Audit contribution readiness and replenish humanitarian work queue

[humanifest-bot]

Keep independent humanitarian research available while maintainer inquiries wait, and distinguish verified current behavior from accepted implementation. The portfolio contains 13 projects and 15 opportunities, with cited sources, explicit coverage limits and actionable next steps. The separate work-view feature in #3 reads these records.

Verified findings and disposition:

• CHT #10155: 431 unchanged admin tests, three controller probes and two DOM probes support the failed-user-loading report. Maintainer confirmation is still required. The separate #11342 calendar/export investigation is complete, with an unsent inquiry awaiting posting approval.
• OWA-EPANET #883: 34 native behavior cases reproduce invalid water-storage curve handling; the unchanged baseline passes 88 Boost cases and two threaded runs. Twelve additional hydraulic simulations found no meaningful effect from a neighboring nominal-diameter discrepancy. Its completed research is removed from available-work metadata; the prepared inquiry awaits human review.
• MetaData Sync #1255: WHO malaria-repository guidance supports this new research path. The global mapping list omits indicator types. After seven isolated mapper checks, 13 unchanged native metadata tests and four original-code workflow probes pass. The builder includes a source type whose fields survive under the mapped destination ID. Actual destination updates remain untested, so reuse/update semantics need confirmation before a picker-only change. Existing mapping PRs #1051/#1251 were checked for overlap. The original in-app browser probe now confirms the missing option and a working category control: selection, save, reload, restored picker selection and replacement. Original mapping use cases/repository run against synthetic metadata and localStorage. The concise inquiry is prepared and awaits human review; no maintainer contact or accepted scope is claimed.
• OpenAQ API #404: original SQL produces the reported incomplete coverage, which fails the API response models. Eight model/HTTP cases and nine SQL controls passed expected assertions. All 24 subsequent original-route scenarios also pass with synthetic database rows, original query construction and unedited result wrapping: incomplete coverage fails, valid/absent coverage succeeds, and empty-result/invalid-ID controls behave as expected. Initialization includes nullable period-column definitions in a separate metadata script, resolving the earlier source-location gap. Route evidence and fixtures document boundaries. Current ingestion, deployed prevalence, contributor eligibility and desired API/database handling remain open; no operational data was queried.
• Open Food Facts, XLSForm, CLIMADA and WFP PRISM: recorded and delivered bounded verification findings. These cover search-contract behavior, nested-repeat output, duplicate wildfire fractions and browser cache transfers. Results do not establish full applications, accepted fixes or realized impact.
• Kobo, IFRC and ODK: preserve existing contribution ownership and record the limits of formatter, transformer and original-function checks. Kobo still needs an isolated service runner; IFRC's measured rounding difference has uncertain practical value.
• Candidate screening: HOT is parked after verifying its bot-account exclusion. OpenLMIS is deferred after checking its contributor-agreement requirement and inaccessible backlog. Other already-owned, stale or unsuitable leads remain unpromoted.

The OpenAQ screen documents the new selection and exclusions. The portfolio report links the current records. The browser audit includes the reusable probe and explicit service/storage boundaries. The MetaData Sync native audit provides reproducible commands and boundaries. It also corrects an earlier setup assumption: Yarn disables dependency scripts, but still ran the workspace localization step. Its generated translation template was preserved and restored; future installs use --mode=skip-build. All 898 pinned target files were verified unchanged after the tests and browser probe.

No new production fix, agreement signing or automatic merging is included. Delivery is not maintainer acceptance. Eight actual inquiries have manual September 15 review dates; these are not automations. Unsent inquiries have no invented reply dates, and waiting reviews do not consume independent research capacity. Humanifest communication uses the verified humanifest-bot identity through a dedicated fork branch.

Validation: all 52 publication-branch tests and all 28 record/schema checks pass; generated status and whitespace checks pass. Target checks are documented separately and prove only their stated coverage. OpenAQ used 17 lockfile-hashed wheels, original model code and PostgreSQL 14.18 single-user execution under OS network denial; 43 fetched source files remain unchanged. Original sensor routing has now run with synthetic database rows; full SQL joins, application middleware and the upstream suite remain untested. The latest 13+4 MetaData Sync tests ran with operating-system network denial and synthetic inputs; the browser probe used loopback serving and CSP denying API connections. Actual DHIS2 storage/import, the full application build and deployment remain unverified.

Assisted by OpenAI Codex: source review, local probes, evidence records, validation and authorized bot coordination. No independent human code review or measured humanitarian benefit is claimed.