Measured on a 2.7 MB production plan (plan-cxpaket.sqlplan, kept out of git):
- rewrite_sql becomes option_type + sql_to_run. The old name told the model 'query rewrite only', so an index option carried no CREATE INDEX anywhere in the pipeline and the report had zero SqlBlocks. It now requires paste-ready T-SQL per option, and the report prompt makes a SqlBlock per non-null sql_to_run a hard requirement.
- verify_semantics keys on option_type === 'rewrite' rather than 'has sql', so index and statistics options are not flagged just for carrying DDL.
- dropPlaceholderSql: 'SELECT ... FROM ... OPTION (MAXDOP 1)' is not runnable and was rendered as if it were. Ellipsis SQL is now nulled, and digestForModel reports sql_truncated so the model is told not to offer a rewrite for a statement it never received.
- max_tokens 2048 -> 4096 for both calls: the larger JSON outgrew the old ceiling and the user waited 38s for 'malformed JSON'. A truncated response now says how many characters it got, and extra params still override the ceiling.
- Real plan, live: 52s, 17 cards, 3 SqlBlocks with real table names and no placeholders, 3 ScoreBars, headline 'Query Plan Timeout on Parallel Join'. 43 unit, 23 e2e.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Measured against a 2.7 MB production plan that failed to import:
- maxNestedTags: real spaghettified plans nest 2729 tags deep and fast-xml-parser caps at 128, which surfaced as 'Maximum nested tags exceeded' and blamed the export. Raised to 20000; the 12 MB input cap bounds the worst case.
- table names live on <Object>, not on <IndexScan>/<TableScan>. The digest reported 0 tables for a plan with 51 objects; the analyst model never saw a table name. Now 37 distinct tables.
- SSMS writes StatementText with numeric character references (
 on every line, 90 in this plan). Attribute/tag processors decode them, so the model gets SQL instead of entity noise.
- 3 new digest tests, each from one of the three measurements above. 39 unit tests pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- lib/plan-file.ts: BOM-aware decode for UTF-8, UTF-16LE/BE (with and without BOM) and UTF-32, with a NUL-density fallback for BOM-less UTF-16. The browser had a two-byte FF FE sniff only, so a UTF-16BE or BOM-less plan arrived as NUL-filled junk and the digest blamed the export.
- digestPlan now quotes fast-xml-parser's first line instead of a fixed sentence, so a truncated file, a stray entity and a wrong file stay distinguishable.
- plan-file.ts holds the decode so the XML parser stays out of the client bundle.
- tests: 6 encodings round-trip through decodePlan into a digest; a truncated plan's message is asserted to carry a reason; e2e drops a real UTF-16LE file. 36 unit, 23 e2e.
- history.spec picked the rail's first row, which is now one of the new failing-input rows; it selects the newest finished run instead.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- the session cookie was Secure whenever NODE_ENV=production, so a production build on http://localhost got no cookie back in Safari: every page after signup looked signed-out. It is now Secure only when X-Forwarded-Proto says https, or when COOKIE_SECURE=1.
- /app and /app/[id] now require an analyst and a Jev key, the same rule /api/analyze enforces with a 412. Closing the guide mid-way used to leave a console that could not run anything.
- Playwright runs the production server (pree2e builds first) and gains a webkit project for session survival; chromium-only QA could never see a Secure-cookie rejection.
- tests-e2e/session.spec.ts pins both; .env.example gains COOKIE_SECURE and stops shipping a blank APP_SECRET.
- e2e: 21 passed (setup, chromium, webkit, mobile).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
- changeEmail validates and refuses an address another account owns (users.email is UNIQUE; a collision was a 500 inside the settings form).
- /app/[id] no longer redirects a 'running' row to itself; it renders what exists and says the run was interrupted.
- ScoreBar / OptionRow / VerdictBlock survive null numbers from the model instead of throwing and losing the block (seen live: '[openui] runtime/render-error: Component ScoreBar render failed').
- drop the ad-hoc render-debug scripts now that Playwright covers the browser.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Dockerfile/compose written but never built or run on this machine (no Docker). smoke.sh now works in both modes and fails loudly.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds @openuidev/lang-core (one-line reason: the prompt-side library must build in a non-React Node runtime; react-lang's createContext crashes the route's server bundle).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
No fixtures/real.sqlplan available (no SQL Server reachable from this machine); the real-plan test skips and says so.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
