You are helping a B2B buyer choose a partner for recurring paid-social creative production. Work only from the material pasted below. Do not browse unless I explicitly ask.

REFERENCE METHOD
1. Separate brand-system work from recurring ad production. This decision covers recurring paid-social concepts, iterations, formats, adaptations, revisions, rights, naming and measurement handoff.
2. A concept is a distinct audience plus problem or angle package. An iteration changes the hook, copy or treatment while retaining that concept. An adaptation changes format, aspect ratio, duration or approved segment. Use the buyer's definitions when supplied.
3. Evaluate every Must have before weighted scoring. Fail on one confirmed Must have produces Exclude unless the buyer explicitly grants an exception. Unknown produces Hold for evidence. Never let a weighted score override a Must-have Fail or Unknown.
4. Public service language supports only the capability it expressly names. A portfolio shows observable production craft, not targeting, delivery, authorship, capacity or performance.
5. Preserve Unknown. Do not convert it to No, zero, a weak score or Not applicable. Not applicable requires an explicit buyer decision.
6. Do not invent pricing, capacity, turnaround, revisions, staffing, source-file rights, results, URLs or owner opinions. Separate agency claims from independent evidence.
7. Use stable source pointers and dates for every material candidate judgment. More public evidence does not make a candidate a better fit.

INPUT A: BUYER CONTEXT
Company and offer: [PASTE]
Approved audience, claims and proof: [PASTE]
Channels: [PASTE]
Internal decision, brand, claim and media owners: [PASTE]
Required start and decision dates: [PASTE]
Commercial bounds: [PASTE]
Work that is outside this engagement: [PASTE]

INPUT B: PRODUCTION SPECIFICATION
Paste the completed specification, including monthly workload, definitions, formats, input checklist, approvals, revisions, turnaround clock, promised duration and delivery milestones, source-file and license rules, AI/data constraints, naming/versioning, archive/offboarding and measurement handoff.
[PASTE]

INPUT C: HARD GATES
For each Must have, paste the pass evidence, fail condition, owner and due date.
[PASTE]

INPUT D: CANDIDATE RECORDS
Repeat for each candidate:
Candidate: [NAME]
Model: [in-house / freelancer or studio / subscription service / broad creative agency / ad-focused partner / other]
Source checked date: [DATE]
Source pointers: [URLs, proposal sections, references]
Documented scope: [PASTE]
Comparable production evidence: [PASTE]
Capacity and turnaround: [PASTE]
Revision rule: [PASTE]
Team and backup coverage: [PASTE]
Source files, licenses and rights: [PASTE]
Naming and measurement handoff: [PASTE]
Fee, software and pass-through costs: [PASTE]
Term, cancellation and offboarding: [PASTE]
Known conflicts or dependencies: [PASTE]
Existing Unknowns: [PASTE]

METHOD
- Audit buyer inputs first. If offer, channel, workload, Must haves, approval owner or decision date is missing, keep the selection provisional and request the exact input.
- For each candidate, build a Must-have gate table using only Pass, Fail, Unknown or Not applicable. Cite the source pointer beside the state.
- Exclude confirmed Must-have failures. Hold decision-critical Unknowns. Score only candidates that pass every Must have.
- For eligible candidates, calculate weighted score as sum(weight percent multiplied by score divided by 5). A score requires cited evidence. If evidence is insufficient, use Unknown and do not compute a final total for that candidate.
- Keep buyer fit separate from evidence confidence.
- Turn every decision-critical Unknown into one exact question, requested proof, owner and due date. Use Unscheduled when no due date is supplied.

REQUIRED OUTPUT
1. INPUT CHECK: present, conflicting and missing decision-critical inputs; say whether final selection is possible.
2. MODEL FIT: best-fit production model and buyer-specific reason; plausible alternative and tradeoff when supported.
3. HARD-GATE MATRIX: one row per Must have and one column per candidate with state and source.
4. ELIGIBLE-CANDIDATE SCORECARD: only candidates that passed every Must have; show inputs, weights and math.
5. DISPOSITIONS: Shortlist, Hold for evidence or Exclude for every candidate, tied to a buyer requirement and source.
6. EVIDENCE REQUESTS: candidate-specific Unknowns ordered by ability to change the decision, each with exact question, proof, owner and due date.
7. CONTRACT CHECKS: scope, definitions, quantities, revisions, clock rules, duration and delivery milestones, team, rights/licenses, AI/data, IDs, measurement, fees, term and offboarding.
8. NEXT DECISION: one concrete next action. If final selection is blocked, say what must arrive first.