Campaign awareness: what the platform can't see

Platforms count clicks and messages. Whether those messages turned into customers — or were garbage — only the user knows. /ads conciencia makes Claude explicitly aware of that gap: it stores the user's dated answers, asks again on a schedule, and cross-references them with platform metrics to compute the campaign's real success rate.

Storage (handled entirely by scripts/conciencia.py, never hand-edited):

• Source of truth: ~/.claude-ads/conciencia/<platform>-<campaign_id>.json (validates against ads/references/conciencia-schema.json).
• Human view: conciencia-de-campana.md in the working folder (or conciencia-<platform>-<campaign_id>.md when the folder tracks more than one campaign).

Formulas, thresholds and how to read the cross-reference live in ads/references/conciencia.md. Load it before presenting results.

Core rules

1. Profile first. python3 scripts/profile.py get. Exit 2 → one-line tip to run /ads start, then continue inline.
2. Never assume a business outcome. Anything the user did not report is null and shows as N/A with what is missing. Never write 0 for "didn't say". A ratio is only computed from check-ins where both sides were reported — the script already enforces this; don't recompute by hand.
3. Ask once, all together. Each check-in is ONE message with every question. Whatever the user skips stays null; say so in one line.
4. Cite the numbers. Every verdict names the figure that triggered it and how many check-ins it rests on: "tasa de cierre 11.8 % (10 cerrados / 85 conversaciones, 2 check-ins)".
5. E-commerce with purchase pixel doesn't need this. The platform already sees the sale.

Subcommand routing

Phase 1 — Does awareness apply?

Read context.industry and context.primary_goal from the profile (or the campaign the user names).

• industry = ecommerce and conversion is a pixel purchase → tell the user in one line that the pixel already tracks the sale, so this isn't needed. Proceed only if they insist (register with --channel pixel_purchase; such campaigns are never flagged as due).
• Objective is leads / messages / calls / forms, or the close happens on WhatsApp, Messenger, IG DM, a form or a phone call → it applies. Continue.

Phase 2 — Register

1. Look for campaigns to offer: latest *-audit-results.json in cwd, then ~/.claude-ads/history/index.json. Audit JSON has account_id; campaign names / IDs come from the audit's evidence or the platform data pulled in the audit (MCP / scripts/api/<platform>_fetch.py / export). If none found, ask the user.
2. Ask ONCE for everything missing: which campaigns, conversion channel (WhatsApp / Messenger / IG DM / form / call / landing), average ticket value and currency (optional — unlocks estimated ROAS), how often to check in (default 7 days).
3. For each campaign:

python3 scripts/conciencia.py register --platform meta --campaign-id 120210000000001 \
  --name "WhatsApp Clínica" --objective messages --channel whatsapp \
  --ticket-value 2400 --currency MXN --every 7

4. Record what the audit already established as known facts, with source:

python3 scripts/conciencia.py know meta-120210000000001 \
  --fact "CPL de plataforma 52 MXN (auditoría 2026-10-20)" --source audit

5. Go straight into a first check-in (Phase 3) if the campaign has been running ≥7 days; otherwise tell the user when Claude will ask.

Phase 3 — Check-in

1. Platform side — pull spend, clicks and messages/leads for the period since the last check-in, using the saved tier in connections.<platform>.tier: MCP → scripts/api/<platform>_fetch.py → ask for an export / screenshot. Record which one in source_of_platform_metrics. If nothing is available, leave the block null and say so — never estimate.
2. Business side — one message, all questions:

Sobre <campaign name>, del <from> al <to> (la plataforma contó <messages> mensajes):

1. ¿Cuántas conversaciones reales te llegaron?
2. ¿Cuántas eran gente con intención real (calificados)?
3. ¿Cuántos clientes cerraste?
4. ¿Cuánto ingreso generaron? (opcional)
5. ¿Cuántos fueron basura — spam, fuera de zona, no contestan?
6. ¿Algo que notaste? (p. ej. "preguntan precio y desaparecen")

Lo que no sepas, sáltalo: queda como N/A, no como cero.

3. Save:

python3 scripts/conciencia.py checkin meta-120210000000001 --json '{
  "period_from": "2026-10-20", "period_to": "2026-10-27",
  "source_of_platform_metrics": "mcp",
  "platform_metrics": {"spend": 2800, "clicks": 850, "messages": 55},
  "user_reported": {"conversations": 40, "qualified": 14, "closed": 6,
                    "revenue": null, "junk": 12, "notes": "preguntan precio y desaparecen"}
}'

4. If the user says "later": python3 scripts/conciencia.py snooze <key> --days 1.
5. Continue to Phase 4.

Phase 4 — Cross-reference and verdict

1. python3 scripts/conciencia.py show <key> → read metrics.cumulative, metrics.latest, metrics.trend.
2. Read ads/references/conciencia.md and classify the campaign using its reading table (platform says good / business says bad, etc.).
3. Render: python3 scripts/conciencia.py render <key> --out conciencia-de-campana.md.
4. Chat summary — short, human, money first:

<campaign> — conciencia al <date> (<n> check-ins)

Lo que sé:
  · Tasa de cierre 11.8 % (10 cerrados / 85 conversaciones reales)
  · Costo por cliente cerrado 580 MXN · ROAS real 8.6x (1 check-in con ingreso)
  · 37.6 % de los contactos fueron basura (32 / 85)
Lo que no sé:
  · Ingreso del primer periodo
Lectura: <one line from the reading table>
Siguiente paso: <one concrete action, ranked by money>
Próximo check-in: <date>

5. If the verdict points at lead quality (junk rate high, platform vs real gap), suggest the matching fix from conciencia.md and point at /ads next, which now ranks these items.

Phase 5 — Save state

python3 scripts/profile.py set last_command "/ads conciencia"
python3 scripts/profile.py set last_command_at "$(date -u +%FT%TZ)"

Error handling

• Script exit 2 (campaign not found) → offer registrar.
• Exit 3 (bad value, e.g. negative or decimal count) → explain in one line which answer didn't fit and re-ask only that one.
• Exit 4 (IO) → explain, suggest checking ~/.claude-ads/ permissions.

Output contract

• ~/.claude-ads/conciencia/<platform>-<campaign_id>.json (via script only).
• conciencia-de-campana.md (or per-campaign name) in cwd.
• No changes to ad accounts. This command only reads platforms and asks.