Choupo-2607

Choupo-2607 — the first public release, consolidated 2026-07-20.

Public name Choupo-2607 · tag v2607 · version 2607 · GPL-3.0-or-later
Run it in the browser at https://www.choupo.org/app/

[Choupo-2607] — 2026-07-14, consolidated 2026-07-19

Three threads under one stabilisation tag: a large open compound-library
expansion with release hygiene, the pristine-electrolyte architecture,
and the 2026-07-19 consolidation wave below.

Consolidation wave (2026-07-15 → 2026-07-19)

• v2-native case grammar. constant/thermoPhysPropDict
  (recordType thermophysicalPropertySystem; schemaVersion 2;) is THE case
  grammar: the builder assembles every equilibrium.formulation natively
  (gammaPhi / gammaGamma / diluteSolution / phiPhi /
  electrolyteGammaPhi), the v1 propertyDict reader and the translateV2
  scaffold are deleted, and a v1 case gets a named refusal pointing at the
  migrator. Active chemistry selection lives in constant/chemistryDict.
• Sealed, self-contained tutorials. bin/choupo-import mirrors each
  case's dependency closure into its own constant/ under a sha256
  propertyManifest (sealed true; forbids catalogue fallback); the corpus
  runs with data/standards/ hidden. A seal-drift gate audits sealed copies
  against the live catalogue.
• Balance diagnostics, three levels — mass, per-element atoms, energy.
  One shared formula parser (ElementComposition, IUPAC/CIAAW 2021 atomic
  weights); the steady elementBalance report is a DEFAULT diagnostic of
  every converged run (opt-out enabled false;), with
  FULL/PARTIAL/UNAVAILABLE honesty states and named refusals; batch carries
  material+energy campaign ledgers (exact state differences on the elements
  datum); choupoCtrl integrates an accepted-state ledger
  (balanceTrajectory.csv + sidecar) for mass + per-element laws, and the
  dynamicCSTR physical-energy claim honestly refuses pending its
  reformulation. elementalComposition{} gives formula-less substances a
  declared, provenance-gated composition.
• Stream-face closure. The aggregated snapshot is streamFaces/ on
  disk and faces{} in dicts; a streams {} block is refused at every node.
• PC-SAFT non-associating core (Gross & Sadowski 2001) validated ~1 %
  against literature; COSMO-SAC 2002 with named multi-set profiles.
• GUI. Global balances are the landing surface; ONE "Element balance"
  view (total atoms in/out + per-element detail, sealed only when every
  element closes); the plots sidebar shows only the open result's views;
  pinch no-recovery states conclude once with every stream drawn; the
  landing is capability-aware and phone-safe; the Explore bench synthesizes
  v2 natively. Incremental WASM build (clean 13 min → ~1 min; correct
  invalidation on sources, flags and standards content).
• Docs from a blank slate. Every LLM surface (docs/ai/, AGENTS.md,
  README, CLAUDE.md) and all seven guides teach only the v2 grammar; the
  theory guide gains "Balances: mass, elements, energy"; F1 deep-links cover
  the Control workspace and the balance family; the retired-name gate scans
  every doc surface for v1 tokens. docs/architecture/2608-handoff.md
  records the state, debts and deferrals.

Added (open compound library)

• ~28.8k group-estimated compounds under data/groupEstimative/ — identity +
  Joback / Lee-Kesler / Ambrose-Walton estimates from the open chemicals /
  thermo + RDKit toolchains, each .dat flagged as an ESTIMATE (never a
  measurement), engine-loadable, gated by check_groups (atom conservation).
  A stable nomenclature base (names / CAS / groups) students cannot silently rename.
• UNIFAC groups filled on 25 standard components (vocab-checked vs
  unifac/groups.dat).

Changed (release hygiene)

• yieldReactor built-in removed — a mass-yield split blind to atoms could
  create or destroy elements; the case-local userOp01 yield reactor stays as
  the teaching example of adding your own unit op.
• data/references/ retired (superseded by data/groupEstimative/): the
  release ships Choupo's own open estimates, not third-party pointers.
• No commercial-simulator interop in the public tree — private benchmark
  tooling and validation suites are local-only; the public repo names no
  commercial simulator.

Electrolyte architecture

The pristine-electrolyte-architecture work. Native 208/0, byte-exact
throughout. Consolidates the 8-home / 2-axis electrolyte data architecture and
retires the legacy electrolyte{} component block — the arity-1 "saco" is gone.

Status: stabilisation tag, not a feature announcement. The electrolyte
architecture is one to two days old, validated by AI-assisted design
reviews and byte-exact regression only — not yet
reviewed by a human co-author nor exercised by real use. Byte-exact 207/0
proves no regression, not architecture correctness. Announce as a milestone
only after human review + a case authored from scratch against the new tree.

Changed

• 8-home electrolyte data architecture (design settled 2026-07-01). A substance's ROLE
  (lumped / dissociated / multi-ion / molten) is chosen by the propertyPackage,
  never stored on the substance, via two orthogonal axes: REPRESENTATION (the
  package activates) × REFERENCE (the method selects one of 4 discrete rungs).
  Homes: components/ (identity + dissociatesTo) · species/ · phases/solid/
  · chemistry/ · parameters/ · propertyMethods/ · propertyPackages/ ·
  propertySets/. Canonical doc: docs/architecture/electrolyte-data-architecture.md.
• components/apparent/* deleted (the duplicate salt records); the builder
  reads salt identity + dissociatesTo from components/, solid from
  phases/solid/, anchor from chemistry/salts/.
• The legacy electrolyte{} component block RETIRED from all 21 salt .dat
  via expand-contract (shim → backfill dissociatesTo → repoint DSPM-DE +
  Crystalliser → delete). cation/anion single-sourced from dissociatesTo;
  dHsoln/solubility from chemistry/salts.
• "true species" purged → model species: species/ (was components/true/),
  phases/solid/ (was species/solids/), dissociatesTo, speciesComposition.
• halite dissolution dH corrected 1370 → 3841 J/mol (phreeqc.dat
  delta_h 0.918 kcal/mol; the old value mis-cited the source).

Added

• ENRTLMixedSolvent — the generalized segment-based eNRTL (Chen & Song 2004,
  DOI 10.1002/aic.10151) for a 1-1 salt in water+alcohol: segment local NRTL
  (ethanol = 1.811 C2H4 + 0.609 OH) + component-scale PDH + mixed-solvent
  infinite-dilution reference. New enrtlMixedSolvent props op + a validation
  tutorial against the PRIMARY Esteso 1989 gamma_pm data (47 points, 0-90 wt%
  ethanol): AAD 4.31% with ZERO refit of the published parameters (aqueous
  1.0%; the +8% at 80/90 wt% is Esteso's own measured ion pairing, declared).
• thermoTest tutorial — one system {NaCl, CaSO4, ethanol, water}, five
  models: lumped vs Pitzer evaporators (BPE 2.11 vs 5.07 K — the representation
  axis on one operation), eNRTL antisolvent crystalliser, NRTL flash, Pitzer-HMW
  speciation+scaling (gypsum/halite SI). In tutorials/{steady,props}/thermoTest/.

Limitations (honest scope — declared, not hidden)

• eNRTL is single-salt in this implementation — the Chen-Song-Evans model is
  multi-salt by design; only Choupo's ENRTLSingleSalt treats one salt per case
  (the shipped eNRTL cases are single-salt, so this is a scope limit, not an error;
  a multi-ion eNRTL is a roadmap item, mirroring PitzerHMW on the Pitzer side).
• eNRTL calorimetricFit=false — the tau(T) is not calorimetrically
  calibrated, so the apparent molar enthalpy (heat of dilution, L_phi) is gated
  OUT of the energy balance; the antisolvent-crystalliser duty is therefore
  materially approximate on the enthalpy side.
• Aqueous-only: no redox (pe), no solid solutions, no non-aqueous/molten
  electrolyte yet (the reference-rung machinery accepts it; no case ships).
• Reference-datum T-dependence is shallow (a single ΔH, not a full T-function)
  for most salts.
• The thermoPackage path and the flat-vs-identity{} component dual-reader are
  retained compat surfaces (all ~200 tutorials use thermoPackage; it is the
  legitimate degenerate propertyPackage).