# Stage B2 verified publication connection — delivery record

The dashboard gained a statewide IDEA publication section inside Developmental Identification & Service Access. The implementation connected only independently checked publication records, retained all eight missing pathway labels and the existing ten domains, and left Stage B1 geography, participation, Census and CDC integrations intact. It did not establish county outcomes or referral-to-first-service completion. No new Part C figure passed the consistency checks.

## Connected records

| Record | Numerator / count | Denominator | Result | Period and native geography |
|---|---:|---:|---|---|
| Part B Indicator 11: evaluations within 45 school days after parental consent | 21,399 | 21,698 after federal exclusions | Published 98.62%; independently rounded from 98.6220…% | Oklahoma statewide, FFY 2024; narrative identifies entire reporting year and school year 2024–2025 |
| Part B Indicator 12(a): children served in Part C and referred to Part B for eligibility determination | 1,638 | Not applicable | Cohort count only, no percentage | Oklahoma statewide, same report/year; not completed transitions |

Indicator 11 excludes qualifying 34 CFR §300.301(d) cases from both counts: repeated parental failure/refusal to produce the child or qualifying transfer between agencies before determination. The report allows no additional state exceptions. The excluded count is not provided in these tables; a pre-exclusion raw rate is unavailable. The 299 outside the timely numerator are not treated as an all-referral unresolved-service cohort. Exact extraction boundaries, carryovers and deduplication remain unconfirmed and are disclosed rather than invented. These limitations permit a narrowly labeled published compliance record, not a new linked outcome definition.

## Independently retrieved official sources

Both PDFs and their official index links were retrieved anew on November 7, 2026. Physical PDF page numbers were used. SHA-256 hashes matched the prior research, but the prior research was not substituted for reading the retrieved reports. Retrieval timestamps, hashes, response codes and version details are in [connection-verification.json](connection-verification.json).

- [Official Part B report index](https://oklahoma.gov/education/services/special-education/data.html).
- [Oklahoma Part B FFY 2024 SPP/APR](https://oklahoma.gov/content/dam/ok/en/osde/documents/services/special-education/data/state-reporting/OK-01%20SPP%20PART%20B%20FFY%202024-25%20FINAL.pdf): p. 1 title/due date; pp. 56–57 Indicator 11 definition, counts, state timeline, exclusions and full-year collection; pp. 59–60 Indicator 12 definition, categories, counts and collection.
- [Official Part C report index](https://oklahoma.gov/education/services/soonerstart/soonerstart-data-and-reporting.html).
- [Oklahoma Part C FFY 2024 SPP/APR](https://oklahoma.gov/content/dam/ok/en/osde/documents/services/soonerstart/data-reporting/SPP%20APR%20FFY24.pdf): p. 1 title/due date; p. 2 July 1, 2024–June 30, 2025 collection year; pp. 29–30 Indicator 7 definition, counts, exceptions, quarter and inconsistencies.

Both reports state a February 2, 2026 due date, not a publication date. HTTP Last-Modified values were May 18, 2026 (Part B) and June 4, 2026 (Part C). First-publication dates were not established. Verification covers published aggregates, definitions and arithmetic; it is not an independent audit of administrative child records. All seven original research files were reviewed, and unsent agency drafts remained unsent.

## Excluded or unresolved figures

Part B Indicator 12 adjusted compliance remained on hold. The printed arithmetic is 969 / (1,638−159−381−50−0) = 969 / 1,048 = 92.46% rounded; residual 79. However, category e describes Part C eligibility less than 90 days before age three in the instructions and referral in the table. The eligibility/referral distinction is unresolved. Neither this rate nor a raw 969/1,638 “completion” rate was connected. Only category a’s unambiguous published referral count was admitted.

Part C Indicator 7 remained on hold, including its counts. Its April 1–June 30, 2025 cohort is a quarter, not an annual census. The published calculation adds 146 family-exception delays to 409 timely cases and retains all 557 in the denominator: (409+146)/557 = 99.64% rounded. Raw 409/557 = 73.43% rounded is distinct from adjusted compliance; neither was published as a connected measure. The remaining 2 cases implied by the table do not reconcile with the narrative’s 32 staff-delay incidences across eight districts. Quarterly counts total 2,230 while printed shares sum to 92.0%; 557/2,230 is about 24.98%, not the printed 23.0%. No unverified explanation was adopted. Agency reconciliation or a corrected official release is still needed.

Existing Part C first-service and exit-data holds remained active. Indicator 1 stayed outside this integration. County referral cohorts, all-referral evaluation completion, complete eligibility outcomes, first delivered services, referral-to-first-service completion, waiting periods and rural/urban completion remained DATA NOT YET CONNECTED. No statewide/district-to-county allocation or suppressed-cell reconstruction was performed. The connected statewide cells contain no suppression symbols; no program-wide or local suppression threshold was inferred.

## Implementation and files

Persistent structured records were placed in Netlify Database via a new Drizzle table and deployment migrations. A read-only endpoint admitted only the reviewed statewide release/counts, with explicit checks for holds, suppression, geography, source version and arithmetic. Unavailable database access returned no records. County query parameters were rejected. No in-memory or local JSON data store was added. The JSON manifest is research evidence, not a runtime data source.

The dashboard included a named keyboard-scrollable table region, caption, row/column headers, source-page links, verification timestamp, definitions, exceptions, reporting-period limitations, suppression notes, Part B/Part C separation and explicit holds. Reload, unavailable and partially available states were provided. Map and county filters did not change these historical statewide records.

Files changed or added:

- `real-time-dashboard.html`
- `css/verified-dashboard.css`
- `js/idea-indicators.js`
- `js/warning-center.js` (developmental evidence text only)
- `data-sources-methods.html`
- `db/schema.ts`
- `netlify/lib/idea-indicators.mts`
- `netlify/functions/idea-indicators.mts`
- `netlify/database/migrations/20261107155424_add_verified_idea_indicators/migration.sql`
- `netlify/database/migrations/20261107155424_add_verified_idea_indicators/snapshot.json`
- `netlify/database/migrations/20261107155500_seed_verified_idea_publications.sql`
- `tests/idea-indicators.test.mts`
- `docs/phase6-b2-research/connection-verification.json`
- `docs/phase6-b2-research/connection-report.md`
- `.netlify/results.md`

## Validation

`node --experimental-strip-types --test tests/public-data.test.mts tests/disability-context.test.mts tests/idea-indicators.test.mts` passed 34 tests. These covered existing Census/CDC parsing, holds, stale/unavailable behavior and restoration; Stage B1 geography, suppression and cache checks; actual deployment seed arithmetic; rejection of held, suppressed, changed, unsupported-geography and count-as-rate records; ten warning-sign domains and eight unchanged missing pathway labels. JavaScript syntax and `git diff --check` passed.

Chromium ran against `netlify dev --port 8889` at 1440, 390 and 320 CSS-pixel widths. The browser used the exact migration seed records as intercepted API responses, with the unrelated feeds placed in unavailable states. Checks passed for section bounds, internal mobile table scrolling, keyboard Tab order and Enter activation of disclosure controls, category/indicator filters, developmental pathway selection, all 77 county map paths, Tulsa selection and statewide filter independence. Unavailable, partial and reload recovery states passed. Axe checks of the new section, including expanded disclosures, reported zero WCAG A/AA violations at the tested widths. Screenshots were visually inspected. Screen-reader semantics were checked through table/region/status markup and automated accessibility checks; no manual assistive-technology user session was performed. This is not a full-site WCAG certification.

The actual local endpoint returned HTTP 503 with an empty records array. Persistence after migration application therefore remained unvalidated. Migrations were generated, not applied manually. No build command was run; platform deployment validation still needs to apply the schema and seed and confirm the database-backed response. Existing live Stage B1 feeds were not refreshed as part of browser testing.

## Git and preview delivery status

The workspace reported branch `main`, with baseline commit `c0efdd09a98c913e0a29217cf3b27fe399141319`. No new commit or pull request was created by this run; the platform manages those artifacts. Their new identifiers and URLs were not yet available.

No new deployment or preview provisioning attempt was initiated by this run. A successful preview for these changes was not established. The historical failures documented in [preview-provisioning.md](preview-provisioning.md), deploys `6aef477165f1288df2b37ad5` and `6aef4795d54b72c0de3ad50c`, failed on `createSiteDatabaseBranch` HTTP 500 for branch `agent-6aef469eab373`. That is a recorded platform provisioning failure, not proof of an application defect. It was not repeatedly retried, and database configuration was not bypassed or reset.

The code and migration records are ready for review. Production publication and successful database-backed preview validation were not claimed. Rollback can remove this section/script and endpoint without changing Stage B1 data; the additive public-record table can remain unused. Held figures require a separately reviewed correction before release.


## Subsequent platform validation result

The platform subsequently attempted deployment `6aef4a849a6a51e4c002a1a8` for preview alias `agent-6aef49728ab89`. It failed while creating the database branch: `createSiteDatabaseBranch` returned HTTP 500 (`internal error`). The log listed six migrations but showed no SQL execution or compilation failure. A successful preview and database-backed validation therefore remained blocked. No retry or application workaround was attempted. The detailed investigation is recorded in [preview-provisioning.md](preview-provisioning.md).

At the follow-up investigation, the workspace reported `main` at commit `3c0811f8868e5a0f528fb7f5d63c618c9f28c19f`; a pull-request URL was not established. The investigation changed only this delivery record, the provisioning investigation and the run summary. The original implementation and its verification/quality holds were preserved.


A second connection-preview attempt, deployment `6aef4aad44e1999e41b0a5e2`, failed with the same `createSiteDatabaseBranch` HTTP 500 for `agent-6aef49728ab89`. It provided no new SQL or compilation diagnostic. The workspace at this follow-up reported `main` at `1ffcc85d663bc8785e41884275f2f2a39376b836`. Preview/persistence validation remained blocked, with no established pull-request URL. The repeated provisioning failure was documented without initiating another attempt or changing the implementation; see [the provisioning investigation](preview-provisioning.md).

## Full development extension

The subsequent full development implementation added the nine-measure accountability framework, geographic evidence availability, rural/urban readiness and a separate future aggregate review schema. It preserved the two existing verified publication records and all source discrepancies/quality holds. See [full-development-report.md](full-development-report.md) for the exact file list, fresh verification, 40 passing data tests, browser/accessibility results, branch-creation limitation and continuing preview blocker. No deployment was attempted.
