Opportunity Agent

Free-safe autonomy

AI off, sending off, zero-source startup, zero-source route-map visibility with pending-deploy fallback, bounded source checks, scheduled source expansion, source-finding fallback playbook, public link graph discovery, sitemap scanning, query-hint generation, query-hint recovery brief guidance, query-result resolving, candidate-source testing, origin metrics, adaptive recommendations, operator brief reasoning, candidate review brief guidance, source health brief guidance, live/fallback/zero-source plan signal status, checked-time freshness, actionable cycle-mode posture, staged safe automation planning, evidence-rich opportunity extraction, calibrated review saves, run audit history, source-health scoring, scoring diagnostics, and review learning.

Assisted discovery

Continuous source expansion, public partner/client/supplier-style link discovery, sitemap/robots discovery, query-pattern source hints, query-hint recovery brief guidance, human-reviewed URL resolving, fallback source-finding playbook, zero-source readiness checks, zero-source route-map visibility with pending-deploy fallback, candidate preview, candidate review brief guidance, confirmed local save, source testing, source health brief guidance, source health review, operator brief reasoning, next-best-action planning, live/fallback plan signal status, checked-time freshness, actionable cycle-mode posture, dedupe, and inserts only after explicit confirmation rules.

Opportunity review

Reviews teach the system which grants, tenders, partner openings, agency signals, and project clues are worth pursuing, while source-origin metrics show which discovery paths produced them.

Draft preparation

Reviewable drafts and response notes only after settings, caps, provider rules, source health, origin learning, scoring diagnostics, guardrail checks, and audit records are configured.

Controlled outreach

Only after approvals, settings, caps, suppression rules, source quality checks, origin metrics, scoring diagnostics, guardrail checks, and run-level audit trail are in place.

built

Planner cockpit

Risk, badges, checklist, compact/full dashboard modes, refresh hints, contract metadata, and self-test.

built

Source memory

Sources can be tracked with state, success/failure counts, cooldowns, quality scoring, notes, created time, saved-source origin, and manual controls.

built

Zero-source startup

The Ops console verifies live Worker readiness, routes no-signal automation plans to start-from-zero mode, highlights the Worker-advertised zero-source route map or a pending-deploy fallback, and explains how to bootstrap candidate-source memory without any supplied manual source list.

built

Source finding fallback playbook

When candidate memory is empty, stale, exhausted, duplicate-heavy, or low-yield, the Ops console now routes the operator through bootstrap seeds, tiny bounded scans, sitemap/robots probing, public-link discovery, and query-hint resolving before increasing depth.

built

Query hint recovery brief

The query-hints panel now explains when to generate hints, which high-score or unused hint to try first, when to open search manually, and when pasted public URLs are ready to resolve into candidate memory.

built

Continuous source expansion

The agent can bootstrap expansion seeds, run bounded seed scans manually or on the scheduled tick, store discovered source candidates, remember quality, and expose candidate memory in Ops without automatically saving live sources.

built

Sitemap and robots discovery

Known source domains can be inspected through robots.txt and sitemap files to discover deeper tender, grant, supplier, funding, and opportunity pages under strict caps.

built

Public link graph discovery

Known public pages can be checked for partner, client, supplier, funding, portfolio, and opportunity-style links. Useful links are scored and saved as candidates only.

built

Query-pattern source hints

The system can generate explainable AU/NZ query hints for tenders, grants, supplier panels, creative funding, and digital-service opportunities without searching the web automatically.

built

Query hint resolver

Operators can paste useful result URLs from a query hint; the resolver scores them, filters noise, dedupes known domains, and stores source candidates for review.

built

Source origin metrics

Live sources are split by origin path so operators can see what came from query hints, public link graph discovery, sitemap discovery, bounded expansion, candidate preview, or manual sources.

built

Ops authentication and confirmation gates

Private operations pages and proxies require operator authentication. Settings writes and manual opportunity runs must carry explicit confirmation flags before reaching the Worker.

built

Adaptive origin recommendations

A read-only recommendation layer compares source-origin yield, active sources, failures, and strategy scores to suggest where the next safe discovery attention should go.

built

Operator brief reasoning

The cockpit condenses the highest-value recommendation into one operator brief with what to do now, why it matters, the relevant evidence, and the guardrail that keeps the action safe.

built

Next-best action reasoning

The cockpit translates origin recommendations into practical operator actions with confidence, risk, payoff, and evidence counters.

built

Plan signal status

The automation plan states whether it is using live recommendations, has no origin signals yet, is loading, has routed to zero-source startup, or has fallen back to conservative guidance, and shows when that check last completed.

built

Actionable cycle-mode posture

The automation plan summarises whether the current operating posture is start-from-zero, learn-first, inspect-first, expand-ready, or review-first and links directly to the recommended next section.

built

Safe automation plan

The Ops console turns live recommendation signals into staged start, learn, inspect, expand, and review steps with guardrails for each phase.

built

Scheduled safe expansion tick

Small recurring expansion scans can run under autonomy caps, source quality, cooldowns, and no-AI/no-email defaults. They store candidate memory only; saving live sources remains confirm-gated.

built

Opportunity intelligence

Dedicated tables, preview, commit-preview, review decisions, and learning summaries for grants, tenders, partner programs, supplier panels, and contract-style demand signals.

built

Preview-only source candidate discovery

Operators can preview scored candidate opportunity sources and duplicate flags before any save step.

built

Candidate review brief

The candidate panel now interprets each batch as promote, inspect, hold, or waiting, then selects only fresh recommended candidates under a capped operator-confirmed save path.

built

Confirmed source candidate save

After preview and review-brief guidance, operators can explicitly confirm selected candidate URLs into opportunity_sources through local metadata-only save while preserving discovery origin.

built

Candidate source testing workflow

Saved candidate sources can be filtered, identified with badges/notes, jumped to from the candidate panel, then tested or previewed from the source manager.

built

Source health brief

The source health panel now interprets the source mix as trust, tighten, pause or repair, review, or waiting before more discovery budget is spent.

built

Review learning loop

Review decisions can feed strategy scores and separate shortlisted, watching, needs-research, rejected, duplicate, and archived opportunities.

built

Run audit history

Opportunity runs can record lifecycle status, source-level results, candidate rejection reasons, and read-only run detail for the Operations Hub.

built

Source health scoring

Run audit history can score sources as strong, useful, noisy, weak, or failing and recommend whether to prioritise, monitor, tighten, or pause them.

built

Source health actions

Operators can confirm local metadata-only actions such as pause, activate, lower priority, raise priority, and reset source error state.

built

Scoring diagnostics

Operators can inspect calibration impact, reason counts, source-health influence, review-learning influence, guardrail counts, warnings, and boosted/penalised opportunities.

built

Calibration guardrails

Scoring now caps adjustment strength, limits low-confidence boosts, applies ceilings without strong signals, and preserves review floors for strong-fit/urgent opportunities.