Simulation table — end-to-end coverage

Antares currently produces two output systems in parallel:

• the historical mc-all / mc-ind text tree, built by the src/solver/variable/ template machinery — this is the production output;
• the flat simulation table (simulation-table-<year>-optim-nb-<n>.csv / .parquet), written only when the solver runs with --output=all or --output=simulation-tables.

The simulation table is not used in production yet. This page tracks the e2e coverage that has to be in place before it can be.

Test assets

Layer	Location
C++ unit tests	src/tests/src/io/outputs/testSimulationTable.cpp, test-parquet-simulation-table-writer.cpp
Worked-example e2e (hand-picked hours, inline expected values)	src/tests/cucumber/features/solver-features/legacy_simulation_table.feature
Systematic legacy ⇔ table equivalence	src/tests/cucumber/features/solver-features/legacy_simulation_table_equivalence.feature + features/steps/common_steps/simulation_table_equivalence.py
Parquet round-trip (pure modeler)	src/tests/cucumber/features/modeler-features/test_launcher_parquet.feature
Folder diff vs inline reference	hybrid_studies.feature (simulation tables match the references)

The equivalence check

the simulation table matches the legacy mc-ind output for year N walks a declarative mapping over every area / thermal cluster / short-term-storage cluster / link of the study and asserts, for each timestep the table covers, that the table value equals its mc-ind counterpart within a per-quantity tolerance.

Design points:

• reads the final optimisation pass — optim-nb-2 when present, else optim-nb-1. optim-nb-1 is the LP relaxation and diverges from mc-ind on unit-commitment studies;
• on a study where a later resolution stage (peak-shaving, adq-patch) changes the numbers mc-ind actually prints, the optim-nb-* default is wrong — the simulation table for stage "<stage>" matches the legacy mc-ind output for year N reads a named stage's table instead. Two scenarios exercise this: "Hydro remix (peak-shaving)" (RemixHydroPostProcessCmd reshuffling hydro generation moves unsupplied_energy / spilled_energy / price) and "Adequacy patch (CSR)" (see below);
• mc-ind folders are 1-based, the table's scenario_index is 0-based (year N ⇒ mc-ind/0000N ⇔ simulation-table-(N-1)-...);
• no frozen reference file — it compares two live outputs, so it keeps working while the table layout/naming still changes;
• basis_status is never compared (see CHANGELOG #3233);
• tolerances are generous because mc-ind hourly columns are printed rounded; the table carries full precision.

Quantity coverage

Quantity	Table output	mc-ind source	Status
Unsupplied energy	unsupplied_energy	area values-hourly UNSP. ENRG	✅ equivalence-checked
Spilled energy	spilled_energy	area values-hourly SPIL. ENRG	✅ equivalence-checked
Area marginal price	price	area values-hourly MRG. PRICE	✅ equivalence-checked — dual-derived, see gaps
Actual load	actual_load	area values-hourly LOAD	✅ equivalence-checked
Thermal generation	generation_power	area details-hourly <cluster> / MWh	✅ equivalence-checked (optim-nb-2)
Units on (ceil)	actual_num_units_on	area details-hourly <cluster> / NODU	✅ equivalence-checked — skipped when [other preferences] unit-commitment-mode is fast (Mapping.requires_accurate_uc); the ST value is ceil(x(NumberOfDispatchableUnits)) while mc-ind's NODU is the fast-UC heuristic's own unit count, two independently-computed integers that can differ by 1 on borderline hours
STS injection / withdrawal / level	injection_power / withdrawal_power / level	area details-STstorage-hourly	✅ equivalence-checked, including studies where the cluster's id and display name differ
Link flow / abs / minus / loop	flow / abs_flow / minus_flow / actual_loop_flow	link values-hourly FLOW LIN. / LOOP FLOW	✅ equivalence-checked
Raw num_units_on (LP value)	num_units_on	—	⛔ not equivalence-checked — limitation: fractional LP value, no integer mc-ind column to compare to (actual_num_units_on is the checked, rounded counterpart)
Derived costs (prop_cost, non_prop_cost, imbalance_cost, …)	—	—	⛔ not equivalence-checked — limitation: no mc-ind column at all (mc-ind never printed these); values also carry anti-degeneracy noise. Tested instead as hand-derived worked examples in legacy_simulation_table.feature
Emissions (co2_emissions, …)	*_emissions	area values-hourly CO2 EMIS. (area total only)	⛔ not equivalence-checked — limitation: mc-ind only prints an area-level total, not per-cluster, so per-cluster ST rows have nothing to diff against. Tested instead as a worked example in legacy_simulation_table.feature
Hydro level_percentage, actual_inflows, hydro_shadow_price, bellman_value	—	H. LEV is absolute, not a direct match	⛔ not equivalence-checked — limitation: H. LEV is a different (absolute, not percentage) quantity. Tested instead as a worked example in legacy_simulation_table.feature
Congestion fees (abs_congestion_fee, alg_congestion_fee)	abs_congestion_fee / alg_congestion_fee	link values-hourly CONG. FEE (ABS./ALG.)	✅ equivalence-checked — dual-derived, see gaps
Reserve area spilled/unsupplied energy	spilled_energy_reserve_<id> / unsupplied_energy_reserve_<id>	area values-hourly <reserve>_SPIL. / <reserve>_UNSP.	✅ equivalence-checked
Reserve thermal participation (on/off units)	units_on_reserve_power_<id> / units_off_reserve_power_<id>	area details-hourly <reserve>_<cluster> / <reserve>_<cluster>_off	✅ equivalence-checked
Reserve costs (reserve_imbalance_cost_<id>, reserve_participation_cost_<id>)	—	—	⛔ not equivalence-checked — limitation: no mc-ind column (same reason as the derived costs above)
Thermal reserve total participation (reserve_power_<id> on a thermal component)	—	—	⛔ not equivalence-checked — redundant, not a limitation: always equals units_on_reserve_power + units_off_reserve_power, both of which are already checked individually
STS reserve participation (net)	reserve_power_<id> on ..._short_term_storage_<cluster>	area details-hourly <reserve>_<cluster>	✅ equivalence-checked — unlike thermal, mc-ind only exposes the net (signed release-minus-store) value for STS, so there's nothing to check reserve_released_power_<id> / reserve_stored_power_<id> against individually (see below)
Hydro reserve participation (net)	reserve_power_<id> on ..._hydro_storage	area details-hourly <reserve>_Hydro	✅ equivalence-checked — same net-only caveat as STS above
STS/hydro reserve released/stored power (reserve_released_power_<id>, reserve_stored_power_<id>)	—	—	⛔ not equivalence-checked — limitation: no mc-ind column; mc-ind only ever exposes the net participation (see reserve_power_<id> rows above)

Known gaps

• MIP / MILP weeks — dual-derived rows (price, STS profit, hydro_shadow_price) are zero on weeks solved as MIP because duals are not extracted there. See docs/architecture/legacy-extra-outputs-spec-checklist.md.
• Adequacy patch — domestic_unsupplied_energy (DENS), is_local_matching_rule_violated (LMR.VIOL) blocked on ANT-5240; CSR outputs out of scope. The rest of LEGACY_TO_ST (unsupplied_energy, price, flow, …) is equivalence-checked on a CSR study, scoped to the adq-patch stage table — see "Adequacy patch (CSR)" in legacy_simulation_table_equivalence.feature, which reads that stage because mc-ind prints the post-CSR numbers, not optim-nb-2's.
• ~~abs_flow / minus_flow / abs_congestion_fee / alg_congestion_fee on the adq-patch stage~~ — fixed. These were wrong on the congested link inside a CSR patch, found by the "Adequacy patch (CSR)" scenario. LegacyExtraOutputEmitter::linkOutputs() computes them from x(variableManager_.DirectFlow(interco, pdt)). The adequacy patch's CSR post-process (HourlyCSRProblem::constructVariableFlows, adequacy_patch_csr/construct_problem_variables.cpp) repoints that hour's CorrespondanceVarNativesVarOptim entry at an index in its own short-lived per-hour LP while building it, and nothing put it back once CSR moved on, so by dump time most CSR-touched hours resolved that index against the original weekly problem's variable table — unrelated to CSR's flow. (Reading ValeursDeNTC[pdt].ValeurDuFlux[interco] instead, mirroring how actual_loop_flow is read two lines below, looked like a tempting one-line fix and is what both the ordinary weekly solve and CSR publish their DirectFlow result to — but it broke the ordinary, non-post-process dump path: that dump runs before this week's own publish-to-address step catches up, so ValeurDuFlux there is one dump cycle stale while x(DirectFlow) is fresh. Do not reintroduce that swap.) Fixed instead at the source, in HourlyCSRProblem::run() (adequacy_patch_csr/adq_patch_curtailment_sharing.cpp): it now saves the hour's original CorrespondanceVarNativesVarOptim indices (still valid, still correctly published) before building its own per-hour problem and restores them once that problem is solved and its result published, so nothing outside CSR's own solve ever resolves against its numbering.
• Coverage is hybrid-only. Pure-classic studies produce a sparse table (SimulationTableWriter throws on an empty table); pure-modeler has only the Parquet round-trip test.
• No frozen-reference regression in src/tests/run-study-tests/ — deferred until the table file layout (feature/st_stage_selection: per-MC-year files, per-stage tables) stabilises.

Adding coverage

• New quantity, direct mc-ind counterpart — add a Mapping(...) row to LEGACY_TO_ST in simulation_table_equivalence.py. Pick a study that exercises it, confirm the sign and the mc-ind print precision, set atol / rtol accordingly.
• New study — add a Scenario to legacy_simulation_table_equivalence.feature pointing at a study already in the Antares_Simulator_Tests_NR submodule; --output=all plus the existing year-by-year forcing in init_simulation is enough.
• Quantity with no mc-ind counterpart (derived costs, hydro %) — keep it as a worked example in legacy_simulation_table.feature with an inline expected value and a comment deriving it.