Pre-Release Notes

Unreleased

This pre-release changes the following contracts without compatibility shims:

Values owned by a Stream and its segment records are read-only. Mutations use explicit stream assignment, indexed-value, or segment-update APIs.
Period weights use one validation policy: omitted trailing weights become 1.0; excess, non-finite, negative, and all-zero vectors are rejected.
Structured process-stream and nested thermal-profile inputs reject unknown fields. Process streams accept the canonical name and heat_capacity_flowrate spellings only.
Workspace bundles use schema version 2 and case_input. Version 1, unknown versions, and the retired payload field are rejected.
Segmented process streams and utilities share the same semantic validation in reports and preparation, including parent aggregate consistency.

Private class helpers are grouped under owner-oriented packages for streams, values, collections, problem tables, problem orchestration, workspaces, and heat exchangers.
Runtime stream-segment, exchanger-period, and exchanger-area-slice record classes are parent-owned implementation details. Construct them through Stream and HeatExchanger mappings; StreamSegmentSchema remains the supported external segment input contract.
The former public record imports and their Python pickle paths are removed without aliases or compatibility shims.
Synthesis schemas now have concrete common, topology, method, task, and result owners. The compatibility-only methods, tasks, and results modules and the synthesis package re-export barrel are removed. Import concrete owner modules instead; old barrel-qualified pickle paths are unsupported.
Intentional public barrels at OpenPinch, OpenPinch.lib, and OpenPinch.lib.schemas remain, but their lazy synthesis exports resolve directly to concrete owner modules.
Process-MVR records, multiperiod HPR period cases, dashboard graph state, graph specifications/metadata, and HEN solver runtime records are private to their owning services. Documented parent components, schemas, direct-MVR models, and HEN equation-model classes remain available.

PDM decompositions expose ordered period_targets with explicit period identities and indices. Clipped temperatures and active-stream flags are period-indexed; the retired singular target and clipped-temperature fields do not exist.
Each operating period uses its own utility targets and pinch temperature. Shared topology is the union of streams and matches active in any period, and a pinch side is solved when any period requires it.
Above- and below-pinch amalgamation retains every period’s duties, temperatures, approach variables, split fractions, and explicit non-isothermal branch outlet temperatures.
Non-isothermal warm starts normalize hot split fractions across cold matches and cold split fractions across hot matches independently in every period.
Segmented utilities use local per-segment dt_cont values. The inlet uses the first segment contribution, the solved match outlet uses the traversed segment contribution, and an exact boundary uses the larger adjacent value.

HeatExchanger retains shared topology, design area, and capital fields. Operational duty, activity, approaches, split fractions, and source/sink temperatures are stored in non-empty ordered period_states containing parent-owned period-state records.
The retired exchanger-level operating scalar fields do not exist. Use exchanger.state(period_id); omission is accepted only for an exchanger with exactly one period state.
Multiperiod duty, temperature, diagram, export, and controllability queries require period_id. No implicit period-zero selection is provided.
Extraction walks every period array, retains matches active only outside the first period, and prefers explicit non-isothermal branch outlet temperatures. Solved branch split fractions keep downstream duty checks physically valid.

Multiperiod summary replay captures one baseline zone and solves every period against a fresh deep copy. The original zone object, cached results, and recorded targeting specification are restored after success or failure.
Shared simulated-HPR candidates are ranked by weighted operating cost plus weighted feasibility penalty plus maximum annualized capital cost. Weighted backend obj is used only for backends that provide no cost breakdown.
Weighted public HPR summaries average operating fields, use the maximum capital, annualized-capital, compressor-capital, and heat-exchanger-capital fields, and recompute total annualized cost as weighted operating cost plus maximum annualized capital. Non-HPR fields retain weighted averaging.