credit-ledger 0.1.0__tar.gz

credit_ledger-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Isaac Teague Frayling
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

credit_ledger-0.1.0/PKG-INFO

@@ -0,0 +1,127 @@
+Metadata-Version: 2.4
+Name: credit-ledger
+Version: 0.1.0
+Summary: Atomic, overdraft-proof, idempotent credit metering over Postgres. Extracted from PANTHEON.
+Author: Isaac Teague Frayling
+License: Apache-2.0
+Project-URL: Homepage, https://pantheonlabs.info
+Keywords: metering,billing,credits,idempotency,postgres,multi-tenant,row-level-security
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: SQLAlchemy>=2.0
+Provides-Extra: postgres
+Requires-Dist: psycopg[binary]>=3.1; extra == "postgres"
+Provides-Extra: test
+Requires-Dist: pytest>=7; extra == "test"
+Dynamic: license-file
+
+# credit-ledger
+
+[![tests](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml/badge.svg)](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![Python](https://img.shields.io/pypi/pyversions/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE)
+
+**An atomic, overdraft-proof, idempotent credit ledger — for metering money on a pay-per-use or multi-tenant service.** ~150 lines of Python over Postgres. Extracted from [PANTHEON](https://pantheonlabs.info), a multi-tenant AI substrate, where it's the money path for autonomous AI agents that spend real credits on every turn.
+
+The interesting part isn't the size — it's that three properties that are usually gotten *subtly wrong* are each guaranteed by a single, boring database mechanism instead of application-level hope:
+
+| Property | How | Not by |
+|---|---|---|
+| **No overdraft, ever** — a balance can never go negative, even under a stampede of concurrent charges | one guarded `UPDATE … WHERE credits >= n` | `SELECT` then `UPDATE` (a race), or a mutex (doesn't survive multiple processes) |
+| **Exactly-once under retries** — a replayed charge (a Stripe webhook fired twice, a client retry) charges once | `UNIQUE(tenant_id, idempotency_key)` on an append-only ledger | de-duping in app code (another race) |
+| **Tenant isolation** — tenant A can't read or write tenant B's balance | Postgres row-level security, `FORCE`d | a `WHERE tenant_id = ?` you have to remember to add every time |
+
+## The core: one statement does the hard part
+
+A spend is a single guarded, atomic `UPDATE`:
+
+```sql
+UPDATE credit_balance
+   SET credits = credits - :n
+ WHERE tenant_id = :t AND credits >= :n
+RETURNING credits;
+```
+
+- The row lock **serialises** concurrent charges against the same tenant — no lost updates.
+- The `credits >= :n` guard makes overdraft **structurally impossible**: the charge either fully succeeds (a row comes back) or changes nothing (zero rows). There is no window between the check and the debit, because they're the same statement.
+
+That's the whole trick, and the test proves it: fire **100 concurrent spends of 1 credit against a balance of 50**, and *exactly 50* succeed — the balance lands at 0, never negative, with 50 ledger rows.
+
+```python
+# tests/test_metering.py
+def test_atomic_decrement_under_concurrency(engine, tenant):
+    metering.grant(tenant, 50, engine=engine)
+    with ThreadPoolExecutor(max_workers=20) as ex:
+        results = list(ex.map(lambda _: metering.decrement(tenant, 1, engine=pool).ok, range(100)))
+    assert sum(results) == 50                              # exactly the balance, no more
+    assert metering.balance(tenant, engine=engine) == 0.0  # never negative
+```
+
+## Idempotency: the ledger *is* the dedup
+
+Every grant and spend appends a row to `credit_ledger`, which carries a `UNIQUE(tenant_id, idempotency_key)`. Pass an `idempotency_key` (a Stripe `event_id`, a request id) and a replay can't insert a second row — the database rejects it, and the code returns the *original* result marked `deduped=True`. Money is never double-counted, and a duplicate never surfaces as a 500.
+
+The subtle bit is the concurrent race: two identical charges arrive at once, both past the "already seen?" check, both try to insert. One wins; the other's `INSERT` violates the constraint, which **rolls back its whole transaction — including the decrement**. The loser then re-reads the ledger, finds the winner's row, and returns that. So a concurrent duplicate is a clean dedup, not a double-charge and not an error. There's a test for exactly this — 20 concurrent replays of one Stripe event → *one* credit applied, zero exceptions.
+
+## The correctness detail most people miss: fail loud, not phantom-success
+
+An `IntegrityError` after retries is ambiguous — it could be a genuine concurrent dedup (the winner committed, there's a ledger row), **or** it could be a foreign-key violation because the tenant doesn't exist (nothing landed). Reporting success in the second case would silently lose money. So the code disambiguates by re-reading the ledger: a row means real dedup → report success; no row means nothing happened → **fail loud** (`ok=False`). This came out of an adversarial self-audit of the substrate; it's the difference between "looks fine" and "correct."
+
+## The RLS footgun this quietly avoids
+
+Row-level security is only a guarantee if the writer is actually subject to it. Metering runs on a *privileged* app role (the money path is infrastructure), and on any managed Postgres that role is **not** `BYPASSRLS`. So before every transaction the code binds the tenant into a session variable the policy reads:
+
+```python
+c.execute(text("SELECT set_config('app.current_tenant', :t, true)"), {"t": tenant_id})
+```
+
+Skip this and, the moment RLS is enforced in production, the guarded `UPDATE` matches zero rows and the `INSERT`'s `WITH CHECK` fails — **billing silently bricks**, while every dev box (superuser, RLS bypassed) looks perfectly fine. Getting this right is the difference between a demo and something you'd run.
+
+## Why so little code
+
+The database does the concurrency, the atomicity, and the isolation. The code's entire job is to *use those primitives correctly* — bind the tenant, guard the decrement, key the ledger, and disambiguate the one genuinely-ambiguous failure. Fewer moving parts is the point, not an accident.
+
+## Use it
+
+```bash
+pip install -e .                       # SQLAlchemy is the only runtime dep
+createdb ledger && psql "$DATABASE_URL" -f schema.sql
+```
+
+```python
+from credit_ledger import metering
+
+metering.grant(tenant_id, 100, idempotency_key="stripe:evt:abc")  # top up (Stripe-retry-safe)
+c = metering.decrement(tenant_id, 1, reason="chat-turn")          # spend
+if not c.ok:
+    ...  # out of credits — nothing was charged
+metering.balance(tenant_id)                                       # -> float
+```
+
+Every function also takes an explicit `engine=` (for your own pool/lifecycle). `schema.sql` is the full Postgres schema including the RLS policies.
+
+## Run the tests
+
+```bash
+DATABASE_URL=postgresql://localhost/ledger pytest -q     # needs Postgres + schema.sql applied
+```
+
+The suite is the spec: overdraft impossibility under 100-way concurrency, idempotent spend and grant retries, concurrent Stripe-replay dedup, and rejection of non-positive grants (a negative "grant" would decrement a tenant — the primitive refuses it before touching the DB).
+
+## Design notes
+
+A few deliberate choices, and the answers to the sharp questions they invite:
+
+- **A failed (insufficient) charge records nothing — so a retry with the same key re-attempts. That's intended.** Idempotency dedups *successful effects*, not attempts. A charge rejected for insufficient balance didn't happen, so the same key should be free to succeed later — e.g. a metered webhook that hit a zero balance and is retried after a top-up. The dedup keys the *committed ledger row*, which only exists for charges that actually landed.
+- **`numeric(18,4)` in the database, `float` at the API boundary.** The stored balance and every ledger delta are exact decimals — no binary-float drift on the money itself. The functions return `float` for ergonomics, which is fine for integer-ish credits; if you were metering *actual currency*, return `Decimal` at the boundary too (a one-line change) so nothing rounds on the way out.
+- **Why two attempts is enough** (`for _attempt in range(2)`). Under N concurrent charges sharing one key, exactly one wins the `UNIQUE(tenant_id, idempotency_key)` insert and commits; every loser's `INSERT` raises, rolling back its whole transaction (decrement included). On the second pass the losers see the winner's now-committed ledger row in the "already seen?" read and return *that* — a clean dedup. A third attempt can't be needed: once the winner commits, the row is visible, so attempt two is always either the winner's own success or a loser's dedup. The re-read *after* the loop only exists to distinguish the genuinely-nothing-landed case (e.g. a foreign-key violation) — so a constraint error is never reported as phantom success.
+
+## Context
+
+This is one self-contained piece of **[PANTHEON](https://pantheonlabs.info)** — a multi-tenant substrate for running AI agents safely on other people's money and data (tenancy + RLS isolation, a governed reasoning loop, an approval queue, this metering ledger, and a quality gate), designed and built solo. The live Studio at [pantheonlabs.info](https://pantheonlabs.info) runs on it; the flagship write-up is at [pantheonlabs.co.uk](https://pantheonlabs.co.uk). Built by Isaac Teague Frayling.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

credit_ledger-0.1.0/README.md

@@ -0,0 +1,109 @@
+# credit-ledger
+
+[![tests](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml/badge.svg)](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![Python](https://img.shields.io/pypi/pyversions/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE)
+
+**An atomic, overdraft-proof, idempotent credit ledger — for metering money on a pay-per-use or multi-tenant service.** ~150 lines of Python over Postgres. Extracted from [PANTHEON](https://pantheonlabs.info), a multi-tenant AI substrate, where it's the money path for autonomous AI agents that spend real credits on every turn.
+
+The interesting part isn't the size — it's that three properties that are usually gotten *subtly wrong* are each guaranteed by a single, boring database mechanism instead of application-level hope:
+
+| Property | How | Not by |
+|---|---|---|
+| **No overdraft, ever** — a balance can never go negative, even under a stampede of concurrent charges | one guarded `UPDATE … WHERE credits >= n` | `SELECT` then `UPDATE` (a race), or a mutex (doesn't survive multiple processes) |
+| **Exactly-once under retries** — a replayed charge (a Stripe webhook fired twice, a client retry) charges once | `UNIQUE(tenant_id, idempotency_key)` on an append-only ledger | de-duping in app code (another race) |
+| **Tenant isolation** — tenant A can't read or write tenant B's balance | Postgres row-level security, `FORCE`d | a `WHERE tenant_id = ?` you have to remember to add every time |
+
+## The core: one statement does the hard part
+
+A spend is a single guarded, atomic `UPDATE`:
+
+```sql
+UPDATE credit_balance
+   SET credits = credits - :n
+ WHERE tenant_id = :t AND credits >= :n
+RETURNING credits;
+```
+
+- The row lock **serialises** concurrent charges against the same tenant — no lost updates.
+- The `credits >= :n` guard makes overdraft **structurally impossible**: the charge either fully succeeds (a row comes back) or changes nothing (zero rows). There is no window between the check and the debit, because they're the same statement.
+
+That's the whole trick, and the test proves it: fire **100 concurrent spends of 1 credit against a balance of 50**, and *exactly 50* succeed — the balance lands at 0, never negative, with 50 ledger rows.
+
+```python
+# tests/test_metering.py
+def test_atomic_decrement_under_concurrency(engine, tenant):
+    metering.grant(tenant, 50, engine=engine)
+    with ThreadPoolExecutor(max_workers=20) as ex:
+        results = list(ex.map(lambda _: metering.decrement(tenant, 1, engine=pool).ok, range(100)))
+    assert sum(results) == 50                              # exactly the balance, no more
+    assert metering.balance(tenant, engine=engine) == 0.0  # never negative
+```
+
+## Idempotency: the ledger *is* the dedup
+
+Every grant and spend appends a row to `credit_ledger`, which carries a `UNIQUE(tenant_id, idempotency_key)`. Pass an `idempotency_key` (a Stripe `event_id`, a request id) and a replay can't insert a second row — the database rejects it, and the code returns the *original* result marked `deduped=True`. Money is never double-counted, and a duplicate never surfaces as a 500.
+
+The subtle bit is the concurrent race: two identical charges arrive at once, both past the "already seen?" check, both try to insert. One wins; the other's `INSERT` violates the constraint, which **rolls back its whole transaction — including the decrement**. The loser then re-reads the ledger, finds the winner's row, and returns that. So a concurrent duplicate is a clean dedup, not a double-charge and not an error. There's a test for exactly this — 20 concurrent replays of one Stripe event → *one* credit applied, zero exceptions.
+
+## The correctness detail most people miss: fail loud, not phantom-success
+
+An `IntegrityError` after retries is ambiguous — it could be a genuine concurrent dedup (the winner committed, there's a ledger row), **or** it could be a foreign-key violation because the tenant doesn't exist (nothing landed). Reporting success in the second case would silently lose money. So the code disambiguates by re-reading the ledger: a row means real dedup → report success; no row means nothing happened → **fail loud** (`ok=False`). This came out of an adversarial self-audit of the substrate; it's the difference between "looks fine" and "correct."
+
+## The RLS footgun this quietly avoids
+
+Row-level security is only a guarantee if the writer is actually subject to it. Metering runs on a *privileged* app role (the money path is infrastructure), and on any managed Postgres that role is **not** `BYPASSRLS`. So before every transaction the code binds the tenant into a session variable the policy reads:
+
+```python
+c.execute(text("SELECT set_config('app.current_tenant', :t, true)"), {"t": tenant_id})
+```
+
+Skip this and, the moment RLS is enforced in production, the guarded `UPDATE` matches zero rows and the `INSERT`'s `WITH CHECK` fails — **billing silently bricks**, while every dev box (superuser, RLS bypassed) looks perfectly fine. Getting this right is the difference between a demo and something you'd run.
+
+## Why so little code
+
+The database does the concurrency, the atomicity, and the isolation. The code's entire job is to *use those primitives correctly* — bind the tenant, guard the decrement, key the ledger, and disambiguate the one genuinely-ambiguous failure. Fewer moving parts is the point, not an accident.
+
+## Use it
+
+```bash
+pip install -e .                       # SQLAlchemy is the only runtime dep
+createdb ledger && psql "$DATABASE_URL" -f schema.sql
+```
+
+```python
+from credit_ledger import metering
+
+metering.grant(tenant_id, 100, idempotency_key="stripe:evt:abc")  # top up (Stripe-retry-safe)
+c = metering.decrement(tenant_id, 1, reason="chat-turn")          # spend
+if not c.ok:
+    ...  # out of credits — nothing was charged
+metering.balance(tenant_id)                                       # -> float
+```
+
+Every function also takes an explicit `engine=` (for your own pool/lifecycle). `schema.sql` is the full Postgres schema including the RLS policies.
+
+## Run the tests
+
+```bash
+DATABASE_URL=postgresql://localhost/ledger pytest -q     # needs Postgres + schema.sql applied
+```
+
+The suite is the spec: overdraft impossibility under 100-way concurrency, idempotent spend and grant retries, concurrent Stripe-replay dedup, and rejection of non-positive grants (a negative "grant" would decrement a tenant — the primitive refuses it before touching the DB).
+
+## Design notes
+
+A few deliberate choices, and the answers to the sharp questions they invite:
+
+- **A failed (insufficient) charge records nothing — so a retry with the same key re-attempts. That's intended.** Idempotency dedups *successful effects*, not attempts. A charge rejected for insufficient balance didn't happen, so the same key should be free to succeed later — e.g. a metered webhook that hit a zero balance and is retried after a top-up. The dedup keys the *committed ledger row*, which only exists for charges that actually landed.
+- **`numeric(18,4)` in the database, `float` at the API boundary.** The stored balance and every ledger delta are exact decimals — no binary-float drift on the money itself. The functions return `float` for ergonomics, which is fine for integer-ish credits; if you were metering *actual currency*, return `Decimal` at the boundary too (a one-line change) so nothing rounds on the way out.
+- **Why two attempts is enough** (`for _attempt in range(2)`). Under N concurrent charges sharing one key, exactly one wins the `UNIQUE(tenant_id, idempotency_key)` insert and commits; every loser's `INSERT` raises, rolling back its whole transaction (decrement included). On the second pass the losers see the winner's now-committed ledger row in the "already seen?" read and return *that* — a clean dedup. A third attempt can't be needed: once the winner commits, the row is visible, so attempt two is always either the winner's own success or a loser's dedup. The re-read *after* the loop only exists to distinguish the genuinely-nothing-landed case (e.g. a foreign-key violation) — so a constraint error is never reported as phantom success.
+
+## Context
+
+This is one self-contained piece of **[PANTHEON](https://pantheonlabs.info)** — a multi-tenant substrate for running AI agents safely on other people's money and data (tenancy + RLS isolation, a governed reasoning loop, an approval queue, this metering ledger, and a quality gate), designed and built solo. The live Studio at [pantheonlabs.info](https://pantheonlabs.info) runs on it; the flagship write-up is at [pantheonlabs.co.uk](https://pantheonlabs.co.uk). Built by Isaac Teague Frayling.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

credit_ledger-0.1.0/credit_ledger/__init__.py

@@ -0,0 +1,6 @@
+# Copyright 2026 Isaac Teague Frayling
+# Licensed under the Apache License, Version 2.0 (see LICENSE).
+"""credit-ledger — atomic, overdraft-proof, idempotent credit metering over Postgres."""
+from .metering import Charge, balance, configure, decrement, grant
+
+__all__ = ["Charge", "balance", "configure", "decrement", "grant"]

credit_ledger-0.1.0/credit_ledger/metering.py

@@ -0,0 +1,169 @@
+# Copyright 2026 Isaac Teague Frayling
+# Licensed under the Apache License, Version 2.0 (see LICENSE).
+"""Atomic, overdraft-proof, idempotent credit metering.
+
+The whole charge is one guarded UPDATE:
+
+    UPDATE credit_balance SET credits = credits - n WHERE tenant_id = :t AND credits >= n
+
+The row lock serialises concurrent charges; the ``credits >= n`` guard makes overdraft
+impossible — a charge either fully succeeds or doesn't happen. Every charge is written
+to an append-only ledger; ``UNIQUE(tenant_id, idempotency_key)`` makes a retried charge a
+no-op (and, on a concurrent dup, the loser's transaction rolls back its decrement too).
+
+Runs on a PRIVILEGED database role (the money path is infrastructure) and scopes by
+``tenant_id`` explicitly; row-level security on the tables is defence-in-depth — hence the
+per-transaction bind of the ``app.current_tenant`` GUC so those privileged writes still
+satisfy the tenant-isolation policy on any Postgres where the role is not BYPASSRLS.
+
+Extracted from PANTHEON (a multi-tenant AI substrate). See README.md for the design notes.
+"""
+from __future__ import annotations
+
+import logging
+import os
+import uuid
+from dataclasses import dataclass
+
+from sqlalchemy import create_engine, text
+from sqlalchemy.exc import IntegrityError
+
+log = logging.getLogger("credit_ledger.metering")
+
+# The Postgres GUC (session/txn variable) that carries the current tenant for row-level security.
+# Set per-transaction by _bind so writes on a non-BYPASSRLS role satisfy the tenant_isolation policy.
+# Override via the env var if your schema uses a different GUC name.
+TENANT_GUC = os.environ.get("CREDIT_LEDGER_TENANT_GUC", "app.current_tenant")
+
+_ENGINE = None
+_DATABASE_URL = os.environ.get("DATABASE_URL")
+
+
+def configure(database_url: str) -> None:
+    """Set the default database URL (alternative to the ``DATABASE_URL`` env var). Optional: every
+    public function also accepts an explicit ``engine=`` for full control over pooling / lifecycle."""
+    global _DATABASE_URL, _ENGINE
+    _DATABASE_URL = database_url
+    _ENGINE = None
+
+
+def _engine():
+    global _ENGINE
+    if _ENGINE is None:
+        if not _DATABASE_URL:
+            raise RuntimeError(
+                "No database configured — set DATABASE_URL, call configure(url), or pass engine=.")
+        _ENGINE = create_engine(_DATABASE_URL, future=True)
+    return _ENGINE
+
+
+def _bind(c, tenant_id: str) -> None:
+    """Bind the RLS tenant GUC for this transaction so a privileged engine's writes satisfy the
+    tenant_isolation policy on any Postgres where the role is NOT BYPASSRLS (every managed offering,
+    and a hardened self-host). Harmless on a superuser dev box. Without it, the decrement matches
+    zero rows and the INSERT's WITH CHECK fails the moment RLS is actually enforced — billing
+    silently bricks in prod."""
+    c.execute(text("SELECT set_config(:g, :t, true)"), {"g": TENANT_GUC, "t": str(tenant_id)})
+
+
+@dataclass
+class Charge:
+    ok: bool
+    balance: float
+    deduped: bool = False
+
+
+def balance(tenant_id: str, *, engine=None) -> float:
+    eng = engine or _engine()
+    with eng.begin() as c:
+        _bind(c, tenant_id)
+        v = c.execute(text("SELECT credits FROM credit_balance WHERE tenant_id = :t"),
+                      {"t": str(tenant_id)}).scalar()
+    return float(v) if v is not None else 0.0
+
+
+def grant(tenant_id: str, amount: float, *, reason: str = "grant",
+          idempotency_key: str | None = None, engine=None) -> Charge:
+    """Add credits. Idempotent on ``idempotency_key`` (e.g. a Stripe ``event_id``): a retried or
+    concurrently-replayed grant is a no-op, not a double-credit or a 500. The ledger's
+    ``UNIQUE(tenant_id, idempotency_key)`` is the real guarantee; the retry loop turns a racing
+    duplicate (which would otherwise surface an IntegrityError) into a clean dedup."""
+    if not amount > 0:                       # invariant at the primitive: a non-positive grant would
+        raise ValueError(f"grant amount must be positive, got {amount!r}")  # decrement/brick a tenant.
+    eng = engine or _engine()
+    for _attempt in range(2):
+        try:
+            with eng.begin() as c:
+                _bind(c, tenant_id)
+                if idempotency_key:
+                    prior = c.execute(text("SELECT balance_after FROM credit_ledger "
+                                           "WHERE tenant_id=:t AND idempotency_key=:k"),
+                                      {"t": str(tenant_id), "k": idempotency_key}).scalar()
+                    if prior is not None:
+                        return Charge(ok=True, balance=float(prior), deduped=True)
+                bal = c.execute(text(
+                    "INSERT INTO credit_balance (tenant_id, credits) VALUES (:t, :a) "
+                    "ON CONFLICT (tenant_id) DO UPDATE SET credits = credit_balance.credits + :a, "
+                    "updated_at = now() RETURNING credits"),
+                    {"t": str(tenant_id), "a": amount}).scalar()
+                c.execute(text("INSERT INTO credit_ledger (id, tenant_id, delta, reason, idempotency_key, balance_after) "
+                               "VALUES (:id,:t,:d,:r,:k,:b)"),
+                          {"id": str(uuid.uuid4()), "t": str(tenant_id), "d": amount,
+                           "r": reason, "k": idempotency_key, "b": bal})
+                return Charge(ok=True, balance=float(bal))
+        except IntegrityError:                        # a constraint blocked it — could be a concurrent
+            continue                                  # same-key dedup OR an FK (no such tenant); decide below
+    # Both attempts hit a constraint. If a ledger row exists for this key, it was a genuine concurrent
+    # dedup (the winner committed). If NOT, nothing landed (e.g. an FK violation: the tenant doesn't
+    # exist) — fail LOUD rather than report a phantom success that silently loses money.
+    prior = None
+    if idempotency_key:
+        with eng.begin() as c:
+            _bind(c, tenant_id)
+            prior = c.execute(text("SELECT balance_after FROM credit_ledger "
+                                   "WHERE tenant_id=:t AND idempotency_key=:k"),
+                              {"t": str(tenant_id), "k": idempotency_key}).scalar()
+    if prior is not None:
+        return Charge(ok=True, balance=float(prior), deduped=True)
+    log.error("grant did NOT land for tenant %s (key=%r) — no ledger row after retries (no such tenant?)",
+              tenant_id, idempotency_key)
+    return Charge(ok=False, balance=balance(tenant_id, engine=eng))
+
+
+def decrement(tenant_id: str, amount: float = 1.0, *, reason: str = "usage",
+              idempotency_key: str | None = None, engine=None) -> Charge:
+    """Atomically spend `amount` credits. Returns ok=False (no change) if insufficient."""
+    eng = engine or _engine()
+    for _attempt in range(2):
+        try:
+            with eng.begin() as c:
+                _bind(c, tenant_id)
+                if idempotency_key:
+                    prior = c.execute(text("SELECT balance_after FROM credit_ledger "
+                                           "WHERE tenant_id=:t AND idempotency_key=:k"),
+                                      {"t": str(tenant_id), "k": idempotency_key}).scalar()
+                    if prior is not None:
+                        return Charge(ok=True, balance=float(prior), deduped=True)
+                bal = c.execute(text(
+                    "UPDATE credit_balance SET credits = credits - :n, updated_at = now() "
+                    "WHERE tenant_id = :t AND credits >= :n RETURNING credits"),
+                    {"t": str(tenant_id), "n": amount}).scalar()
+                if bal is None:                       # insufficient — no change
+                    cur = c.execute(text("SELECT credits FROM credit_balance WHERE tenant_id=:t"),
+                                    {"t": str(tenant_id)}).scalar()
+                    return Charge(ok=False, balance=float(cur) if cur is not None else 0.0)
+                c.execute(text("INSERT INTO credit_ledger (id, tenant_id, delta, reason, idempotency_key, balance_after) "
+                               "VALUES (:id,:t,:d,:r,:k,:b)"),
+                          {"id": str(uuid.uuid4()), "t": str(tenant_id), "d": -amount,
+                           "r": reason, "k": idempotency_key, "b": bal})
+                return Charge(ok=True, balance=float(bal))
+        except IntegrityError:                        # concurrent same idempotency_key
+            continue                                  # txn rolled back (incl. decrement); retry → dedup
+    with eng.begin() as c:
+        _bind(c, tenant_id)
+        prior = c.execute(text("SELECT balance_after FROM credit_ledger "
+                               "WHERE tenant_id=:t AND idempotency_key=:k"),
+                          {"t": str(tenant_id), "k": idempotency_key}).scalar()
+    if prior is None:                                    # IntegrityError but NO prior row → not a real dedup;
+        return Charge(ok=False, balance=balance(tenant_id, engine=eng))   # don't report a phantom success
+    return Charge(ok=True, balance=float(prior), deduped=True)

credit_ledger-0.1.0/credit_ledger.egg-info/PKG-INFO

@@ -0,0 +1,127 @@
+Metadata-Version: 2.4
+Name: credit-ledger
+Version: 0.1.0
+Summary: Atomic, overdraft-proof, idempotent credit metering over Postgres. Extracted from PANTHEON.
+Author: Isaac Teague Frayling
+License: Apache-2.0
+Project-URL: Homepage, https://pantheonlabs.info
+Keywords: metering,billing,credits,idempotency,postgres,multi-tenant,row-level-security
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: SQLAlchemy>=2.0
+Provides-Extra: postgres
+Requires-Dist: psycopg[binary]>=3.1; extra == "postgres"
+Provides-Extra: test
+Requires-Dist: pytest>=7; extra == "test"
+Dynamic: license-file
+
+# credit-ledger
+
+[![tests](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml/badge.svg)](https://github.com/Igfray/pantheon-credit-ledger/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![Python](https://img.shields.io/pypi/pyversions/credit-ledger.svg)](https://pypi.org/project/credit-ledger/)
+[![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](LICENSE)
+
+**An atomic, overdraft-proof, idempotent credit ledger — for metering money on a pay-per-use or multi-tenant service.** ~150 lines of Python over Postgres. Extracted from [PANTHEON](https://pantheonlabs.info), a multi-tenant AI substrate, where it's the money path for autonomous AI agents that spend real credits on every turn.
+
+The interesting part isn't the size — it's that three properties that are usually gotten *subtly wrong* are each guaranteed by a single, boring database mechanism instead of application-level hope:
+
+| Property | How | Not by |
+|---|---|---|
+| **No overdraft, ever** — a balance can never go negative, even under a stampede of concurrent charges | one guarded `UPDATE … WHERE credits >= n` | `SELECT` then `UPDATE` (a race), or a mutex (doesn't survive multiple processes) |
+| **Exactly-once under retries** — a replayed charge (a Stripe webhook fired twice, a client retry) charges once | `UNIQUE(tenant_id, idempotency_key)` on an append-only ledger | de-duping in app code (another race) |
+| **Tenant isolation** — tenant A can't read or write tenant B's balance | Postgres row-level security, `FORCE`d | a `WHERE tenant_id = ?` you have to remember to add every time |
+
+## The core: one statement does the hard part
+
+A spend is a single guarded, atomic `UPDATE`:
+
+```sql
+UPDATE credit_balance
+   SET credits = credits - :n
+ WHERE tenant_id = :t AND credits >= :n
+RETURNING credits;
+```
+
+- The row lock **serialises** concurrent charges against the same tenant — no lost updates.
+- The `credits >= :n` guard makes overdraft **structurally impossible**: the charge either fully succeeds (a row comes back) or changes nothing (zero rows). There is no window between the check and the debit, because they're the same statement.
+
+That's the whole trick, and the test proves it: fire **100 concurrent spends of 1 credit against a balance of 50**, and *exactly 50* succeed — the balance lands at 0, never negative, with 50 ledger rows.
+
+```python
+# tests/test_metering.py
+def test_atomic_decrement_under_concurrency(engine, tenant):
+    metering.grant(tenant, 50, engine=engine)
+    with ThreadPoolExecutor(max_workers=20) as ex:
+        results = list(ex.map(lambda _: metering.decrement(tenant, 1, engine=pool).ok, range(100)))
+    assert sum(results) == 50                              # exactly the balance, no more
+    assert metering.balance(tenant, engine=engine) == 0.0  # never negative
+```
+
+## Idempotency: the ledger *is* the dedup
+
+Every grant and spend appends a row to `credit_ledger`, which carries a `UNIQUE(tenant_id, idempotency_key)`. Pass an `idempotency_key` (a Stripe `event_id`, a request id) and a replay can't insert a second row — the database rejects it, and the code returns the *original* result marked `deduped=True`. Money is never double-counted, and a duplicate never surfaces as a 500.
+
+The subtle bit is the concurrent race: two identical charges arrive at once, both past the "already seen?" check, both try to insert. One wins; the other's `INSERT` violates the constraint, which **rolls back its whole transaction — including the decrement**. The loser then re-reads the ledger, finds the winner's row, and returns that. So a concurrent duplicate is a clean dedup, not a double-charge and not an error. There's a test for exactly this — 20 concurrent replays of one Stripe event → *one* credit applied, zero exceptions.
+
+## The correctness detail most people miss: fail loud, not phantom-success
+
+An `IntegrityError` after retries is ambiguous — it could be a genuine concurrent dedup (the winner committed, there's a ledger row), **or** it could be a foreign-key violation because the tenant doesn't exist (nothing landed). Reporting success in the second case would silently lose money. So the code disambiguates by re-reading the ledger: a row means real dedup → report success; no row means nothing happened → **fail loud** (`ok=False`). This came out of an adversarial self-audit of the substrate; it's the difference between "looks fine" and "correct."
+
+## The RLS footgun this quietly avoids
+
+Row-level security is only a guarantee if the writer is actually subject to it. Metering runs on a *privileged* app role (the money path is infrastructure), and on any managed Postgres that role is **not** `BYPASSRLS`. So before every transaction the code binds the tenant into a session variable the policy reads:
+
+```python
+c.execute(text("SELECT set_config('app.current_tenant', :t, true)"), {"t": tenant_id})
+```
+
+Skip this and, the moment RLS is enforced in production, the guarded `UPDATE` matches zero rows and the `INSERT`'s `WITH CHECK` fails — **billing silently bricks**, while every dev box (superuser, RLS bypassed) looks perfectly fine. Getting this right is the difference between a demo and something you'd run.
+
+## Why so little code
+
+The database does the concurrency, the atomicity, and the isolation. The code's entire job is to *use those primitives correctly* — bind the tenant, guard the decrement, key the ledger, and disambiguate the one genuinely-ambiguous failure. Fewer moving parts is the point, not an accident.
+
+## Use it
+
+```bash
+pip install -e .                       # SQLAlchemy is the only runtime dep
+createdb ledger && psql "$DATABASE_URL" -f schema.sql
+```
+
+```python
+from credit_ledger import metering
+
+metering.grant(tenant_id, 100, idempotency_key="stripe:evt:abc")  # top up (Stripe-retry-safe)
+c = metering.decrement(tenant_id, 1, reason="chat-turn")          # spend
+if not c.ok:
+    ...  # out of credits — nothing was charged
+metering.balance(tenant_id)                                       # -> float
+```
+
+Every function also takes an explicit `engine=` (for your own pool/lifecycle). `schema.sql` is the full Postgres schema including the RLS policies.
+
+## Run the tests
+
+```bash
+DATABASE_URL=postgresql://localhost/ledger pytest -q     # needs Postgres + schema.sql applied
+```
+
+The suite is the spec: overdraft impossibility under 100-way concurrency, idempotent spend and grant retries, concurrent Stripe-replay dedup, and rejection of non-positive grants (a negative "grant" would decrement a tenant — the primitive refuses it before touching the DB).
+
+## Design notes
+
+A few deliberate choices, and the answers to the sharp questions they invite:
+
+- **A failed (insufficient) charge records nothing — so a retry with the same key re-attempts. That's intended.** Idempotency dedups *successful effects*, not attempts. A charge rejected for insufficient balance didn't happen, so the same key should be free to succeed later — e.g. a metered webhook that hit a zero balance and is retried after a top-up. The dedup keys the *committed ledger row*, which only exists for charges that actually landed.
+- **`numeric(18,4)` in the database, `float` at the API boundary.** The stored balance and every ledger delta are exact decimals — no binary-float drift on the money itself. The functions return `float` for ergonomics, which is fine for integer-ish credits; if you were metering *actual currency*, return `Decimal` at the boundary too (a one-line change) so nothing rounds on the way out.
+- **Why two attempts is enough** (`for _attempt in range(2)`). Under N concurrent charges sharing one key, exactly one wins the `UNIQUE(tenant_id, idempotency_key)` insert and commits; every loser's `INSERT` raises, rolling back its whole transaction (decrement included). On the second pass the losers see the winner's now-committed ledger row in the "already seen?" read and return *that* — a clean dedup. A third attempt can't be needed: once the winner commits, the row is visible, so attempt two is always either the winner's own success or a loser's dedup. The re-read *after* the loop only exists to distinguish the genuinely-nothing-landed case (e.g. a foreign-key violation) — so a constraint error is never reported as phantom success.
+
+## Context
+
+This is one self-contained piece of **[PANTHEON](https://pantheonlabs.info)** — a multi-tenant substrate for running AI agents safely on other people's money and data (tenancy + RLS isolation, a governed reasoning loop, an approval queue, this metering ledger, and a quality gate), designed and built solo. The live Studio at [pantheonlabs.info](https://pantheonlabs.info) runs on it; the flagship write-up is at [pantheonlabs.co.uk](https://pantheonlabs.co.uk). Built by Isaac Teague Frayling.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE).

credit_ledger-0.1.0/credit_ledger.egg-info/SOURCES.txt

@@ -0,0 +1,11 @@
+LICENSE
+README.md
+pyproject.toml
+credit_ledger/__init__.py
+credit_ledger/metering.py
+credit_ledger.egg-info/PKG-INFO
+credit_ledger.egg-info/SOURCES.txt
+credit_ledger.egg-info/dependency_links.txt
+credit_ledger.egg-info/requires.txt
+credit_ledger.egg-info/top_level.txt
+tests/test_metering.py

credit_ledger-0.1.0/credit_ledger.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

credit_ledger-0.1.0/credit_ledger.egg-info/requires.txt

@@ -0,0 +1,7 @@
+SQLAlchemy>=2.0
+
+[postgres]
+psycopg[binary]>=3.1
+
+[test]
+pytest>=7

credit_ledger-0.1.0/credit_ledger.egg-info/top_level.txt

@@ -0,0 +1 @@
+credit_ledger

credit_ledger-0.1.0/pyproject.toml

@@ -0,0 +1,24 @@
+[project]
+name = "credit-ledger"
+version = "0.1.0"
+description = "Atomic, overdraft-proof, idempotent credit metering over Postgres. Extracted from PANTHEON."
+readme = "README.md"
+requires-python = ">=3.10"
+license = { text = "Apache-2.0" }
+authors = [{ name = "Isaac Teague Frayling" }]
+keywords = ["metering", "billing", "credits", "idempotency", "postgres", "multi-tenant", "row-level-security"]
+dependencies = ["SQLAlchemy>=2.0"]
+
+[project.optional-dependencies]
+postgres = ["psycopg[binary]>=3.1"]
+test = ["pytest>=7"]
+
+[project.urls]
+Homepage = "https://pantheonlabs.info"
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools]
+packages = ["credit_ledger"]

credit_ledger-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

credit_ledger-0.1.0/tests/test_metering.py

@@ -0,0 +1,125 @@
+# Copyright 2026 Isaac Teague Frayling
+# Licensed under the Apache License, Version 2.0 (see LICENSE).
+"""Atomic credit metering — no overdraft under concurrency, idempotent retries.
+
+Needs a Postgres reachable at $DATABASE_URL with schema.sql applied:
+    createdb ledger_test
+    psql "$DATABASE_URL" -f schema.sql
+    DATABASE_URL=postgresql://localhost/ledger_test pytest -q
+"""
+from __future__ import annotations
+
+import os
+import uuid
+from concurrent.futures import ThreadPoolExecutor
+
+import pytest
+from sqlalchemy import create_engine, text
+
+from credit_ledger import metering
+
+DB_URL = os.environ.get("DATABASE_URL")
+
+
+@pytest.fixture(scope="module")
+def engine():
+    if not DB_URL:
+        pytest.skip("set DATABASE_URL to a Postgres with schema.sql applied")
+    eng = create_engine(DB_URL, future=True)
+    try:
+        with eng.connect() as c:
+            c.execute(text("SELECT 1 FROM credit_balance LIMIT 0"))   # schema applied?
+    except Exception as exc:  # noqa: BLE001
+        pytest.skip(f"apply schema.sql first: {exc}")
+    return eng
+
+
+@pytest.fixture()
+def tenant(engine):
+    tid = uuid.uuid4()
+    with engine.begin() as c:
+        c.execute(text("INSERT INTO tenants (id, slug, name) VALUES (:id,:s,:n)"),
+                  {"id": str(tid), "s": f"t-{tid.hex[:8]}", "n": f"t-{tid.hex[:8]}"})
+    yield str(tid)
+    with engine.begin() as c:
+        c.execute(text("DELETE FROM tenants WHERE id=:id"), {"id": str(tid)})   # cascades
+
+
+def test_grant_and_spend(engine, tenant):
+    assert metering.grant(tenant, 10, engine=engine).balance == 10.0
+    r = metering.decrement(tenant, 3, engine=engine)
+    assert r.ok and r.balance == 7.0
+    assert metering.balance(tenant, engine=engine) == 7.0
+
+
+def test_insufficient_is_rejected_without_change(engine, tenant):
+    metering.grant(tenant, 2, engine=engine)
+    r = metering.decrement(tenant, 5, engine=engine)
+    assert r.ok is False and r.balance == 2.0              # unchanged
+    assert metering.balance(tenant, engine=engine) == 2.0
+
+
+def test_idempotent_retry_charges_once(engine, tenant):
+    metering.grant(tenant, 10, engine=engine)
+    a = metering.decrement(tenant, 4, idempotency_key="charge-1", engine=engine)
+    b = metering.decrement(tenant, 4, idempotency_key="charge-1", engine=engine)   # retry
+    assert a.ok and b.ok and b.deduped is True
+    assert metering.balance(tenant, engine=engine) == 6.0  # charged once, not twice
+
+
+def test_grant_idempotent_retry_credits_once(engine, tenant):
+    a = metering.grant(tenant, 100, idempotency_key="stripe:evt:abc", engine=engine)
+    b = metering.grant(tenant, 100, idempotency_key="stripe:evt:abc", engine=engine)   # Stripe retry
+    assert a.ok and b.ok and b.deduped is True
+    assert metering.balance(tenant, engine=engine) == 100.0   # credited once, not 200
+
+
+# ── ★ atomic under concurrency: no overdraft, no double-spend ─────────────────
+def test_atomic_decrement_under_concurrency(engine, tenant):
+    metering.grant(tenant, 50, engine=engine)
+    # 100 concurrent spends of 1 against a balance of 50 → exactly 50 succeed
+    pool_engine = create_engine(DB_URL, future=True, pool_size=20, max_overflow=20)
+
+    def spend(_i):
+        return metering.decrement(tenant, 1, reason="race", engine=pool_engine).ok
+
+    with ThreadPoolExecutor(max_workers=20) as ex:
+        results = list(ex.map(spend, range(100)))
+
+    pool_engine.dispose()
+    assert sum(results) == 50                              # exactly the balance, no more
+    assert metering.balance(tenant, engine=engine) == 0.0  # never negative
+    with engine.begin() as c:
+        spent = c.execute(text("SELECT count(*) FROM credit_ledger "
+                               "WHERE tenant_id=:t AND reason='race'"), {"t": tenant}).scalar()
+    assert spent == 50                                     # one ledger row per successful spend
+
+
+# ── ★ grant is retry-safe under concurrency: a replayed Stripe event credits once ─
+def test_grant_idempotent_under_concurrency(engine, tenant):
+    # 20 concurrent grants of the SAME event_id → exactly one credit, and none raises / 500s
+    pool_engine = create_engine(DB_URL, future=True, pool_size=20, max_overflow=20)
+
+    def give(_i):
+        return metering.grant(tenant, 100, reason="stripe", idempotency_key="stripe:evt:race",
+                              engine=pool_engine)
+
+    with ThreadPoolExecutor(max_workers=20) as ex:
+        results = list(ex.map(give, range(20)))
+
+    pool_engine.dispose()
+    assert all(r.ok for r in results)                     # every call returned a clean Charge, no 500
+    assert metering.balance(tenant, engine=engine) == 100.0   # exactly one grant applied, not 20
+    with engine.begin() as c:
+        rows = c.execute(text("SELECT count(*) FROM credit_ledger "
+                              "WHERE tenant_id=:t AND idempotency_key='stripe:evt:race'"),
+                         {"t": tenant}).scalar()
+    assert rows == 1                                       # one ledger row, not 20
+
+
+@pytest.mark.parametrize("bad", [0, -1, -100.0])
+def test_grant_rejects_non_positive_amounts(bad):
+    # A non-positive grant would DECREMENT (brick) a tenant or 500; the primitive must refuse it loudly,
+    # before touching the DB — so the invariant lives at the money primitive, not only in each caller.
+    with pytest.raises(ValueError):
+        metering.grant("00000000-0000-0000-0000-000000000000", bad, reason="bad", engine="UNUSED")