Darkroom review

Dedicated performance-creative page states a specific throughput: '250 to 600 production-ready assets per cycle' delivered 'every two weeks', across static, motion, UGC and AI formats, with a two-week onboarding. A stated, specific volume and cadence — but the rubric's top band requires corroboration and none exists: no scoped Meta Ad Library check could be performed (the instrument could not render the Ad Library, which is inconclusive, never negative), and no independent source repeats the figure.

What this dimension measures: How many assets does this vendor actually ship per client per month, and how fast is a variant round? A stated, specific throughput (e.g. 'X assets/month', published turnaround of 24-72h per round) with corroboration scores 4-5. Vague 'unlimited requests' claims with a one-at-a-time queue score 2-3 — unlimited requests with a single concurrent slot is a throughput of one. No stated volume anywhere scores 2. Where the vendor names a client, check the client's live ads in the Meta Ad Library — POSITIVE EVIDENCE ONLY, and scope it properly. Correct procedure: run the keyword search, harvest the advertiser's page_id from the rendered HTML, then re-query with view_all_page_id=<id>. Scoped counts are 100% brand-attributed; keyword counts are not scoped to the advertiser and overstated by 41% in one measured case (HexClad, 410 keyword vs 290 scoped), while a search for a generically-named client returned 1,400 results consisting of Viator, Klook and Tripadvisor. A high-volume, recent, scoped result confirms a real programme and may raise this score. A low or absent result is INCONCLUSIVE and must never lower it. Critically: NEVER record an absence from a page ID you did not harvest from a live ad — the userID shown on a logged-out Facebook page is not the Ad Library page_id, and querying it returns a clean, convincing zero. That was proven by control on 2026-08-11 against a brand known to be running 290 ads. Absence of ads is never scored against an agency. See library/_notes/meta_ad_library_method_2026-08-11.json. Counts for 20% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

The page describes a 'two-week integration period to align with brand' and 'human-supervised AI' with senior creatives supervising ideation and variant production — a learn-your-brand onboarding, which the rubric scores 2-3. No explicit brand-guidelines intake, no named preflight/QC step before client review, and no statement about who catches an off-brand asset anywhere on the pages read.

What this dimension measures: Does the vendor work from the client's brand system — logo lockups and clear space, exact colour and type, tone of voice — or from a mood board and taste? Look for an explicit intake of brand guidelines, a named check step before client review, and any statement about who catches an off-brand asset. A documented preflight or QC step scores 4-5. 'We'll learn your brand over the first month' scores 2-3. Rotating contractor pools with no guideline intake score 1-2. Silence on the question scores 2: at volume, brand drift is the default outcome and a vendor that has not thought about it has not solved it. Counts for 20% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

Named format diversity (UGC, static, motion, AI-generated) with 'platform-specific optimization', and the Olipop case shows one campaign adapted across CTV, YouTube pre-roll and Meta — real multi-placement delivery. But no multi-language or multi-market transcreation capability is mentioned anywhere despite the 'global team / 4 hubs' claim, so this scores as format/placement coverage without demonstrated localization.

What this dimension measures: Can one approved concept become 100+ assets across aspect ratios, placements, languages and markets? Named format/placement coverage plus real multi-market or multi-language delivery scores 4-5. Resizing only, single language, scores 3. 'We make ten great ads' scores 2 — excellent and out of category. Distinguish genuine transcreation (copy reworked for a market) from machine translation dropped into a template; the latter is not localization capacity and should be scored as resizing. Counts for 15% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

Stated model is 'hypothesis-led' with senior creatives designing 'testing roadmaps', biweekly delivery 'then continuous iteration on what scales', explicitly optimizing CTR, CVR and ROAS. Darkroom also sells paid media management in-house ($5,000/mo line), so ad-account access for the feedback loop is structurally plausible and a cadence is named. All of this is the vendor's own description; no independent client confirms the loop operates as described.

What this dimension measures: Does anyone on this vendor's team read the ad account? Look for a stated feedback loop: winners scaled, losers cut, next round briefed off results. Access to platform performance data plus a named cadence scores 4-5. Iteration on client-relayed feedback only scores 3. Ship-and-stop production, where the vendor delivers files and never learns what happened, scores 1-2 — this is the most common failure in the design-subscription tier and it is what separates production from performance creative. Counts for 15% of the weighted score.

Evidence: vendor stated — the agency’s own claim, recorded as theirs rather than ours.

No mention of legal review, claim substantiation, rights/licensing on creator or AI-generated content, or accessibility anywhere on the pages read — notable given AI-generated ads are a headline offering. Health & Wellness is a listed vertical (ProHealth is a named client) but no regulated-category process is described. Scored per rubric guidance: absence here is weak evidence, low confidence, not a penalty beyond Weak.

What this dimension measures: Legal disclaimers, claim substantiation, platform specs and safe zones, accessibility contrast, rights and licensing on music/footage/creator content. Explicit process, or regulated-category experience (health, finance, alcohol, children's), scores 4-5. Platform-spec competence only scores 3. No evidence either way scores 2 with LOW confidence rather than a penalty — most SMB-facing shops never discuss this publicly, and absence here is weaker evidence than absence on volume or ownership. Do not fail a small shop for lacking pharma review workflows it was never asked for. Counts for 10% of the weighted score.

Evidence: inferred — our reading of indirect evidence, not a documented fact.

Stronger here: TubeScience scores Strong on the same dimension.

Complete silence on who owns final files, source/working files, and usage rights on creator or AI-generated content — the FAQ covers onboarding and meetings but not ownership. On continuity, the about page says leadership is 'heavily involved in stewarding client accounts' but nothing guarantees the same creatives month to month; one independent Google review alleges an unapproved substitute was sent on a shoot. At an $8,000/mo starting price, silence on rights resolves against the buyer.

What this dimension measures: Who owns the output — final files, source/working files, raw footage, and the rights to creator content? Explicit transfer of source files and full usage rights scores 4-5. Finals only, working files withheld, scores 2. Silence scores 2; at this price point silence usually resolves against the buyer. Separately, continuity: the same named designers month to month scores 4-5, a rotating pool with no continuity guarantee scores 1-2. Documented designer discontinuity in independent reviews is a verified Poor, not an inference. Counts for 10% of the weighted score.

Evidence: inferred — our reading of indirect evidence, not a documented fact.

Stronger here: Sweat Pants Agency scores Strong on the same dimension.

Published starting prices for all 14 service lines on /services, directly observed: Performance Creative 'Starts at $8,000/mo' (Paid Media $5,000/mo, Creator Content $5,000/mo, etc.). That is a real public anchor, but there is no per-asset pricing, no tier ceiling, and no published contract terms — minimum commitment, month-to-month vs annual, and pause policy are all unstated. Clutch lists minimum project size $5,000+.

What this dimension measures: Published per-asset, per-month or per-round pricing scores 4-5. Published tiers with an unstated volume ceiling score 3. Bespoke-only quoting with no public anchor scores 1-2. Twelve-month lock-in on a production service scores 2 — production is inherently variable in volume and a long lock is a transfer of that risk to the buyer. Month-to-month or per-project scores 4-5. Note any minimum spend, and note whether pausing is permitted, since seasonal advertisers pause. Counts for 10% of the weighted score.

Evidence: verified — checked against a named source you can open; the links under Sources below are where to check it yourself.