[anth-volk]

Fixes #1428

What changes

The Public API is adding missing_parameters to GET /{country}/policy/{id} (PolicyEngine/policyengine-api#4039): the parameter paths a policy sets that the current tax-benefit model no longer has, or null when it could not check. v3 now reads it wherever a user chooses or runs a policy. When the list is not empty, v3 names the parameters in a notice and runs nothing with that policy. The notice's wording is plain: it says the reform or report "cannot be run, because it changes parameters that PolicyEngine’s model no longer has:" and lists the paths (they have no labels in the model any more). For the user's own saved reforms and saved reports, a dialog also offers to transfer them to a new policy without those parameters (see "Transfer a saved reform or report to a new policy" below).

This has an effect only once that Public API change is deployed. While the field is absent, and whenever it is null or empty, v3 behaves as before and blocks nothing.

Where it applies

• Saved reforms ("Your reforms"). Both reform stores read the field with each reform's policy and keep it as Reform.missingParameters. The card reads "Uses parameters the model no longer has"; the reform's page names them and disables "View full impact report". Opening such a reform also opens its transfer dialog.
• Re-run of a saved report. Re-run is decided when clicked, from the report's own reform and baseline policies (reportMissingParameters), so it covers a report compared with a deleted or edited saved reform. A policy that cannot be read does not hide the others' result; when the report itself cannot be read, the saved comparison reform a re-run would fall back to decides. Re-run is disabled while it checks. When it finds missing parameters, the notice names them and the saved report's transfer dialog opens.
• Build. The "Compared with" list shows a reform with missing parameters, names them, and cannot select it; a draft already compared with such a reform names the parameters and disables "Save and run". The draft's own changes come from the current model's parameter list and are not checked. Build never opens the transfer dialog.
• The report page. useFlagshipReport reads each report policy as soon as the simulations name it (not after the parameter metadata) and returns policiesChecked and missingParameters. The page shows "Loading report" until the policies are checked and starts nothing before then. With missing parameters:
  • without stored results, it shows the title and the notice, and calculates nothing;
  • with stored results, it shows them under the notice, for the report's own year only. No other budget-window year, budget window or UK local-area breakdown runs, including the local-area breakdown that main now starts in the background; the Household tab (where all household calculations and their variation views live) says households cannot be calculated; Recompute and Edit in Build are disabled.
  • when the report is the user's own saved report, its transfer dialog opens as the page opens, whichever section the link names. A report the user has not saved, such as another user's shared link, shows the notice only.

Transfer a saved reform or report to a new policy

Like the previous app's DeprecationModal, a dialog names the parameters the model no longer has and offers to move the user's saved item onto a new policy without them. It has two actions: the transfer, and "Not now", which closes it and leaves the item blocked as above. There is no delete action. When nothing would be left without those parameters, the transfer is disabled and the dialog says why. A failed transfer shows its message in the dialog and changes nothing; nothing retries by itself.

• Saved reform: "Transfer to a new policy". It calls reformStore.update with the reform's other changes only. The store saves them as a v1 policy (POST /{country}/policy, which returns an identical one if it exists) and repoints the reform: PATCH /v2/user-policies/{id} with only legacy_policy_id, or the same change in the browser store. The reform keeps its name, baseline and provenance, and its reports keep their results. The reforms are read again, so the notice and the block lift.
• Saved report: "Transfer and run again".
  • It runs the report again as Re-run does: the same year, population, budget window and comparison. If the report's record or its baseline simulation cannot be read, it stops rather than guess them.
  • It reads every policy the report ran again and leaves their missing parameters out of the provisions and out of the comparison policy. A policy that cannot be read stops the transfer, and a comparison that would be left with no changes is refused with a message.
  • It repoints the same saved report at the new run instead of saving a new one: PATCH /v2/user-reports/{id} with legacy_report_id, provisions, validation: null and last_run_at, or the same change in the browser store. Its links to saved reforms stay as they are. FlagshipReportStore gains update for this.
  • An account report keeps its link (the row's ID), which then shows the new results; transferred from its own page, the page stays and reads the new run. A report kept in this browser (sign-in off, or outside the US and UK) is linked by its v1 report, so its saved report opens the new run's link instead, and the dialog says so.
  • The repointed report is dated by its new run (an account record's date is now the row's last_run_at when it has one), and its old validation pin is cleared so the new run is pinned.
• Where it opens. On a saved reform's page, opened from its card in Your library; from Re-run on a reform's reports; and on the user's own saved report's page. Lists open nothing until clicked.

The Public API is unchanged: it already accepts both PATCH bodies.

Behaviour that changes even while the field is absent

• The report page reads each of its policies as soon as the simulations name them, and waits for those reads before it shows stored results or starts the run. The policy details, which used to read each policy after the parameter metadata, reuse that read, so each policy is still requested once. A read that fails is not retried for this check and counts as nothing missing, so a failure does not hold the page.
• The browser reform store (sign-in off, or a country without accounts) reads each saved reform's policy when it lists them, through the session's query cache. A policy that cannot be read lists as saved.
• An account saved report's date is its row's last_run_at when set. A new row sets it when it is created, so only a repointed report reads differently.

Overlap with other open PRs

• v3 household section: UK suggestions, US SPM geography and calculation provenance #1386 and v3 standalone household report and saved households, unreachable until v3 moves to the v2 routes #1392 change the household report components. This PR touches none of the files they change: the Household tab is replaced in ReportView, not inside HouseholdImpactSection.

Environment variables

None. The change reads no environment variables and touches no workflow, build or deployment configuration.

What this PR waits for

PolicyEngine/policyengine-api#4039 (stacked on PolicyEngine/policyengine-api#4036) must be deployed before this has any effect. This PR can merge before or after it.

Testing

Each commit was written test-first; the failing run came before the code change (vitest, Node 26).

• ef2debaa (stores read the field): 6 store tests failed (the stores did not read missing_parameters); then 49 passed with the store hook tests.
• 07eeeb3b (saved reforms): 3 page tests failed (no "Uses parameters the model no longer has", no notice); then passed.
• ab8023a8 (Build): 2 tests failed (Unable to find role="alert"); then 43 passed.
• eb5c9aab (report page): 6 tests failed (no policiesChecked/missingParameters, the run started, no notice); then passed.
• Review round 1: 1772c24e (hold every run until checked; no retry delay; one read per policy; browser findById), 244f1754 (Re-run checks the report's policies), 2a1376d5 (UK local-areas reason; helper and notice tests). 5 tests failed first, among them a complete report starting another year and a window before the check, and two reads per policy; then passed.
• Review round 2: 54021cdd (a settled check stays settled on refetch; renaming reads nothing) and 1ae167b2 (Re-run decided by the report's own policies, disabled while checking; the picker names parameters). 5 tests failed first; then passed.
• Review round 3: ed1e2b17 (Re-run with a partly failed check or an unreadable report; UK overview card). 3 tests failed first; then passed.
• On every commit up to ed1e2b17: bun run typecheck, bun run lint (ESLint and Stylelint), bun run prettier, and the flagship, hooks, api and libs tests. After an earlier rebase onto main, the whole app/ suite: 459 files, 5,926 passed, 10 skipped.
• Rebased onto main again on 2026-10-29 (9f542782). main's new background start of UK local-area breakdowns is held while a report is blocked or its policies are still being checked, with one new ReportRunLifecycle test that fails without it. The type check, lint, Prettier, the changed test files and the local-area tests passed. A whole-suite run had 3 failures, none in a file this PR touches: two header tests also fail on main, and a chart test timed out under load and passes alone.
• Transfer dialog (dec8c6a4, 3edd38c2, dcfb1975): 28 tests failed first (no store update, no repointing run, no dialog); then passed.
• Its review round 1: 5b659735 (refuse a transfer that cannot read the run it replaces), 6e567ec4 (dialog on any section), 2c549f2e (link wording for each store, one dialog for both pages), 97a606ba and 4e3e9768 (date and re-pin a repointed report; a refused browser write fails). Review round 2: 5b9c83bc (read every policy of the report before transferring), 1f21b640 (drop the transferred report's cached page data), 391116c8 (re-read the pin only when a run is replaced), 11c86633 (keep fields passed as undefined). Review round 3: f6d346e0 (stay on the page when the link is kept). Each fix had a failing test first.
• For the transfer dialog and its fixes, only the tests these commits add or change were run (by name for the later fixes), with bun run typecheck, bun run lint and bun run prettier; the whole suite was not run.
• Earlier GitHub Actions runs on this branch were cancelled at the user's request, until 2026-10-27.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
policyengine-calculator-next	Ready	Preview	Oct 30, 2026 3:52pm UTC
policyengine-website	Ready	Preview	Oct 30, 2026 3:52pm UTC