Level 0 of Arc Ops made a retried tool call safe to repeat. Level 1 stopped the retries amplifying an outage. Neither bounds the bill you run up on yourself, and a max cost of five dollars is a number without a place. The place decides the outcome.
The reference workload is a month-end close: 6 steps, 8 calls, 72 billing units, 741 cents - and 400 of those 741 sit in one indivisible unit, a bulk reprice that bills once and cannot be stopped once started. That single attribute drives the whole level.
python -m arc_ops.levels.l2_cost_caps.demo # every number below, recomputed
python -m pytest -q # 369 tests, no network, no model
Repo: https://github.com/dev48v/arc-ops - PUBLIC, MIT, dependencies = [], 369 pytest with 119 in the five new L2 files, and a page that fetches nothing off itself: https://dev48.infy.uk/arcops/level2-cost-caps.html
A dashboard is not a cap. At a ceiling of 250, the observer that reads the meter once at the end bills the full 741; the best observer money can buy - poll every tick, zero aggregation lag, publish per chunk - bills 521, and still overruns, because the 400-cent call had already started. Hold the lag at exactly one tick and sweep the ceiling across all 741 places it could sit: the marginal cost of that identical delay runs from 3 cents to 400, a 133x spread with ticks_late == 1 in every row. Seconds are the wrong unit for a budget guarantee.
Fine granularity was meant to be the fix.
| check at | spend at ceiling 250 | mean overrun over 741 ceilings | worst |
|---|---|---|---|
| run end (the control) | 741 | 370.00 | 740 |
| per step | 581 | 163.65 | 459 |
| per call | 521 | 129.96 | 399 |
| per chunk | 521 | 108.66 | 399 |
Per-chunk is byte-identical to per-call at 445 of the 741 ceilings and shares its worst case, because the expensive call is the uninterruptible one.
The L1 inversion repeated, and I did not see it coming
The control is no cap at all: 200 tasks delivered, 209,244 cents, 1,046.22 each. Sweep the ceiling and not one value improves that number. Seven make it worse.
| ceiling | completed | wasted | cost per delivered task |
|---|---|---|---|
| 400 | 24 | 90% | 3,917.96 (3.74x) |
| 1,000 | 119 | 50% | 1,461.47 |
| 2,000 | 195 | 5% | 1,065.83 |
| none | 200 | 0% | 1,046.22 |
At cap 400, ninety per cent of what the fleet spent bought nothing. A cap does not reduce cost. It reclassifies it, from delivered work into waste - and whoever absorbs that is the person who asked for the task that died at ninety per cent.
Then the wreckage. Restarting the 58 aborted runs under the same ceiling finishes 0 of 58: 696 attempts, 891,924 cents, every attempt billing identically, so more attempts is not slow progress. A checkpointed resume finishes 58 of 58 at 1.08x the uncapped price. And the L0 door is right there - unkeyed restart duplicates 662,900 cents of effects, unkeyed resume 44,900, either one keyed 0.
Concurrency defeats a naive counter. Eight workers against a 1,000-cent budget commit 1,600 while the counter reports 200 - overspend invisible on its own meter.
206 in-page assertions. Next is L3, approval gates: a cap decides how much an agent may spend, and nothing here decides what it is allowed to do. Nine levels, one at a time: https://dev48.infy.uk/arcops.php
Top comments (0)