Training completion records, dated hiring steps and scheduled check-ins are available in code, not proof of effectiveness.
Hire, train, retain · sources before conclusions
Begin with departments and sourcing/channel results: staffing gaps, hiring time, channel spend and retained hires. Carry a sourced question into the position record, training completion and retention follow-up. These are available records and views, not demonstrated improvements.
Explore the available workflows →Published process example; no position outcome supplied
The Austin Police Department Recruiting and Hiring Audit is a narrow reference for manual and inconsistent applicant tracking. Publication: April 2025 per research synthesis; document metadata date not independently verified. Reference reviewed . Full audit verification is required before quoting figures or recommending an intervention.
No vacancy, candidate, named official, local delay or completed outcome has been supplied. Austin is not presented as a Staff Sustain customer.
Read the audit's scope and distinguish its tracking observation from any claim about a particular position. The input here is a public document, not private applicant information.
A proposed local review question is: “Can the authorized reviewer reconstruct the next step from dated records?” The audit does not answer that question for your agency. A local position identifier, applicable rule and actual dated evidence are still needed.
Ask the agency to identify the HR process owner and department requester who can reconcile the local records. No responsible person is named by this example. A role label alone does not grant decision authority.
The implemented Commit flow reviews the proposed result, confidence and what could change it before typed-name confirmation, checking the current version and an active delegation for confirmation. Ownership alone does not grant authority. It is not independently accepted. In this public example, the honest result remains Needs your decision: no local reviewer or authority evidence has been supplied.
The proposed owner would record what happened, its source and observation date, plus the next review date. For this public example the outcome is unknown. Do not convert a reviewed document into a claimed hiring improvement.
Implementation boundaries
Source snapshot reviewed : current position-case, document, coach and memory routes, plus data and market routes. Implemented does not mean independently accepted or agency browser-accepted.
Start / Watch / Commit, append-only records, version checks, typed-name confirmations, reminders and outcomes; case-level state JOLTS evidence; authenticated verification and audience-projected HTML/CSV briefs.
Versioned attachments, expiring revocable sharing, consent-controlled coach memory, dated citations, history and change orientation are implemented. Provider success remains a separate check.
No automatic ATS or finance connector, email-delivery service or live occupational pay feed is connected. BLS OEWS wage import and occupation/area lookup are implemented; they require actual imported source records and do not provide an automatic live pay feed.
The accepted end-to-end position experience has not been verified. Do not infer browser completion from an API route.
Agency acceptance, independent verification of delegation sources, automatic ATS/finance/HRIS connectors, consent-checked messaging, independently accepted causal outcome evaluation, procurement acceptance and measured U5 usability remain unestablished. Pilot trial and effect recording is implemented, not proof of effectiveness. MFA enforcement and OCR are not implemented.
Supporting workflows · available in code today
Implementation reviewed against src/routes/provision.js, src/routes/workforce.js, src/routes/reporting.js, src/routes/market.js and docs/reporting-powerbi.md. Independent acceptance and effectiveness are not established.
A trusted platform owner uses the provisioning console to create an agency with its initial admin, add accounts and record initial delegations with source references and effective dates. Platform ownership does not grant tenant-content access. This is controlled provisioning, not self-selected agency membership.
Enter authentic sourcing channel spend and counts, referral counts, positions and candidates with source labels and observation dates. Candidate changes retain dated events and calculate elapsed time since the preceding recorded event. A channel's cost per retained hire divides spend by retained hires; a missing denominator stays unknown. Referral spend is not collected, so referral cost per retained hire is not supplied.
Timing is only as accurate as the supplied events. Current candidate transitions can populate the start field at acceptance; reconcile actual starts before treating that field as an observed first day of work.
Record the supported check-in intervals—30, 90, 180 and 365 days—and review the next uncompleted check-in with its due date. The due list uses the stored start date and excludes separated candidates. It does not send candidate messages or establish that support improved retention.
Record assignment method, comparison group, outcome measure, trial arms and a pre-registered decision rule. Record effect estimates and supplied uncertainty with sources and dates; missing confidence intervals are explicitly unknown. Recording a trial is not proof that assignment was followed, confounding was controlled or an intervention caused an outcome.
An authorized admin can import BLS OEWS wage rows with source-file details and observation dates. Lookup uses occupation and area codes and returns the latest matching imported year with annual wage evidence labels. Missing or suppressed values are not invented pay figures. Check occupation, geography and period fit before local compensation review.
BLS Occupational Employment and Wage Statistics is the external wage-reference program. Publication date varies by release; reference reviewed . Import support does not establish the correctness or freshness of an agency-supplied file.
An agency admin creates a reporting token and stores its one-time raw value securely. Power BI can request the HTTP JSON API through its Web connector using that token in the Authorization header. Expand each measure's value and evidence together, retaining source, observation date, state and limits. Preserve unknown values rather than converting them to zero.
The connection guide is maintained at docs/reporting-powerbi.md. Reporting reads cover departments, channels, referrals, aggregate check-ins, pilots and position cases. Root extraction time is not each measure's observation date. Empty agencies return empty lists, not invented records. Verify Desktop connection and Service refresh in your own environment; no tested Power BI Service integration is claimed.
Rotate or revoke credentials through the agency admin controls. Revocation blocks future reads, not files already downloaded. Aggregate check-ins have no minimum-group suppression; small groups, referral identifiers and pilot unit keys may remain sensitive. Restrict report and workbook access.
The implemented flow · not independently accepted
Enter the actual position and jurisdiction within your assigned agency. Save a draft, then assign the owner, backup and review date. State transitions are checked server-side; version checks reject stale edits instead of silently replacing newer work.
Record known and missing information, references and a proposed action. State JOLTS is optional live context with its own missing or failed-source state—not a local staffing measure. Attach PDF, DOCX, XLSX, CSV, TXT or MD files with source-owner and effective-date details. Replacement versions retain the earlier attachment. Only TXT, MD and CSV text is extracted; PDF, DOCX and XLSX text is unavailable, and there is no OCR.
Open the coach beside the case and review its case/version orientation and previous answers. Perplexity or Anthropic proposals are unverified AI; rule-based guidance is explicitly not AI. Perplexity source metadata is not independent citation verification; Anthropic does not add web-search verification. Missing provider credentials return an unavailable response. With memory enabled, orientation compares the current case with the last coach-viewed version; this is not tracking every browser visit. View, correct, export or delete your own agency-scoped memories. Proposed memories require your confirmation before storage.
Memory starts off. Turning it off stops memory use and new saves but does not delete existing memories or coach history. Memories persist until you delete them; coach history is kept for traceability. Document excerpts are not sent to a provider unless the agency's separate excerpt permission is enabled, independently of memory consent; responses disclose this boundary. Review provider terms before submission.
Compare immediate, intermediate and enduring proposed actions while the case is in draft, then explicitly submit for review. A reviewer can return a proposed action for revision with a reason; it remains in the record. Only actions allowed for the current person and state are offered.
Review result, confidence, what could change it and typed name, then explicitly confirm. The server requires the current version, write permission and an active delegation for confirmation, including for admins. Owner and backup assignments route the work; they do not grant approval authority. The authenticated verify endpoint reports the latest confirmation and case version, not independent proof of legal delegation.
Download HTML or CSV for position, HR/finance or council. Council exports omit private free text records and person-level fields; the CSV is deliberately limited. Spreadsheet-formula prefixes are neutralized. Public sharing is limited to a council one-pager: review the current version and explicitly acknowledge its projection before creating an immutable snapshot link. Anyone holding that public link can open the snapshot until expiry or revocation; later edits do not change it. Position and HR/finance links require sign-in and agency access. Documents are not exposed through the public snapshot.
Add an in-app reminder or acknowledgment, then a sourced, dated outcome with a typed name. An unresolved outcome can return a monitored case to reassessment. Corrections append a superseding record; earlier records remain. Revoke an obsolete share link separately. A download, link access or acknowledgment is not proof that a message was delivered or an intervention worked.
Prepare actual authorized records
Send comma-separated text with a header row, using an authenticated analyst or admin account assigned to the agency. These are limited flat imports—not a full history transfer. Do not upload sensitive background content.
Required: name, work_type, authorized, filled, turnover_pct, retained_12m_pct, days_stage1, days_stage2, days_stage3, days_stage4, days_stage5, qualified, withdrew, cost_per_hire.
Optional: source.
Work-type codes: PS, COR, TECH, PRO, IT, CRT, CMT, HHS. Filled cannot exceed authorized; withdrew cannot exceed qualified. Counts must be nonnegative integers; percentages must stay within percentage bounds.
No invented values: a row missing any required measure is rejected with the field named; nothing is defaulted to zero or full retention. Import only reconciled measures you would defend as agency evidence.
Repeating a department name within the same agency and source updates its aggregate record. The original measurement period and source owner are not separate import columns; keep the original evidence for review.
Beyond vacancy and retention: the platform reports the stage-to-stage selection funnel per department, per-requisition cycle time and offer acceptance, withdrawal reasons by channel, early-stage pool coverage against openings, outreach attribution per effort (audience, ambassador, message, spend, cost per hire), and candidate-experience ratings by stage. Rates with a zero denominator and figures without a source date are reported as unknown, never as zero. Candidate comments are record-level content and are never aggregated into analytics.
Required: external_id, source_channel; source channel cells cannot be blank.
Optional: department_id, stage, applied_at, consent_comm, source, source_tag. A valid campaign tag can associate an application with a recorded outreach effort.
Accepted stage codes: sourced, applied, screened, interviewed, offered, accepted, onboarded, active, separated, withdrew.
Department references must belong to the assigned agency. Supply a real ISO calendar date or a timestamp with an explicit zone. Consent accepts true, false, 1, 0; a supplied field is not independent consent evidence.
Important default warning: omitted stage becomes applied; omitted application date becomes the import time; omitted consent becomes false. Supply the actual event date. Reconcile updates against the original file rather than treating an import as independent evidence.
Before and after import: review required headers, duplicate columns, blank numeric cells and row errors. Validation errors reject the batch before writes; database errors can produce partial writes. Reconcile accepted and rejected rows against the original file before retrying.
The source field defaults to the route's real-data label; that label is a caller assertion, not independent verification. Use only authentic authorized records. A CSV import does not establish consent, the origin of a document or a completed appointment.
Questions worth asking
No. Current imports accept limited flat CSV records, not an automatic ATS connection or a full dated event history.
A proposed action can be returned for revision with a reason; corrections append superseding records; agency acceptance is not established. Edit a draft or reassessed position with a current-version check, compare proposed actions, then explicitly submit for review. Export an audience-projected HTML or CSV brief for position, HR/finance or council.
No. Generated proposals require human review. The coach does not select candidates, approve funds or send communications.
No. It is a published process example; no position outcome supplied. No Austin agency relationship or Staff Sustain intervention result is claimed.
Choose a real position, identify the governing authority and agree what evidence would change the decision before evaluating the workflow.
Evaluation requirements →