Park-Enterprise Fit & Target Portfolio
Convert audited candidates into a portfolio of enterprises that the park can verify and advance. First deliver candidates that have already met qualification and evidence requirements for the current batch, then continue according to the frozen quantity target or coverage requirements. When the user has not specified or confirmed priority criteria, use only a transparent, stable display order and declare that it does not represent business priority; only when priority criteria exist should the sorting be based on the current goal, structured judgments, and evidence-derived factors. Under no circumstances use self-reported scores from candidates, nor package them as landing probabilities.
Prerequisites
- A valid
ParkBriefand aTaskGoalthat is frozen or can be clearly revised. - The
TaskGoalis frozen bypark-brief-builder; this Skill only reads that target, it does not create, supplement, or rewrite it. - Candidates have completed entity normalization.
- Claims referenced by frozen rules as key items for qualification, delivery, or completion have auditable evidence; "key" must be derived from
TaskGoal / ParkBriefand their reference relationships, not from the candidate's or caller's self-reportedcritical. - Hard failures, conflicts, and unknowns are explicitly recorded.
When the target is not yet frozen and there are blocking unknowns that would change the candidate pool, return the minimal set of blocking information; merge where possible, and when multiple independent blocking items cannot be safely assumed, ask honestly or pause, and never declare completion or market shortage.
Enterprise Qualification Baseline
Before a candidate can enter the qualified pool, it must simultaneously satisfy:
- The legal entity is real, distinguishable, and deduplicated.
- No confirmed park hard constraint failure.
- Matches the target conditions of the current task.
- Key judgments have auditable publicly available evidence.
Any failure on the qualification baseline cannot be offset by high soft scores. Needing channels, public entry points, or verification scripts does not belong to the enterprise qualification baseline.
Judgment Items Based on Task Applicability
For judgment methods and examples, read ../../references/fit-and-portfolio.md; for degradation and quality gates, read ../../references/quality-and-degradation.md.
Generate applicable judgment items from TaskGoal, hard constraints, and coverage requirements. Common but not mandatory judgment items include:
- Industry or chain-node relevance.
- Venue and engineering conditions.
- Enterprise stage and public signals.
- Park resource exchange value.
- Regional and supply chain synergy.
- Business and operational acceptance conditions.
- Evidence strength, freshness, and conflicts.
- Counter-evidence, unknowns, and failure conditions.
A fixed checklist does not need to be filled for every task nationwide. For each substantive judgment item, record applicability, conclusion, reasoning, and evidence references; if not applicable, explicitly mark as not_applicable; unknowns must not be disguised as passed. When the user has specified or confirmed priority criteria, external terms like "priority / considered / to observe" may be used; otherwise, only express qualification and evidence status, do not imply business priority.
Sorting, Batches, and Cumulative Portfolio
- First exclude hard failures and entity conflicts.
- Place candidates lacking sufficient evidence for frozen rules into the watchlist; they do not count toward the qualified quantity; a candidate's self-reported
critical=falsecannot hide gaps. - Retain only one row per legal entity; one entity can serve multiple coverage roles.
- The current batch only includes candidates that pass the qualification and evidence baseline; if the user has specified or confirmed priority criteria, sort according to that criterion and its evidence to form a priority batch; otherwise, use a transparent and stable display order and declare that it does not represent business priority.
- Batch size comes from user requirements, park team processing capacity, or host display capability; it only controls delivery cadence.
- No product-level candidate cap; the total number requested by the user can be fulfilled across any number of batches.
- Preserve user-specified quantity relationships:
exactstops when N is reached and must not exceed;at_leaststops when the lower bound is reached and coverage is satisfied, and may exceed the lower bound if coverage requires;betweenstops when the lower bound is reached and coverage is satisfied, and must not exceed the upper bound;at_mostcan only be frozen when a positive coverage condition also exists, and must not exceed the upper bound;nonecompletes only by coverage. Ranges can only come from the user, not auto-generated; additionally verified enterprises go into internal alternates unless the user requests otherwise; do not expand delivery on the front-end without authorization. - Coverage targets first select feasible combinations that can meet the requirements; the same enterprise can satisfy multiple coverages simultaneously, but counts only once; among multiple feasible combinations, only compare by business priority when the user has specified or confirmed priority criteria; otherwise, use a transparent and stable display order.
- Chain nodes and roles are completion conditions only when written into
TaskGoal.coverageRequirements; otherwise, they serve only as portfolio diagnostic information. - When keys sources are deleted, or targets or hard constraints change, only recalculate the affected objects and batches, and keep a record of the changes.
- After the first batch delivery, continue research as long as the target is unmet and there is still a clear path; user feedback can adjust sorting, but is not an authorization gate for continuing work.
Action Readiness vs. Enterprise Qualification
Action readiness does not determine whether an enterprise fits the park. Enterprises without a public entry point can still enter the qualified pool, but must be marked as needing supplementation and list supplementary pathways.
For candidates that already meet the conditions, recommend channels, public entry points, channel rationale, alternative paths, and customer operation sequence based on the task and evidence. Channel types are an extensible catalog, not mandatorily mapped by park prototype, scale, or stage; when current evidence is insufficient, primaryChannel=null is allowed.
Public entry points are only used for manual verification and preparation. Evidence pages and action entry points are recorded separately: project, filing, environmental assessment, news, recruitment pages, etc., by default can only prove the corresponding facts; they cannot become contact, cooperation, application, or procurement entry points just because they are accessible. Action entry points must close with the same legal entity, the recommended channel for this task, and a specific public_entry_purpose; when the purpose is inconsistent, short URLs are not expanded, redirects are not verified, or the body text is not read, keep as "pending supplementation". executionBoundary=prepared_not_contacted; do not claim that contact has been made, a relationship established, registration completed, a reply received, or that the enterprise is interested.
State Determination
Task state must be determined relative to the frozen TaskGoal:
in_progress: Target is not yet met, and there is still a clear actionable path.completed: The applicable quantity relationships, coverage, qualification, item-by-item assessment, delivery requirements, and acceptance criteria for this task are all satisfied.stopped_with_gap: Target is not met, and within the frozen scope there is auditable proof of exhaustion.paused: User pause, time window, budget checkpoint, search unavailable, tool failure, or missing target information.superseded: Target or key constraints are replaced by a new version.
A pause is not a shortage. Whether batch delivery completion equals task completion is determined solely by the frozen TaskGoal; it cannot be automatically judged as completed or incomplete. Time or query budget expiration can only form checkpoints and continuation cursors; by itself it cannot prove that there are no more qualified enterprises in the market.
stopped_with_gap must include terminalProof. The open internet, search spaces that can still be expanded, or a caller's self-reported "finished searching" cannot prove exhaustion; only a user-bound closed list, or an enumerable full set with trusted source version, query conditions, pagination/cursor closure, total count, and receipts, can form termination proof. The actual full set can be zero; in that case, "auditing an empty list item by item" can complete, but enterprise-finding tasks with positive enterprise targets can only report a target gap or pause; they cannot fabricate enterprises, queries, or market shortages. The termination proof should at least state:
Unmet target requirements
Frozen search scope and query ledger summary
Evidence of exhaustion or saturation
Still unresolved evidence gaps
Why there are no more continuable directions within the current scope
Prohibitions:
- Lower evidence thresholds.
- Count watchlist as qualified.
- Duplicate groups, subsidiaries, or the same entity to pad numbers.
- Treat a single batch as total task completion.
- Write pauses, tool failures, or budget expiration as market shortages.
- Use park type to directly determine task, customer strategy, or channel.
Output
TargetPortfolio uses the formal Schema for the required candidate portfolio state; nullable, optional, or not applicable collections remain truly empty, not fabricated to fill a form. Core relationships include:
parkBriefRef
taskGoal
batchPolicy
taskState
deliveryBatches[]
currentBatchRef
cumulativeUniqueDeliveredCount
countGoalAssessment
coverageAssessments[]
unmetRequirementRefs[]
priorityCandidateIds[]
qualifiedCandidates[]
alternates[]
watchlist[]
excluded[]
searchCoverage
terminalProof
pauseReason
nextCursor
continuationRequirements[]
portfolioStrategy (only when task basis exists)
qualityGateReceipt
nextBestAction (compatible projection, can be null)
Each qualified candidate includes the legal entity, fit judgments related to the current task, key evidence references, counter-evidence/unknowns, action readiness, and subsequent verification suggestions. performanceObservation is formed by actual observation of execution and visible delivery within the host/orchestration boundary; this Skill does not self-report time consumption or "already visible to user". Normal chat must translate machine state into business language; do not directly display field names or enums.
Write Authority and Prohibitions
- Solely writes candidate judgments, final fit, sorting, batches, cumulative portfolio, target assessment, and task state.
- Does not write
ParkBrieforTaskGoal; target changes must go back topark-brief-builderto form a new version. - Does not modify
EntityRecord,SourceRecord, orClaimEvidenceLink. - Does not add unaudited enterprises.
- Does not call internal scores "landing probability".
- Does not claim enterprises have real intent, budget, or timeline.
- Does not execute external actions.
- Does not write draft preparation as executed result.
Stop
- The current batch has been delivered and cumulative progress saved.
- If the task target has been met, the state is
completed. - If the target is not met and can still continue, the state is
in_progresswith a cursor and direction for the next batch. - If temporarily interrupted, the state is
pausedwith a description of resumption conditions. - Only when exhaustion proof is established can
stopped_with_gapbe used. - If the user requests a file, hand off to
artifact-renderer; otherwise, deliver directly in chat.