is too large.\n- `posting.py` writes one entry. It validates first, via\n\
+    \  `validate.validate_entry`, and defines no exception of its own.\n- `report.py` renders a summary.\
+    \ It reads the registry and imports no handler.\n- `validate.py` is the only module that decides an\
+    \ entry is malformed.\n- `errors.py` defines every exception class in the service. No other module\n\
+    \  declares one.\n\nEvery tunable is a key in `src/ledger/config.py`; that module is the only one\n\
+    that touches `os.environ`. Modules ask `load_settings()` for values, so an\nattribute read on a `Settings`\
+    \ object is how you tell that a module consumes a\nconfig key.\n"
+  docs/runbook.md: '# Runbook
+
+
+    Attempt counts, backoff intervals and batch sizes are **not written on this
+
+    page** and never will be. Every one of them is a key with its default in
+
+    `src/ledger/config.py` — read that file for the current value. Do not answer an
+
+    operational question about a number from this page; it will be stale.
+
+
+    Escalation: three consecutive settlement failures page the on-call. Foreign
+
+    currency entries are converted, not rejected. A malformed entry is a caller bug
+
+    and is never retried past the wrapper''s own attempt budget.
+
+    '
+  issues/142-settlement-timeout.md: "# 142 — settlement aborts on a zero-amount entry\n\nReported by ops\
+    \ on 2026-08-02 against `main` (`9f2c1ab`). One entry in a batch\nof 40 carried `amount: 0`; the whole\
+    \ batch aborted after the wrapper exhausted\nits attempts. Captured traceback, verbatim from the worker\
+    \ log:\n\n```\nTraceback (most recent call last):\n  File \"src/ledger/settle.py\", line 15, in settle_batch\n\
+    \    posted.append(with_retry(post_entry, entry))\n  File \"src/ledger/retry.py\", line 17, in with_retry\n\
+    \    raise last\n  File \"src/ledger/posting.py\", line 9, in post_entry\n    validate_entry(entry)\n\
+    \  File \"src/ledger/validate.py\", line 13, in validate_entry\n    raise ValidationError(f\"amount\
+    \ must be non-zero: {entry!r}\")\nledger.errors.ValidationError: amount must be non-zero: {'amount':\
+    \ 0, 'ccy': 'EUR'}\n```\n\nOps question: which change put this check in, and should a caller bug really\n\
+    burn the whole attempt budget? Runbook says a malformed entry is never retried\npast the wrapper's\
+    \ budget, so behaviour matches the runbook; the argument is\nabout whether the budget should apply\
+    \ at all.\n"
+  patches/0009-retry-budget.patch: "From 9f2c1ab5e2d84c1f0a7b6935cc41d0e2a8f37b41 Mon Sep 17 00:00:00\
+    \ 2001\nDate: Wed, 29 Jul 2026 09:14:02 +0000\nSubject: [PATCH] fix(retry): read retry_max_attempts\
+    \ from settings, not the\n constant\n\nThe wrapper had its own hard-coded attempt count, so raising\
+    \ the deployed\nbudget changed nothing. It now asks load_settings() like every other module.\n---\n\
+    \ src/ledger/retry.py | 8 +++++---\n 1 file changed, 5 insertions(+), 3 deletions(-)\n\ndiff --git\
+    \ a/src/ledger/retry.py b/src/ledger/retry.py\n--- a/src/ledger/retry.py\n+++ b/src/ledger/retry.py\n\
+    @@ -1,14 +1,15 @@\n-\"\"\"Retry wrapper. Attempts are fixed at the module constant below.\"\"\"\n\
+    +\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\"\"\n\n import\
+    \ time\n\n-_MAX_ATTEMPTS = 3\n+from ledger.config import load_settings\n\n\n def with_retry(fn, *args):\n\
+    +    settings = load_settings()\n     last = None\n-    for _attempt in range(_MAX_ATTEMPTS):\n+ \
+    \   for _attempt in range(settings.retry_max_attempts):\n         try:\n             return fn(*args)\n\
+    \         except Exception as exc:\n             last = exc\n-            time.sleep(0.25)\n+    \
+    \        time.sleep(settings.retry_backoff_ms / 1000)\n     raise last\n--\n2.45.2\n"
+  src/ledger/__init__.py: '"""svc-ledger: a double-entry ledger service."""
+
+
+    __version__ = "0.9.0"
+
+    '
+  src/ledger/config.py: "\"\"\"The only module in svc-ledger that reads os.environ.\n\nEvery tunable is\
+    \ a key in CONFIG_KEYS with its default beside it. docs/runbook.md\ndeliberately does not repeat these\
+    \ values: one place, one default.\n\"\"\"\n\nimport os\n\nCONFIG_KEYS = {\n    \"retry_max_attempts\"\
+    : 5,\n    \"retry_backoff_ms\": 250,\n    \"settle_batch_size\": 200,\n    \"posting_currency\": \"\
+    EUR\",\n    \"report_top_n\": 10,\n}\n\n\nclass Settings:\n    \"\"\"Attribute access over CONFIG_KEYS,\
+    \ environment first.\"\"\"\n\n    def __init__(self, values):\n        self.values = values\n\n  \
+    \  def __getattr__(self, name):\n        if name not in self.values:\n            raise AttributeError(f\"\
+    unknown config key: {name}\")\n        return self.values[name]\n\n\ndef load_settings():\n    values\
+    \ = {}\n    for key, default in CONFIG_KEYS.items():\n        raw = os.environ.get(\"LEDGER_\" + key.upper())\n\
+    \        values[key] = type(default)(raw) if raw is not None else default\n    return Settings(values)\n"
+  src/ledger/errors.py: "\"\"\"Exception hierarchy for svc-ledger. Every module raises from here.\"\"\"\
+    \n\n\nclass LedgerError(Exception):\n    \"\"\"Base class. Nothing raises this directly.\"\"\"\n\n\
+    \nclass ValidationError(LedgerError):\n    \"\"\"An entry failed validate_entry.\"\"\"\n\n\nclass\
+    \ SettlementError(LedgerError):\n    \"\"\"A batch could not be settled.\"\"\"\n"
+  src/ledger/posting.py: "\"\"\"Post one entry to the ledger. Validates before writing; raises nothing\
+    \ itself.\"\"\"\n\nfrom ledger.config import load_settings\nfrom ledger.validate import validate_entry\n\
+    \n\ndef post_entry(entry):\n    settings = load_settings()\n    validate_entry(entry)\n    if entry[\"\
+    ccy\"] != settings.posting_currency:\n        return {\"status\": \"converted\", \"ccy\": settings.posting_currency}\n\
+    \    return {\"status\": \"posted\", \"ccy\": entry[\"ccy\"]}\n"
+  src/ledger/registry.py: "\"\"\"Handler registry. Resolved once at import time by the service entry point.\"\
+    \"\"\n\nHANDLERS = {\n    \"settle\": \"ledger.settle:settle_batch\",\n    \"post\": \"ledger.posting:post_entry\"\
+    ,\n    \"report\": \"ledger.report:build_report\",\n}\n"
+  src/ledger/report.py: "\"\"\"Per-handler summary. Reads the registry; imports no handler and no settings.\"\
+    \"\"\n\nfrom ledger.registry import HANDLERS\n\n\ndef build_report(rows):\n    ranked = sorted(rows,\
+    \ key=lambda r: r[\"amount\"], reverse=True)\n    # Left over from 6b1a903: this 10 was never moved\
+    \ into CONFIG_KEYS.\n    return {\"handlers\": sorted(HANDLERS), \"top\": ranked[:10]}\n"
+  src/ledger/retry.py: "\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\
+    \"\"\n\nimport time\n\nfrom ledger.config import load_settings\n\n\ndef with_retry(fn, *args):\n \
+    \   settings = load_settings()\n    last = None\n    for _attempt in range(settings.retry_max_attempts):\n\
+    \        try:\n            return fn(*args)\n        except Exception as exc:\n            last =\
+    \ exc\n            time.sleep(settings.retry_backoff_ms / 1000)\n    raise last\n"
+  src/ledger/settle.py: "\"\"\"Batch settlement. Drives posting through the retry wrapper.\"\"\"\n\nfrom\
+    \ ledger.config import load_settings\nfrom ledger.errors import SettlementError\nfrom ledger.posting\
+    \ import post_entry\nfrom ledger.retry import with_retry\n\n\ndef settle_batch(entries):\n    settings\
+    \ = load_settings()\n    if len(entries) > settings.settle_batch_size:\n        raise SettlementError(f\"\
+    batch of {len(entries)} exceeds settle_batch_size\")\n    posted = []\n    for entry in entries:\n\
+    \        posted.append(with_retry(post_entry, entry))\n    return posted\n"
+  src/ledger/validate.py: "\"\"\"Entry validation. The only module that decides an entry is malformed.\"\
+    \"\"\n\nfrom ledger.errors import ValidationError\n\nREQUIRED_FIELDS = (\"amount\", \"ccy\")\n\n\n\
+    def validate_entry(entry):\n    for field in REQUIRED_FIELDS:\n        if field not in entry:\n  \
+    \          raise ValidationError(f\"missing field {field!r}: {entry!r}\")\n    if entry[\"amount\"\
+    ] == 0:\n        raise ValidationError(f\"amount must be non-zero: {entry!r}\")\n    return entry\n"
+  tests/test_posting.py: "import pytest\n\nfrom ledger.errors import ValidationError\nfrom ledger.posting\
+    \ import post_entry\n\n\ndef test_post_entry_rejects_zero_amount():\n    with pytest.raises(ValidationError):\n\
+    \        post_entry({\"amount\": 0, \"ccy\": \"EUR\"})\n\n\ndef test_post_entry_converts_foreign_currency():\n\
+    \    assert post_entry({\"amount\": 10, \"ccy\": \"USD\"})[\"status\"] == \"converted\"\n"
+  tests/test_settle.py: "import pytest\n\nfrom ledger.errors import SettlementError\nfrom ledger.settle\
+    \ import settle_batch\n\n\ndef test_settle_batch_rejects_oversized_batch():\n    with pytest.raises(SettlementError):\n\
+    \        settle_batch([{\"amount\": 1, \"ccy\": \"EUR\"}] * 201)\n"
+prompt: 'For each of the three handlers registered in src/ledger/registry.py, report whether the file
+  that defines its entry function imports ledger.config. Read the registry, then read each of the three
+  files it points at. Do not answer from memory. Answer with ONLY this JSON, nothing else: {"settle":
+  <true|false>, "post": <true|false>, "report": <true|false>}'
+scoring:
+  kind: json_equal
+  expected:
+    settle: true
+    post: true
+    report: false

bantamkit/assets/evals/devteam/tasks/dt-patch-before-after.yaml

@@ -0,0 +1,182 @@
+# GENERATED by tools/devteam/build_tasks.py from assets/evals/devteam/manifest.yaml.
+# Do not hand-edit: `build_tasks.py check` fails if you do. The task's selection
+# rationale, reference walk and exclusion record live in the manifest.
+name: dt-patch-before-after
+family: dev-repo-history
+tools:
+- read_file
+- list_files
+workspace:
+  HISTORY.md: '# Release history
+
+
+    `git log --oneline --date=short --pretty=''%h %ad %s''`, newest first. Patch files
+
+    for selected commits are under `patches/`.
+
+
+    ```
+
+    9f2c1ab 2026-07-29 fix(retry): read retry_max_attempts from settings, not the constant
+
+    7d4e88c 2026-07-22 feat(settle): batch settlement with per-entry retry
+
+    6b1a903 2026-07-15 refactor(config): move every tunable into CONFIG_KEYS
+
+    5c0de41 2026-07-08 fix(validate): reject zero-amount entries
+
+    4a9bb27 2026-07-01 feat(report): render a per-handler summary
+
+    3e88f10 2026-06-24 feat(posting): post_entry writes one entry
+
+    2d5a6c9 2026-06-17 feat(errors): ValidationError and SettlementError
+
+    1c4f0b8 2026-06-10 chore: initial layout
+
+    ```
+
+
+    Branch `main` is at `9f2c1ab`. Tags: `v0.9.0` on `9f2c1ab`, `v0.8.0` on
+
+    `5c0de41`.
+
+    '
+  README.md: '# svc-ledger
+
+
+    Double-entry ledger service. Three handlers, one settings module, no globals.
+
+
+    - Module map and the call order between modules: `docs/architecture.md`
+
+    - Operating procedure, escalation, and where the tunables live: `docs/runbook.md`
+
+    - Release history, newest first, with commit ids: `HISTORY.md`
+
+    - Patch files for selected commits: `patches/`
+
+    - Open incidents, including captured tracebacks: `issues/`
+
+
+    Handlers are resolved at import time from `src/ledger/registry.py`. Nothing else
+
+    in the service knows the handler names.
+
+    '
+  docs/architecture.md: "# Architecture\n\nThe three entry points are the handlers listed in `src/ledger/registry.py`.\n\
+    \n- `settle.py` drives a batch. It wraps `posting.post_entry` in\n  `retry.with_retry`, and raises\
+    \ its own error when a batch is too large.\n- `posting.py` writes one entry. It validates first, via\n\
+    \  `validate.validate_entry`, and defines no exception of its own.\n- `report.py` renders a summary.\
+    \ It reads the registry and imports no handler.\n- `validate.py` is the only module that decides an\
+    \ entry is malformed.\n- `errors.py` defines every exception class in the service. No other module\n\
+    \  declares one.\n\nEvery tunable is a key in `src/ledger/config.py`; that module is the only one\n\
+    that touches `os.environ`. Modules ask `load_settings()` for values, so an\nattribute read on a `Settings`\
+    \ object is how you tell that a module consumes a\nconfig key.\n"
+  docs/runbook.md: '# Runbook
+
+
+    Attempt counts, backoff intervals and batch sizes are **not written on this
+
+    page** and never will be. Every one of them is a key with its default in
+
+    `src/ledger/config.py` — read that file for the current value. Do not answer an
+
+    operational question about a number from this page; it will be stale.
+
+
+    Escalation: three consecutive settlement failures page the on-call. Foreign
+
+    currency entries are converted, not rejected. A malformed entry is a caller bug
+
+    and is never retried past the wrapper''s own attempt budget.
+
+    '
+  issues/142-settlement-timeout.md: "# 142 — settlement aborts on a zero-amount entry\n\nReported by ops\
+    \ on 2026-08-02 against `main` (`9f2c1ab`). One entry in a batch\nof 40 carried `amount: 0`; the whole\
+    \ batch aborted after the wrapper exhausted\nits attempts. Captured traceback, verbatim from the worker\
+    \ log:\n\n```\nTraceback (most recent call last):\n  File \"src/ledger/settle.py\", line 15, in settle_batch\n\
+    \    posted.append(with_retry(post_entry, entry))\n  File \"src/ledger/retry.py\", line 17, in with_retry\n\
+    \    raise last\n  File \"src/ledger/posting.py\", line 9, in post_entry\n    validate_entry(entry)\n\
+    \  File \"src/ledger/validate.py\", line 13, in validate_entry\n    raise ValidationError(f\"amount\
+    \ must be non-zero: {entry!r}\")\nledger.errors.ValidationError: amount must be non-zero: {'amount':\
+    \ 0, 'ccy': 'EUR'}\n```\n\nOps question: which change put this check in, and should a caller bug really\n\
+    burn the whole attempt budget? Runbook says a malformed entry is never retried\npast the wrapper's\
+    \ budget, so behaviour matches the runbook; the argument is\nabout whether the budget should apply\
+    \ at all.\n"
+  patches/0009-retry-budget.patch: "From 9f2c1ab5e2d84c1f0a7b6935cc41d0e2a8f37b41 Mon Sep 17 00:00:00\
+    \ 2001\nDate: Wed, 29 Jul 2026 09:14:02 +0000\nSubject: [PATCH] fix(retry): read retry_max_attempts\
+    \ from settings, not the\n constant\n\nThe wrapper had its own hard-coded attempt count, so raising\
+    \ the deployed\nbudget changed nothing. It now asks load_settings() like every other module.\n---\n\
+    \ src/ledger/retry.py | 8 +++++---\n 1 file changed, 5 insertions(+), 3 deletions(-)\n\ndiff --git\
+    \ a/src/ledger/retry.py b/src/ledger/retry.py\n--- a/src/ledger/retry.py\n+++ b/src/ledger/retry.py\n\
+    @@ -1,14 +1,15 @@\n-\"\"\"Retry wrapper. Attempts are fixed at the module constant below.\"\"\"\n\
+    +\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\"\"\n\n import\
+    \ time\n\n-_MAX_ATTEMPTS = 3\n+from ledger.config import load_settings\n\n\n def with_retry(fn, *args):\n\
+    +    settings = load_settings()\n     last = None\n-    for _attempt in range(_MAX_ATTEMPTS):\n+ \
+    \   for _attempt in range(settings.retry_max_attempts):\n         try:\n             return fn(*args)\n\
+    \         except Exception as exc:\n             last = exc\n-            time.sleep(0.25)\n+    \
+    \        time.sleep(settings.retry_backoff_ms / 1000)\n     raise last\n--\n2.45.2\n"
+  src/ledger/__init__.py: '"""svc-ledger: a double-entry ledger service."""
+
+
+    __version__ = "0.9.0"
+
+    '
+  src/ledger/config.py: "\"\"\"The only module in svc-ledger that reads os.environ.\n\nEvery tunable is\
+    \ a key in CONFIG_KEYS with its default beside it. docs/runbook.md\ndeliberately does not repeat these\
+    \ values: one place, one default.\n\"\"\"\n\nimport os\n\nCONFIG_KEYS = {\n    \"retry_max_attempts\"\
+    : 5,\n    \"retry_backoff_ms\": 250,\n    \"settle_batch_size\": 200,\n    \"posting_currency\": \"\
+    EUR\",\n    \"report_top_n\": 10,\n}\n\n\nclass Settings:\n    \"\"\"Attribute access over CONFIG_KEYS,\
+    \ environment first.\"\"\"\n\n    def __init__(self, values):\n        self.values = values\n\n  \
+    \  def __getattr__(self, name):\n        if name not in self.values:\n            raise AttributeError(f\"\
+    unknown config key: {name}\")\n        return self.values[name]\n\n\ndef load_settings():\n    values\
+    \ = {}\n    for key, default in CONFIG_KEYS.items():\n        raw = os.environ.get(\"LEDGER_\" + key.upper())\n\
+    \        values[key] = type(default)(raw) if raw is not None else default\n    return Settings(values)\n"
+  src/ledger/errors.py: "\"\"\"Exception hierarchy for svc-ledger. Every module raises from here.\"\"\"\
+    \n\n\nclass LedgerError(Exception):\n    \"\"\"Base class. Nothing raises this directly.\"\"\"\n\n\
+    \nclass ValidationError(LedgerError):\n    \"\"\"An entry failed validate_entry.\"\"\"\n\n\nclass\
+    \ SettlementError(LedgerError):\n    \"\"\"A batch could not be settled.\"\"\"\n"
+  src/ledger/posting.py: "\"\"\"Post one entry to the ledger. Validates before writing; raises nothing\
+    \ itself.\"\"\"\n\nfrom ledger.config import load_settings\nfrom ledger.validate import validate_entry\n\
+    \n\ndef post_entry(entry):\n    settings = load_settings()\n    validate_entry(entry)\n    if entry[\"\
+    ccy\"] != settings.posting_currency:\n        return {\"status\": \"converted\", \"ccy\": settings.posting_currency}\n\
+    \    return {\"status\": \"posted\", \"ccy\": entry[\"ccy\"]}\n"
+  src/ledger/registry.py: "\"\"\"Handler registry. Resolved once at import time by the service entry point.\"\
+    \"\"\n\nHANDLERS = {\n    \"settle\": \"ledger.settle:settle_batch\",\n    \"post\": \"ledger.posting:post_entry\"\
+    ,\n    \"report\": \"ledger.report:build_report\",\n}\n"
+  src/ledger/report.py: "\"\"\"Per-handler summary. Reads the registry; imports no handler and no settings.\"\
+    \"\"\n\nfrom ledger.registry import HANDLERS\n\n\ndef build_report(rows):\n    ranked = sorted(rows,\
+    \ key=lambda r: r[\"amount\"], reverse=True)\n    # Left over from 6b1a903: this 10 was never moved\
+    \ into CONFIG_KEYS.\n    return {\"handlers\": sorted(HANDLERS), \"top\": ranked[:10]}\n"
+  src/ledger/retry.py: "\"\"\"Retry wrapper. Reads its attempt count from settings, never from a constant.\"\
+    \"\"\n\nimport time\n\nfrom ledger.config import load_settings\n\n\ndef with_retry(fn, *args):\n \
+    \   settings = load_settings()\n    last = None\n    for _attempt in range(settings.retry_max_attempts):\n\
+    \        try:\n            return fn(*args)\n        except Exception as exc:\n            last =\
+    \ exc\n            time.sleep(settings.retry_backoff_ms / 1000)\n    raise last\n"
+  src/ledger/settle.py: "\"\"\"Batch settlement. Drives posting through the retry wrapper.\"\"\"\n\nfrom\
+    \ ledger.config import load_settings\nfrom ledger.errors import SettlementError\nfrom ledger.posting\
+    \ import post_entry\nfrom ledger.retry import with_retry\n\n\ndef settle_batch(entries):\n    settings\
+    \ = load_settings()\n    if len(entries) > settings.settle_batch_size:\n        raise SettlementError(f\"\
+    batch of {len(entries)} exceeds settle_batch_size\")\n    posted = []\n    for entry in entries:\n\
+    \        posted.append(with_retry(post_entry, entry))\n    return posted\n"
+  src/ledger/validate.py: "\"\"\"Entry validation. The only module that decides an entry is malformed.\"\
+    \"\"\n\nfrom ledger.errors import ValidationError\n\nREQUIRED_FIELDS = (\"amount\", \"ccy\")\n\n\n\
+    def validate_entry(entry):\n    for field in REQUIRED_FIELDS:\n        if field not in entry:\n  \
+    \          raise ValidationError(f\"missing field {field!r}: {entry!r}\")\n    if entry[\"amount\"\
+    ] == 0:\n        raise ValidationError(f\"amount must be non-zero: {entry!r}\")\n    return entry\n"
+  tests/test_posting.py: "import pytest\n\nfrom ledger.errors import ValidationError\nfrom ledger.posting\
+    \ import post_entry\n\n\ndef test_post_entry_rejects_zero_amount():\n    with pytest.raises(ValidationError):\n\
+    \        post_entry({\"amount\": 0, \"ccy\": \"EUR\"})\n\n\ndef test_post_entry_converts_foreign_currency():\n\
+    \    assert post_entry({\"amount\": 10, \"ccy\": \"USD\"})[\"status\"] == \"converted\"\n"
+  tests/test_settle.py: "import pytest\n\nfrom ledger.errors import SettlementError\nfrom ledger.settle\
+    \ import settle_batch\n\n\ndef test_settle_batch_rejects_oversized_batch():\n    with pytest.raises(SettlementError):\n\
+    \        settle_batch([{\"amount\": 1, \"ccy\": \"EUR\"}] * 201)\n"
+prompt: 'patches/0009-retry-budget.patch changed how src/ledger/retry.py obtains its attempt count. Report
+  the attempt count the code used BEFORE that patch, and the attempt count it uses at HEAD. Read the patch,
+  then the file at HEAD, then whatever that file reads its value from. Do not answer from memory. Answer
+  with ONLY this JSON, nothing else: {"before": <number>, "head": <number>}'
+scoring:
+  kind: json_equal
+  expected:
+    before: 3
+    head: 5