# Social Sales Safari: evidence, synthesis, and platform translation

Use this reference when public social research replaces interviews for Cosmos Luminosity audience validation.

## Evidence record

For every author, retain:

- exact quote, direct URL, date, and platform;
- explicit female evidence (`woman`, `F`, or first-party/editorial pronouns, graded separately);
- self-stated location only; never infer geography from name, photo, aesthetics, or community;
- role, stage, and resource signals only when directly stated;
- trigger, push, pull, what must be preserved, fear, doubt, alternatives tried, purchases, and action readiness;
- source-access status and any platform bias.

Separate four layers in the final report:

1. direct self-description;
2. observable action or purchase;
3. repeated pattern;
4. analytical hypothesis.

## Verification grades

- **Strong:** original page opens, quote can be reread, identity evidence is explicit.
- **Supported:** original page opens and profile/editorial bio supplies pronouns, but no first-person `I am a woman` statement.
- **Archive-only:** quote came from a public archive/search endpoint and the original platform blocks reopening. Keep the permalink and disclose the limitation.
- **Provisional:** identity or quote context cannot be adequately verified; exclude from primary counts.

An HTTP `403` from a social platform is not proof that a post is absent. Record it as blocked repeat verification, not a dead link. An HTTP `200` confirms page availability, not quote accuracy; retain that distinction.

## Research execution pattern

Large 25-record jobs often exceed a single agent window. Use bounded, incremental passes:

1. ask for 5 records per checkpoint and require a physical partial artifact;
2. cap retries on blocked sources and switch platforms quickly;
3. audit record count, unique authors, unique URLs, fields, platforms, and explicit-geography count after every batch;
6. run separate streams for career transition, founders/leaders, and actual adjacent-tool purchases;
7. never treat a timed-out stream without a saved artifact as evidence.

For a worked codebook and corrective validation examples, see `references/work-transition-voc-and-demand-falsification.md`.

Do not repeatedly re-dispatch an unsaved broad task. Narrow it to 5–15 records and one evidence objective.

## Bias controls

- Reddit/FIRE sources are strong for salary, runway, capital, and fear of leaving, but weak as a whole-market sample.
- Sabbatical archives are strong for identity narratives, but overrepresent people able to take breaks.
- A title or net worth is a resource proxy, not proof of willingness to buy transformation work.
- Likes, free consumption, and expressed interest are not willingness to pay.
- Travel destinations are not residence.
- Do not merge acute clinical crisis, divorce/legal triage, relocation, job-search tactics, practitioner-business needs, and post-success identity transition into one immediate JTBD.

## Demand-falsification checkpoint

Before turning a synthesis into positioning, ask whether the audience actually requests the synthesis itself.

- Multi-use of therapy, coaching, retreats, astrology, meditation, or other modalities does **not** prove demand for one integrated product. It may indicate a preference for role separation.
- Separate: population prevalence → expressed problem → service use → self-paid purchase → explicit desired outcome → preference for an integrated delivery model.
- Never calculate a TAM by adding or multiplying percentages from surveys with different years, countries, age bands, definitions, or unknown overlap.
- If no dataset contains `woman × age band × market × request × payer × transaction`, report the denominator as unavailable rather than naming a paid top-three.
- Treat an elegant category name as a proposed category until customers use it or behavior validates it.
- Lead product tests with a concrete job, not the internal mechanism. For work-transition research, test distinct doors such as `burnout/work redesign`, `stay–redesign–leave`, and `post-achievement next chapter` against the same delivery system.
- Validate with qualified signups, completions, deposits/payments, experiment completion, and review return—not resonance or compliments.

## Voice-of-customer corpus design

For a broad request such as “I can’t keep working like this,” split collection by behavioral stage instead of treating all distress as one persona:

1. **Still employed:** what specifically is unsustainable; whether she wants leave, boundaries, role redesign, a new container, or exit.
2. **After change:** before/action/after; relief, duration, unexpected losses, economics, identity, and what remains unresolved.
3. **Solutions and gaps:** what she actually used or bought, what it gave her, what failed, whether she repeated/switched, and the current desired outcome.

Use five-record physical checkpoints for web-heavy work. If a large batch times out, inspect and continue from the saved record number; do not restart or discard partial evidence.

Code five layers separately: direct voice, observable action, reported outcome, analytical pattern, product hypothesis. Never rewrite an analytical residual gap as “Cosmos Luminosity solves this” without behavioral validation.

Important distinctions:

- “I need a break” may mean one sick day, protected leave, or a workload conversation—not a new identity.
- Leave may enable recovery, planning, or both; do not conflate them.
- Travel, retreats, relocation, entrepreneurship, and quitting can help, fail, or recreate the original problem. Collect mixed and negative outcomes deliberately.
- Returning to employment or a prior field can be a successful transition, not failure.
- Structural workload, span of control, caregiving, benefits, and financial constraints cannot be reduced to mindset or resilience.
- Severe insomnia, functional impairment, substance use, homelessness, abuse, suicidality, or acute psychiatric/medical risk are referral/stabilization signals, not acquisition language.

## Confidence

- **High:** 5+ independent women or 3+ source types, with action/purchase support.
- **Medium:** 2–4 independent women or one source type with behavioral support.
- **Low:** one author, prompted answer, unclear identity, or intention without action.

A dedicated purchase corpus that failed to materialize remains a data gap. Do not inflate scattered mentions of therapy, coaching, meditation, retreats, somatics, spirituality, or energy work into a market-wide preference.

## Translating research into the CL platform

Preserve the validated recognition language, then build a grounded bridge before symbolic tools:

1. **Recognition:** name the functioning-but-too-small life.
2. **Transition Check-in:** identify trigger, what became too small, what must be preserved, live decision, prior attempts, acceptable tools, and safety exclusions.
3. **One-Page Transition Map:** `too small / still alive / preserve / decision / know / assume / test / observe / review date`.
4. **Magic Cloud:** optional symbolic perspective; never the decision-maker or replacement for facts.
5. **First paid Gateway:** a bounded mapping intensive with reversible experiments and observable deliverables.
6. **LumoSide:** implementation and integration after a direction is chosen.
7. **Premium 1:1:** for high-stakes intersections of career, money, geography, relationships, body, and state.

Use transition states rather than personality labels, for example: recovery before redesign; preserve and redesign; bridge rather than leap; between credible futures; post-exit blank page; ready for a real-world test.

## Product guardrails

- Keep the client’s authorship and permission central.
- Offer body/state/symbolic/energy layers by consent and after grounded recognition.
- Do not lead with energy language until evidence supports it for that cohort.
- Do not promise purpose, one correct future, clinical healing, guaranteed income, or manifestation.
- Validate price and demand through deposits/payments and completed experiments, not compliments.
- Research quotes are evidence, not testimonials.

## CL-specific reusable artifact paths

Session-specific research may live under `cosmos_universe_media/audience-research/`; treat those files as mutable evidence artifacts, while this reference contains the durable method.