ADR-0015: Make budgets explicit, separate and actionable¶

• Status: Accepted — baseline plus owner capacity decision on 2026-09-28.
• Recorded: 2026-09-29. This is the recording date, not an invented original approval date.
• Authority and implementation boundaries: see Evidence below and the register conventions.

Context¶

Hard-coded helper ceilings and borrowed profiling limits cannot establish end-to-end transformation capacity. Raising disclosure budgets to fit a large dataset would weaken a different boundary.

Decision¶

Bound bytes, rows, columns, cells, scalar width, parsing complexity, expansion and cumulative work/time at their owning stages. Distinguish input/result/output/transport budgets and source-free profiling from transformation capacity. Fail closed without silent truncation or automatic limit increases.

Transformation target: 1,000,000 rows × 100 columns; required fictional acceptance: 300,000 × 50. Arbitrary-width values still must fit byte/time limits. Support session/run and saved behavior-profile settings. Errors distinguish requested_above_limit from actual limit_exceeded and include safe dimension, amount, threshold, units, origin and supported recovery settings. Invalid configuration is separate.

Alternatives and consequences¶

Removing caps or increasing only one constant does not meet the decision. Adding unapproved hard-RSS/exact-wire guarantees is also not selected. Decoded payload measurements and cooperative deadlines must not be described as stronger guarantees.

The private installed CSV replacement route has 300k×50 evidence; SQL throughput, all routes and public activation are not thereby accepted. Local query-worker changes remain implementation work, not a released contract.

Evidence¶

Sources: configuration, operational budgets, owner capacity contract. Evidence: CSV scale acceptance; limit tests.

Revisit when¶

Any capacity change needs end-to-end evidence and separate disclosure review; defaults, supported capacity and measured capacity must stay distinguishable.