Cint
Cint (Cint Exchange, formerly Lucid Marketplace) is a programmatic sample exchange connecting buyers to hundreds of supplier panels across 130+ markets. Flashpoint.AI integrates with the Cint Demand API to let you buy sample at exchange scale — large multi-country studies, low-incidence audiences, and markets where a single panel can't fill on its own.
How it works
Cint uses an exchange model: account → business unit → project → target group → fielding run. The target group is the unit that matters — it carries your targeting, quotas, sample size and price, and it's what fields. The project above it is a thin container.
When you launch a panel, Flashpoint.AI creates a project and a draft target group with your audience expressed as Cint profiling conditions. Cint routes respondents from whichever supplier panels match. Because a live target group draws real panelists from many suppliers at once, launching is a deliberate step — see Approve and launch.
Respondent tracking uses Cint's rid parameter. When a respondent finishes, Flashpoint.AI reports the outcome server-to-server and redirects the browser as confirmation.
Key differences from Prolific and Dynata
| Dimension | Prolific | Dynata | Cint |
|---|---|---|---|
| Model | Study | Project + line items | Project + target group + fielding run |
| Auth | Token header | Password grant | OAuth2 client credentials |
| Feasibility | Eligibility count | Draft project required | calculate-feasibility on the target group |
| Pricing | We set the reward | Rate card | Rate card; Cint also returns a suggested price |
| Quota model | No vendor quotas | Quota cells with counts | Quota cells as percentages |
| Launch | Async, polled job | Buy call | Async, polled job |
| Status delivery | Completion codes | Signed exit URLs | S2S + RIS redirect codes |
| Best for | Academic, UX, consumer | Large-scale MR, B2B | Multi-country, low incidence, high volume |
Targeting
Cint targeting is expressed as profiling conditions against question IDs. Flashpoint.AI translates canonical keys into Cint's profiles[].conditions[] format automatically.
Standard targeting keys
| Key | Question ID | Format | Description |
|---|---|---|---|
gender | 43 | Enum | Biological sex. |
age | 42 | Range — pass age_min / age_max | Age range (1–99). |
zip | 45 | String or list | US ZIP codes. Up to 10,000 entries. |
hispanic | 47 | Enum | Hispanic / Latino origin. |
ethnicity | 113 | Enum | Self-identified ethnicity. |
household_income_us | 14785 | Enum | Annual household income (US). |
employment_status | 2189 | Enum | Current employment situation. |
education | 48741 | Enum | Highest education completed. |
marital_status | 632 | Enum | Relationship status. |
household_composition | 641 | Enum | Total people in household. |
parental_status | 7064 | Enum | Whether the respondent has children. |
Question IDs are global and stable across locales — only the question text translates.
Canonical values
gender: male, female
Cint's gender question has exactly two options. There is no non-binary code on question 43; a request for one surfaces as a warning rather than silently mistargeting.
hispanic: yes, no, plus specific origins — mexican, puerto_rican, cuban, argentina, colombia, ecuador, el_salvador, guatemala, nicaragua, panama, peru, spain, venezuela, other_hispanic, prefer_not_to_answer
This is an origin question, not a yes/no. Passing yes targets the union of every affirmative origin.
ethnicity: white, black_or_african_american, native_american_or_alaska_native, asian_indian, chinese, filipino, japanese, korean, vietnamese, native_hawaiian, guamanian, samoan, other_pacific_islander, some_other_race, other, prefer_not_to_answer
household_income_us: lt_5k, 5k_to_9k, 10k_to_14k, 15k_to_19k, 20k_to_24k, 25k_to_29k, 30k_to_34k, 35k_to_39k, 40k_to_44k, 45k_to_49k, 50k_to_54k, 55k_to_59k, 60k_to_64k, 65k_to_69k, 70k_to_74k, 75k_to_79k, 80k_to_84k, 85k_to_89k, 90k_to_94k, 95k_to_99k, 100k_to_124k, 125k_to_149k, 150k_to_174k, 175k_to_199k, 200k_to_249k, 250k_and_above, prefer_not_to_answer
employment_status: employed_full_time, employed_part_time, self_employed_full_time, self_employed_part_time, active_military, inactive_military_veteran, temporarily_unemployed, between_jobs_looking_for_work, full_time_homemaker, bringing_up_a_family, unpaid_full_time_carer, working_unpaid, parental_leave, leave_of_absence, unable_to_work, disabled, retired, student, other_paid_work, pursuing_own_goals, prefer_not_to_answer
education: some_high_school_or_less, high_school_graduate, vocational_training, some_college_no_degree, associates_degree, bachelors_degree, masters_or_professional_degree, doctorate, none_of_the_above
marital_status: single_never_married, married, engaged, domestic_partnership, separated_divorced_or_widowed, prefer_not_to_answer
household_composition: 1_person, 2_persons, 3_persons, 4_persons, 5_persons, 6_or_more_persons, prefer_not_to_answer
parental_status: pregnant_or_expecting, has_minor_children, has_adult_children, no_children
Attribute search for niche targeting
The keys above are the demographics we translate locally. Cint's full profiling library is far larger and locale-specific, with option codes that differ per market. Use attribute search to reach it — the agent queries Cint's live questions-translation data and returns real question IDs and option codes you can pass straight through.
Agent prompt:
Search Cint for home ownership questions in the UK.
Markets and languages
Cint sells into far more locales than its feasibility engine can price in, and this is the most common source of confusion on international buys.
Two things to know:
The preferred language flag is the market default, not the local language. It reads eng for Pakistan. Target the language your survey is actually written in — a Pakistan study quoted in English succeeds and fields English speakers, which is a wrong answer with no error attached.
A listed locale means Cint sells there, not that it can price there. Uzbekistan is listed in three languages and priceable in none. Use the feasibility probe on a single country to find out which locales respond.
Agent prompt:
What languages can Cint actually field in Kazakhstan?
API call:
curl https://api.flashpoint.ai/api/v1/panels/cint/locales \
-H "Authorization: Bearer $TOKEN"
The full flow
Step 1: Check targeting filters
curl https://api.flashpoint.ai/api/v1/panels/cint/filters \
-H "Authorization: Bearer $TOKEN"
Step 2: Get a quote
Feasibility and price come from the same Cint call. The response carries a suggested completes range and a suggested price per complete.
Agent prompt:
Quote 500 US adults 25-54, household income $75k+, on Cint.
API call:
curl -X POST https://api.flashpoint.ai/api/v1/surveys/$SURVEY_ID/panels/quote \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d '{
"provider_id": "CINT_PROVIDER_UUID",
"sample_size": 500,
"country": "US",
"targeting": {
"age_min": 25,
"age_max": 54,
"household_income_us": ["75k_to_79k", "80k_to_84k", "85k_to_89k"]
}
}'
If Cint's feasibility engine doesn't model the locale you asked for, the quote says so explicitly rather than reporting zero availability.
Step 3: Review the draft
Before any charge, Flashpoint.AI re-reads the draft target group from Cint and verifies the profiling conditions actually landed. Cint's read endpoint lags writes by a second or two, so this check retries before it blocks.
Step 4: Approve and launch
Payment is collected via Stripe inline in chat. The target group is not launched automatically. Once payment clears, the buy is parked as paid-and-ready and our operations team launches it from the Cint Exchange after a sanity check.
This is deliberate. A launched target group draws real panelists from live supplier panels within seconds, and an unreviewed launch is expensive in a way that can't be undone — a misconfigured test buy once sent 353 real prescreens to the wrong survey URL before Cint auto-paused it for poor conversion.
Your panel card shows the buy as live from the moment payment clears. Status flips to actively fielding when the target group starts.
Step 5: Monitor status
Agent prompt:
How's the Cint panel doing?
Response shape:
{
"status": "active",
"completes": 418,
"starts": 1240,
"disqualified": 762,
"over_quota": 60
}
Step 6: Pause and resume
Agent prompts:
Pause the Cint panel.
Resume recruiting on Cint.
How quotas work
Cint's quota model is percentage-based, unlike Dynata's counts. Flashpoint.AI translates your survey's quota records into Cint target cells, converting counts to percentages of the requested sample size.
Not everything translates, and what doesn't is reported as a warning rather than dropped silently:
| Quota type | Fields on Cint? |
|---|---|
| Gender, age range, standard demographic labels | Yes |
| The overall total | No — that's the sample size itself |
| Quotas on your own survey questions | No — Cint doesn't know your questionnaire |
| Interlocked quotas (age and gender in one cell) | Not yet — needs Cint blended profiles |
Anything that doesn't field on Cint is still enforced on our side at survey time, so your quota targets hold either way. The difference is efficiency: a split Cint can't field means respondents reach your survey and get screened out as over-quota, which wastes exchange sample and bounces panelists.
Status delivery and redirects
Cint is the one provider where we report outcomes server-to-server rather than relying on the browser.
On finish, Flashpoint.AI posts the final status to Cint's fulfillment API and then redirects the respondent as confirmation. S2S is the more robust path — a browser redirect can be lost, blocked, or spoofed, and losing one means a ghost complete.
Redirect outcomes use Cint's RIS client status codes:
| Outcome | RIS code |
|---|---|
| Complete | 10 |
| Terminate (screener or survey logic) | 20 |
| Quality terminate | 30 |
| Over-quota | 40 |
Over-quota is reported as 40, never collapsed into 20 — Cint prices and paces on the distinction. Code 30 is reserved for genuine data-quality failures: it affects panelist reputation scores and can trigger supplier disputes, so an ordinary terminate never uses it.
Reference data
Cint's profiling library, locale list, and option codes are fetched live from the Demand API and cached. Question IDs are stable across locales; option codes and question text are not, so niche targeting is resolved per market through attribute search rather than from a static map.
Next steps
- Prolific — academic and UX research, smaller and faster.
- Dynata — large-scale and B2B research with a deep attribute taxonomy.
- Synthetic — pre-flight your survey with AI personas before spending on real respondents.
- Data quality — fraud and quality screening applied to every panel source.