# Step-10 - Verified simulated programming and recovery ## STEP IDENTIFIER Step-10. Step-09 was accepted on 2026-10-30 with no corrective assignment. This is the sole active implementation assignment. Step-11 local AI is blocked. ## STEP TITLE Rehearse approved logical calibration writes against an isolated simulation image, verifying backup, changed/unchanged readback and separately approved restore. ## OBJECTIVE One user outcome: take a saved authoring proposal through an explicit simulated target/backup/approval/write/readback/restore workflow with truthful failure states. One primary new contract: a guarded target mutation bound to an exact image revision and immutable approval, not a change to the authoring configuration. Read [STATE.md](STATE.md), [ROADMAP.md](ROADMAP.md), [authoring contracts](../docs/calibration-authoring.md) and [WORKER-QUALITY-GATE.md](WORKER-QUALITY-GATE.md). Make the frozen matrix executable before coding, self-review after first green, write `prompts/Step-10-Report.md`, and STOP. Do not begin local AI or any physical operation. ## ARCHITECTURAL CONTEXT Keep all nine projects, one Client and generic Components. Core holds reusable calibration values/definitions; Diagnostics exposes a narrow typed simulated programming target port; Simulation owns its real mutable target image. Application owns backup/approval/command/reconciliation/verification and UI owns interaction. Client composes the explicit simulation provider. Application/UI must not reference Simulation implementation, read its private state/oracle, or manufacture target readback from proposal/request/configuration. Use logical scalar/table operations, not a new wire protocol, binary offset or unrestricted command channel. No new project/package/host or generic programming/ workflow framework. Source/tests stay under `src`, tests under `src/tests`. ## CURRENT IMPLEMENTATION STATE Accepted authoring offers one immutable source-bound workspace, Baseline/Proposed revisions, exact scalar/table values/axes/diffs and guarded head commits. Worker baseline: 179 unit + 226 integration passes (405 total, zero failed/ skipped), clean nine-project Release build and native authoring checks. The orchestrator did not run them. No programming target/service exists yet. Use the actual [authoring service](../src/AutoForge.Application/Features/Vehicles/Calibration/CalibrationAuthoringService.cs), [calibration contracts](../src/AutoForge.Core/Calibration/SyntheticEngineCalibration.cs) and [configuration service](../src/AutoForge.Application/Features/Vehicles/Simulation/DynamicsConfigurationService.cs). The supported image is exactly `engine.warm-idle-target` (rpm) and `engine.torque-curve` (fixed rpm axis, N*m cells), with existing template validation. The source warm baseline is 650 rpm; the saved proposal can be 1250 rpm and 360->350 N*m at 2000 rpm. Those are synthetic values, not manufacturer data. A configuration-derived authoring baseline is not the target backup. Inspect Git/prior art and preserve accepted behavior, dirty work and historical reports. ## REQUIRED IMPLEMENTATION ### Frozen contract matrix | ID | Input/precondition | Required result | |---|---|---| | C-01 | Explicit current supported authoring workspace/config/module plus Simulation provider | Open one fresh isolated target instance at image revision 0, with complete supported source values. No implicit open/write; physical/replay/wrong-provider/source rejects before target creation or mutation. | | C-02 | Read target, create original backup and verify with a second independent read | Identity/revision/shape/axes/units/values match exactly; immutable verified volatile backup covers both keys/all cells. Wrong/incomplete/mismatched read rejects before approval; baseline is not used as a fake response. | | C-03 | Prepare/approve immutable current proposal 650->1250, then Apply | Display exact target-to-proposal diff; approval binds target/source/definition, actual before image/revision, proposal/head and action. Re-read/CAS before dispatch, one write, fresh readback; scalar=1250, every torque cell unchanged, authoring baseline/config remain 650. | | C-04 | Next saved proposal additionally changes 2000-rpm torque 360->350 | New approval binds fresh target revision; exactly that cell changes, axes and all other values stay intact; full readback equality is checked, not just acknowledgement or selected cell. | | C-05 | Missing/mismatched backup/confirmation/token/proposal/head/source/target revision, no-op proposal, invalid candidate or wrong target kind | Typed visible rejection before approval consumption/write/target mutation; current backup/image/approval/history remain intact except error notification. No auto-rebase, broad writes or inferred authority. | | C-06 | Head/workspace/target identity or image changes after approval but before dispatch | Rejected stale approval with zero writes; exact accepted source/proposal/before-image cannot silently follow current data. Reopen creates a new instance and invalidates old approvals/backups. | | C-07 | Same operation requested twice; competing operation; cancellation/close before or after dispatch | No second mutation or lost ownership. Exact repeated operation shares/returns its outcome; conflicts reject before mutation. Cancel-before-dispatch writes zero; once dispatch may have happened, outcome is not falsely labeled unchanged/success. | | C-08 | Negative acknowledgement, dropped acknowledgement/exception after actual commit, mismatched fresh readback or read failure | Full fresh readback determines verification. Never label Applied from acknowledgement alone. Unverified/uncertain mutation remains visible RecoveryRequired with reason/evidence; no blind resend or automatic restore. | | C-09 | Explicit reconcile after an uncertain outcome | Read-only exact current-image evidence identifies unchanged, candidate-matching or divergent/unavailable state; no write or stale cached proof. Only exact target/revision/content evidence may resolve verification; original uncertainty remains in the operation record. | | C-10 | Prepare and explicitly approve restore of original backup on same live rehearsal target | Bind fresh current revision/image and original verified backup; one guarded restore and full independent readback. Original 650/360 values and all other cells/axes restored, revision advances; no authoring/history/config rewrite. | | C-11 | Wrong-target backup, failed/unknown restore, pending dispose/late read or failed test | No restore permission, false Restored result or old-generation publication; errors persist, owner/target cleanup drains outside required gates and allows safe close. Barriers/resources released in finally. | | C-12 | Bounded audit/results, real UI/unit and coexistence paths, full verification | At most 32 dispatched mutation outcomes per target session; exact boundary rejects a new attempt before write. All 405 cases plus new coverage pass; real simulation port and native UI work; no existing diagnostics/model/archive/replay or physical side effects. | ### A. Explicit isolated simulation target, not a dynamics application Introduce one explicit simulation-only provider with a unique identifier such as `autoforge.simulation.calibration-rehearsal`. Opening requires the exact selected configuration/profile revision and its actual EngineModuleIdentifier/definition. Reject PhysicalHardware, RecordedReplay and any unregistered/mismatched provider or descriptor; metadata/capability claims alone cannot authorize a write. The target owns an independently stored immutable-at-read mutable-at-commit supported image, seeded once from the selected configuration and validated template. Thereafter Read returns actual target state, and Write/Restore use compare-and-swap against the exact instance/image revision and full validated candidate. Revision starts 0 and increments once per successful changed commit. Validation/conflict/cancellation before commit leaves the image unchanged. Preserve unique instance identity across all reads. Each read defensively copies both supported keys, axis, units and cells. The typed port supplies explicit outcomes/errors, not a success-shaped fallback. Actual Simulation implementation is exercised in normal/failure tests; test-only fault forwarding may simulate lost replies/divergent readback below Application without mocking the workflow. This target is an isolated programming rehearsal, not a live dynamics session. Do not modify `DynamicsConfiguration`, its fingerprint/catalogue, an existing DynamicsEngine, signal source, diagnostics schema or profile calibration identity. No model telemetry may claim 1250 rpm merely because the rehearsal image stores that number. Wiring to a fresh run/functional performance proof is later work. Target/backup/results are process-memory only. Close destroys that target; reopen recreates source values under a new identity. No persist-after-close, crash recovery, firmware block transfer or real-device backup claim is assigned. ### B. Independent supported-image backup and explicit approval Prepare the original target backup once before the first mutation. Read through the real target port, validate identity/applicability/revision/complete supported shape/values and compare against a second independent read of that same target. If either differs or is unavailable, do not publish VerifiedBackup/Ready status. Copy the actual observed image; never call an authoring baseline a verified read. The backup is explicitly "Verified in-memory backup of supported simulation image": one scalar plus the complete fixed-axis table, not an entire ECU. Keep its target instance/source/definition/image revision and immutable contents. It remains the original backup across multiple writes on that instance. Prepare a candidate only from an immutable saved **current Proposed head** of the accepted authoring workspace, with exact source applicability. Read current target state to form the actual before-image and precise target-to-candidate diff. Different head/workspace/source invalidates pending apply approval; do not replace it silently. An identical target/candidate is a visible no-op rejection, not a new Applied or verified mutation. Explicit user confirmation produces a single-use opaque approval bound to: action Apply or Restore, target provider/kind/instance/source/definition, verified original backup identity, expected current image revision/contents, exact candidate values and selected proposal/workspace/head (for Apply). Authoring revisions themselves remain Baseline/Proposed, not Approved/Applied. This local approval records a deliberate rehearsal action, not cryptographic authorization, manufacturer permission or a physical-write capability. Before dispatch, revalidate approval, source/head and a fresh target read against the approved before-image/revision. The target also atomically compares its actual revision at commit so a change after that read cannot slip through. Never consume approval or count a dispatched result before all preconditions pass. Once dispatch begins, prevent token reuse; if cancellation loses certainty, do not re-enable the same approval as though no write happened. ### C. Fresh verification, uncertain outcomes and separately approved restore One operation owner serializes target preparation/mutation/reconciliation/close. Guard eligibility before state changes. Commands carry an opaque operation ID; repeating the exact active/completed request shares/returns that result rather than issuing another write. Reusing an ID with different inputs rejects. Conflicting actions while work/drain is pending reject visibly; no gate is held across an await needed by cleanup. Close cancels and drains with a shared completion/generation guard; no stale success may publish into a new target. After mutation dispatch, request an independent full Read. Compare target instance/source/definition/revision, both keys, units, axes and every changed/ unchanged cell to the exact expected image. A positive acknowledgement or echoed candidate is not verification. Keep acknowledgement versus readback separate in the result; do not hide protocol/port/readback errors. If dispatch may have committed but fresh proof is missing/divergent, label RecoveryRequired/Unknown with precise actual evidence and block new Apply. Never retry a write or restore automatically. Explicit Reconcile is read-only: compare a fresh complete image to approved before/after plus expected revision. Classify unchanged/candidate-matching/divergent/unavailable honestly. Exact identity/revision/content proof may establish VerifiedAppliedAfterReconciliation; preserve the original lost-ack/verification failure in audit. Divergence never becomes success merely because one requested value matches. Restore is a separate prepared/confirmed operation using the immutable original backup of that same target instance/source/definition. Read fresh current state, bind its revision in a new restore approval and use the same atomic CAS/full readback rules. Restore must work after multiple successful writes; it restores the original supported values, not the author's latest proposal. Current revision advances; restored value equality does not pretend time/version went backwards. Wrong/closed/new instance or invalid backup rejects, and failed/uncertain restore retains RecoveryRequired rather than a healthy completion. Preserve a bounded immutable result/audit list in the owning rehearsal session: at most 32 **dispatched mutation outcomes** (Apply/Restore, successful or uncertain). At capacity reject a new dispatch before consuming approval/write; exact operation replays still return their existing result. Preparation/reconciliation update bounded current evidence, not a hidden unbounded history. Include operation/action, proposal/head and target/backup/revision identities, approval, full before/expected/actual evidence, acknowledgement, verification/ recovery status/error and explicit timestamps. This audit is volatile/local, not a new disk format or an external security attestation. ### D. Actual UI, provenance and existing source isolation Extend the existing calibration area or add a clearly named rehearsal route. Use real Application services and the Simulation target provider composed by Client. Explicitly open a target, read/prepare backup/proposal, inspect exact diff/source/preconditions, confirm Apply, inspect separate acknowledgement/ readback, reconcile if required, prepare/confirm Restore and close. Show synthetic/isolated/process-memory scope prominently, current target instance/ revision, original backup coverage, proposal/head, pending/uncertain/error status and bounded audit. Never call authoring source metadata a verified ECU identity. No implicit programming on proposal Save, page entry, unit switch or render. Unconfirmed/rejected actions leave real target and saved authoring data unchanged. Use existing display-unit conversion for human-facing torque; programming and audit compare canonical rpm/N*m with exact fixed axes. Unit changes cannot mutate candidate/approval/before-image or dispatch commands. Old live diagnostics/dynamics/recording/replay remain separate; merely opening the rehearsal cannot connect/reset/stop or rewrite them. A closed rehearsal's immutable backup/result can remain inspectable as inactive evidence, but cannot authorize a new target. Reset/new rehearsal is explicit. ## FILES OR PROJECTS EXPECTED TO CHANGE Narrow Core evidence/target-value contracts if needed, typed Diagnostics target ports, Simulation-owned rehearsal state, Application guarded workflow/readback/ audit, calibration UI/Client DI, focused tests and related README/docs. No existing dynamics model/wire/configuration/schema or authoring status rewrite. No new project/package/host or private oracle exposed to Application/UI. Under `prompts`, create only `Step-10-Report.md`. ## INTERFACE AND DEPENDENCY REQUIREMENTS Preserve nine projects/references, one Client, pure Components, accepted exact source/template/unit/head/value validation, immutable configurations/archives, recording/replay and lifetime/clock semantics. No physical provider implementation, unrestricted hardware command, emissions/safety/security bypass or inferred write permission. Simulation-only actions grant no authority to physical targets. No broad catch/silent failure/false backup/ack-only verification/automatic restore. Do not fix the separately disclosed recording navigation/refresh issue here. No dependency/system/workload install/update/repair/rollback, staging/commit/push or destructive Git. Preserve protected/history/orchestration files and dirty work. ## TESTING REQUIREMENTS Map C-01..C-12 to meaningful named tests before handoff: - Real configuration/profile/authoring plus actual Simulation target reads. Verify two independent backup reads, complete defensive images, explicit target/kind/source/definition validation and missing/corrupted reads. - Apply scalar 650->1250 with all torque cells unchanged; second proposal changes only 2000-rpm torque 360->350; full changed/unrelated readback and separately approved original Restore recover 650/360 plus every other value. - Independent invalid source/head/token/revision/shape/unit/value vectors; stale mutation between precheck and target CAS; exact prior-state/write-count assertions. Rejected calls and unit switches cannot consume an approval. - Faults injected below the real workflow: rejected before commit, commit then lost ack/cancel, fresh read failure, wrong readback/target identity, divergent unchanged cell and restore failure. Read-only reconciliation proof uses actual target values/revision, not a mocked successful service. - Controlled pending-operation/drain/close/late-publication, same operation ID, conflicting requests and bounded 32/33 result capacity. Barriers establish interleavings; no sleeps/polling/stress timing as correctness proof. Release resources in finally and await owned work. - Real rendered backup/approval/apply/reconcile/restore/invalid actions, safe error/status/unit presentation and real diagnostics/dynamics/recording/replay coexistence with zero unassigned source/model/store writes. Run focused cases, then sequentially: - `dotnet build .\src\AutoForge.slnx --configuration Release --no-restore` - `dotnet test .\src\AutoForge.slnx --configuration Release --no-build` Preserve 405 baseline cases (179 unit, 226 integration) plus meaningful new coverage. Map necessary UI-only assertion changes without deleting safety/ error/source checks. Build all nine projects/native Client. Install/restore nothing absent a genuine chosen-command missing dependency. Native smoke on rebuilt Client: explicitly create source/proposals, open isolated target, prepare verified volatile backup, approve/apply scalar then table edit, inspect full readback/audit, explicitly approve/verify original Restore, close/ reopen and confirm new-instance/no-old-approval behavior. Verify original configuration/live source remains unchanged and old routes/unit preferences work. Native normal success is not controlled port/uncertain/race fault proof. Clean only exact task-owned profiles/artifacts; never reset unrelated app data. ## DOCUMENTATION REQUIREMENTS Update README/index/architecture/build guide and rehearsal usage/contracts: isolated Simulation typed target (not a new wire/real driver), supported-image volatile backup coverage, approval/source/CAS/single-use operation semantics, fresh readback/uncertain/reconcile/approved Restore, bounded audit and actual normal/control evidence. Step-09 is accepted; Step-10 remains pending manual review. No crash durability, model application, complete ECU backup, tested performance, manufacturer compatibility or physical programming is claimed. ## ACCEPTANCE CRITERIA | ID | Criterion | Evidence | |---|---|---| | AC-01 | C-01/C-02 actual isolated simulation target, source-kind/applicability gates and independent supported-image backup | Real target/config/port tests with complete defensive reads and no fake source echo | | AC-02 | C-03..C-06 exact immutable approval/head/backup/revision and atomic validated mutation | Exact scalar/table/unrelated-value results, stale/CAS/negative vectors and zero-write rejected actions | | AC-03 | C-07..C-11 one operation owner, fresh verification/unknown/reconciliation and separately approved Restore | Controlled real-port/lifetime/idempotence/fault cases and full original-image readback | | AC-04 | C-12 bounded audit, real UI/native workflow/source isolation, 405 regressions and full Release build | Exact 32/33/no-call/state assertions, native/control distinction and worker commands/results | | AC-05 | Matrix/self-review/docs/README/history/protection/manual gate accurately describe implemented limits | Own post-green source-path review, named executed evidence and `prompts/Step-10-Report.md` | ## OUT-OF-SCOPE WORK Physical programming/driver/protocol/firmware blocks, dynamics application or performance reruns, durable calibration/backup/audit format/storage, AI/runbooks, new template/unit family, arbitrary offsets/commands, new package/project/host and unrelated cleanup. No future assignment, agents, automatic handoff, orchestration/shared-gate edit, staging, commit or push. ## REPORTING REQUIREMENTS Save `prompts/Step-10-Report.md` wholly inside exactly one plain triple-backtick fence, no language tag/outside text/nested fences. Include identifier/title/status, functions/files/interfaces, C-01..C-12/AC-01..AC-05 matrix with named executed evidence, exact backup/approval/write/readback/unknown/reconcile/restore/negative/ owner/audit outcomes and source-isolation counts, commands/exits/warnings/ per-project counts, native versus controlled coverage, docs/README, own post-green source-path self-review, limitations/deviations/blockers and preservation review. COMPLETE requires all assigned checks; otherwise report INCOMPLETE/BLOCKED with the precise unmet criterion. Save and STOP; do not begin Step-11.