Add a pupil not found on my DfE pupil roll (simple path) AB#297310

[ajs-uk-dev]

Adds the Add a pupil not found on my DfE pupil roll journey (simple path) for KS4 June, KS4 Autumn and KS2. AB#297310.

Simple path means no dataset matching against the roll — the soft-match story (AB#297780) sits in front of this one and its interception point is named and commented, but no matching logic exists here.

docs/add-pupil-journey.md is the full write-up; this is the short version.

What it does

A fourth radio, Add a pupil to data, appears on What to change for the three window types that have a flow. Post16 has no Add_Post16.json and never offers it. The journey is three pages — learner details, admission details, evidence — then the shared summary and confirmation, both reused unchanged.

The one genuinely new mechanism is the synthetic pupil. Every other amendment journey opens with a pupil search that resolves a real dataset pupil into RequestState.SelectedPupil. This one can't: the pupil isn't on the roll yet, which is the whole point of the ticket. So learner-details carries a pupilFromAnswers flag, and AddPupilJourney.BuildPupil mints a PupilDto from its own typed answers on a successful POST. Every downstream consumer — summary templating, drafts, the amendment grid, withdraw — reads SelectedPupil exactly as it already does.

No rules-engine outcomes (ticket B2). Submission writes the same ChangeRequests row every amendment writes (Amendment + AmendmentType.Add, SubmittedUnCommitted) plus the journey blob, but skips the enqueue, and the window-close Zendesk replay skips Add rows. Both sites carry a PARKED [AB#297310](https://dfe-gov-uk.visualstudio.com/b947688b-53a4-4c90-9c43-5a5db31e313d/_workitems/edit/297310) comment, mirroring the AB#296648 ResultsEnquiry precedent. QuestionFlowOutcomeKeyAlignmentTests lists Add in FlowPrefixesThatDoNotRouteToTheRulesEngine and asserts it has no outcome key, so it can't be quietly bound to routing later. Downstream is the LDS egress, a separate story.

Two things a reviewer should look at first

WhatToChangeController.Confirm is shared by all five journeys. It now clears the per-request identity — reference, selected pupil, matched pupil, selected result, answers, history — for any flow with no PageType.PupilSearch page. Every other journey has always started clean by accident of its shape, because PupilSearchPost regenerates all of that on every pupil selection; a flow without one inherits whatever the last journey left behind. Two live consequences: a submitted request's reference was reused and its row overwritten by the upsert, and an abandoned Merge journey's matched pupil surfaced on the Add summary as "Second record to merge" — the merge rows are gated on MatchedPupil != null && SelectedPupil != null with no WhatToChange check — and that unrelated pupil's full PupilDto was persisted into the Add request's blob.

It's keyed on the flow's shape rather than on WhatToChange.Add, so the next pupil-search-less journey inherits the guarantee instead of the bug. Every existing flow has at least one pupil-search page, and Confirm_ForAFlowWithAPupilSearchPage_LeavesTheSessionIdentityAlone pins that, but the blast radius is the thing to test.

The merge with main is non-trivial in one file. ad80cbe7 resolves WhatToChangeController, where main's #318 checking-exercise gate and this branch's reset landed on the same two methods. The gate runs first — a closed exercise is turned away before anything touches session state. Index also needed a fix git did not flag: the automatic merge kept both sides' GetCheckingWindowAsync calls and left a duplicate local.

Verified

Against the merged tree, with the dev stack rebuilt so main's six new migrations applied:

Suite	Result
Unit	4391 passed, 0 failed
Integration	724 passed, 0 failed
E2E (Category!=VisualRegression)	118 passed, 0 failed, 41 skipped

A full Add walk in a browser submits as CYPMD_KS4June_* with zero queue messages, and a Remove journey started afterwards still routes to select-pupil.

AddPupilJourneyTests covers the happy path, required-field errors, the future-date block and Post16 not offering the option. It does not guard the two session-carryover defects above: PageTest gives each test a fresh browser context, so a test that runs one journey in an empty session has nothing to carry over. Those are pinned at the unit level instead. Closing that gap needs a seeded pupil for the Merge search, which the dev org currently lacks — /pupils/suggestions returns [] for every query.

Open, not blocking

• Copy is placeholder. The radio label, the sex option U labelled "Not known", the SEN order E→K→N, and every page/question/error string were implemented verbatim and pinned by tests — Figma Epic B1 wasn't reachable from this environment. Needs content-design sign-off.
• Two flagged assumptions the ticket's validation column doesn't specify: future-date rules on DOB and admission date, and the 150/13 character caps. Cheap to keep, one deletion each if the BA strikes them.
• KS2 and KS4 Autumn have no seeded dev window, so those two flows ship pinned by AddFlowTests but unwalked in a browser.

[github-actions]

Deployments

Review app is available at these URLs:
https://check-performance-data-pr-345.test.teacherservices.cloud

[ajs-uk-dev]

Thanks for the review — all ten points are addressed across five commits.

Blocking — null flow config wipes journey state — f89174c. An unknown flow no longer takes the reset branch, so a radio with no flow file, or a forged enum value, leaves an in-progress journey alone as it did before this PR. Two regression tests cover it, including your repro.

Fix before merge

• Reset missed the origin-country fields — f89174c. Both are now nulled, and SeedStaleIdentity seeds them so the test can fail.
• Post16 exclusion enforced only in Razor — f89174c. AddPupilJourney.SupportedWindowTypes is now the single source of truth for the radio and for a guard on the post, checked before the flow service or the session. The new test stubs a non-null Add config, so it pins the window-type rule rather than the missing blob.
• Worker AddMemoryCache — c0458f9. Nothing in the worker's graph consumes IMemoryCache since Stop registering IStudentResultsClient in the worker's dependency bundle #328, so the call and its comment are gone. The worker validates on build; it comes up clean with 0 restarts.
• Admission date could precede date of birth — 5524828. The two dates must now be in order, equal allowed. They are on different pages, so the rule reads the journey's answers with the posted page's overlaid and anchors the error to whichever question the user is looking at, modelled on PageDateRules.

Minor

• FreeText character limits were server-only — 69ae835. The GOV.UK character count component is textarea-only, so the single-line input carries maxlength; the server rule stays the enforcement.
• Actionless "Pupil name" row — 3e1893c. Rendered only when the journey has a pupil-search page for its Change link.
• Merge rows keyed on session state — 3e1893c. Now keyed on the flow declaring a match pupil search, covering the submitted view too.
• EditAdviceService Add arm — 3e1893c. Adds the Add wording; the summary target was already correct.
• Constants asserted against themselves — 3e1893c. AddPupilJourney and AddJourneyDateRules ids are pinned to string literals, and the supported window types to the flow files that exist.

c90eeb3 moves one E2E assertion onto the first/last name rows, since the combined row it read is the one removed above.

Verification after each commit: solution builds clean, unit (4415) and integration (726) green. Mutation-checked the state-reset guard and the window-type rule — reverting either fails the new tests and nothing else. Rebuilt worker and web containers; full non-visual E2E is 118 passed / 0 failed / 41 skipped, matching the pre-change baseline.

The four IncorrectGradeEnquiryTests failures you saw are environment-local — they fail on main too and pass in CI.