Review — ignaciocarcavallo

← overview

ventana últimos 7d vs 7d previos · tz America/Argentina/Buenos_Aires · último cambio Ninjo: 2026-09-23 18:49 (deploy config, hace 5d) · 119 untracked
47
Nuevas (7d)
prior 19
2
Calls booked
— de link
0
Link sent
0.0% de activas
$0
Revenue variable
119 msgs

Resumen — período vs período

Métricathis 7dprior 7dΔ
Activas8137+118.9%
Nuevas4719+147.4%
Link sent00—
Calls booked21+100.0%
Booked/Link %———
Booked/Activas %2.5%2.7%-0.2pp
Booked/Nuevas %4.3%5.3%-1.0pp
Revenue variable$0.00$0.00—

Funnel — semana actual vs anterior

PeríodoActivasNuevasLink sentCalls bookedBooked/LinkBooked/ActivasBooked/Nuevas
Semana actual (7d)814702—2.5%4.3%
Semana anterior (7d)371901—2.7%5.3%
Ver desglose por día
DíaActivasNuevasLink sentCalls bookedBooked/LinkBooked/ActivasBooked/Nuevas
09-2710600—0.0%0.0%
09-2612800—0.0%0.0%
09-253000—0.0%—
09-2410900—0.0%0.0%
09-2310401—10.0%25.0%
09-22171001—5.9%10.0%
09-21191000—0.0%0.0%
09-200001———
09-197400—0.0%0.0%
09-189400—0.0%0.0%
09-177500—0.0%0.0%
09-164300—0.0%0.0%
09-156200—0.0%0.0%
09-144100—0.0%0.0%

⏳ Sin re-auditar hoy (sin cambios de prompt) — mostrando la última auditoría disponible. Estado vigente hasta nueva corrida.

run 2026-07-22T15:07:02+00:00 · fuente: live DB state (daily_audit pipeline)
10/10
BP semánticos PASS
top-10 best practices
11/16
Determinísticos PASS
2 FAIL · 3 WARN
C0 · H0 · M0
Fails por severidad
critical / high / medium
—
Drift
DB vs repo local

BP semánticos (top-10 best practices)

BP-1
BP-2
BP-3
BP-4
BP-5
BP-6
BP-7
BP-8
BP-9
BP-10
BPSevEstadoDetalle
BP-1HIGHPASSTemplate compliance
All 7 required sections present in system.md: Identity, Voice, Flow, Calibration, Principles, Tools, NO_RESPONSE. Content matches principles template specification.
BP-2MEDIUMPASSExample coverage and cap (7 base scenarios + conditional, ≤20 total)
15 examples provided. All 7 base scenarios covered: happy path (Ex 1), resource delivery (Ex 4, 6), objection handling (Ex 5), disqualification (Ex 7), emotional lead (Ex 13), re-engagement (Ex 8), off-topic (Ex 10). Both conditionals met: keyword trigger (Ex 4, 6) and price objection (Ex 5). Count is within cap.
BP-3HIGHPASSExamples align with the flow
Every example follows the 6-step flow (Calificar → Dolor → Contexto → Posicionar → Cerrar → Post-booking) in order. All examples reach terminal states (link sent, NO_RESPONSE, or booking confirmed).
BP-4CRITICALPASSNo internal contradictions in system.md
System.md is internally consistent. The 25-word limit on Voice is stated once and applies uniformly. Flow steps are ordered and non-contradictory. The post-booking rule ('Solo si la respuesta es genuinamente ambigua... preguntá UNA vez') is an explicit exception to the NO_RESPONSE rule, not a contradiction. Example 1 demonstrates this carve-out correctly.
BP-5CRITICALPASSNo cross-file contradictions
All SDK files align. Keywords.json declares send_without_permission flags that govern delivery timing; system.md's Flow step 1 ('Si el lead pregunta por un keyword... entregá el recurso primero') is consistent with keyword conditions. Examples.md demonstrates both rules coexisting (Ex 4, 6 show keyword delivery then qualification). Program.json pricing rules match system.md's Principles (no Elite/Group prices in chat; AI Brain and Hiring Course allowed). No factual contradictions on revenue bands, URLs, or program details.
BP-6MEDIUMPASSVoice consistency across examples
All examples use consistent voice: sharp, direct, no BS, rioplatense voseo, max 25 words per message (except intro), no emojis, no em dashes. Tone is uniform across scenarios (happy path, objections, disqualification, emotional). A reader would recognize the same agent throughout.
BP-7CRITICALPASSUnguarded actions (PAT-004)
All risky actions are guarded. Sending links is guarded by revenue band (Step 4 posicioning) and explicit lead acceptance ('dale', 'sí', 'me interesa'). NO_RESPONSE is guarded by 7 explicit conditions in Principles §7. Keyword delivery is guarded by send_without_permission flags in keywords.json. Example 1 demonstrates the guard on post-booking ambiguity ('pudiste agendar?'). No unguarded actions exist.
BP-8HIGHPASSNo stale version language in examples
All examples use current flow language and CTAs. Closing language ('va + URL + cuando reserves avisame', 'bárbaro, te llega ahí el material', 'NO_RESPONSE') matches system.md's Flow and Principles. No outdated phrasing detected.
BP-9MEDIUMPASSNo duplicated instructions
Each behavioral instruction has a single canonical home. Revenue routing lives in knowledge_base.md (not repeated in system.md). NO_RESPONSE triggers are stated once in Principles §7. Post-booking sequence is in knowledge_base.md. Keywords.json notes cross-reference the flow ('resume normal qualification flow') rather than duplicating. No instruction appears in two places.
BP-10MEDIUMPASSRule-vs-principle balance
Behavioral guidance (tone, flow, conversational style) is expressed via Voice examples and Principles. Hard rules are reserved for facts (revenue bands, pricing, URLs, eligibility). The 25-word limit is stated as a rule but is taught by Voice examples. Examples demonstrate all key behaviors (objection handling, disqualification, emotional leads) rather than prescribing them as rules.

Checks determinísticos

6a
6c
6d
6e
6g
6h
6i
6j
6k
6l
6m
6n
6q
6r
6s
6t
CheckSecEstadoNotas
6a1.8PASSNO_RESPONSE present (limits not checked, --no-db)
6c3.xPASS113 lines
6d1.2PASS20 examples
6e1.3FAIL14 principles, cognitive overload
6gPAT-006WARNnon-STEP-0 keyword handling
6hAnti-patPASSno 'burbuja'
6iAnti-patPASSno HTML comments
6jAnti-patWARN2 em-dash(es) in 2 file(s); likely incidental
6kPAT-006PASSno tool syntax leak
6lResourcesPASSno placeholder URLs
6mHygienePASSno audit metadata in prompt files
6nPAT-008PASSURL forms consistent across files
6qPAT-001PASSno instruction duplication into examples.md, keywords.json, program.json, resources.json
6rKeyword-driftFAIL5 in keywords.json + recent digest, never named in system.md: ['ceo', 'ops', 'fire', 'foco', 'testing']; (1 stale drift entr(ies) ignored — no recent digest mention)
6sHygienePASSno expired dates in live SDK
6tCoverageWARN10/14 principle(s) without example: #2 No des el precio, #3 Si alguien parece menor, #6 No des coaching, estrategia, #7 Respondé con la palabra, #8 Si el mensaje solo; 4/6 flow step(s) without example: #2 Dolor, #3 Contexto, #4 Posicionar, #6 Post-booking
canal ignaciocarcavallo-ninjō · sync 2026-09-28
136
Digests
11 en últimos 14d
1
Cambios pedidos / pendientes
del checklist
17
Threads
en el canal
2026-09-28
Última sync
del digest

Cosas nuevas (digests recientes)

FechaTemaTags
2026-09-28weekly funnel report sep 21 28 42 percent link conversionmetrics, funnel, conversion, leads, keywords
2026-09-23gender detection error system prompt reviewagent-behavior, system-prompt, bug-fix
2026-09-23external message pause config investigationagent-configuration, external-messages, troubleshooting
2026-09-23agent activation bug external pausebug, agent-behavior, critical
2026-09-22reports distribution setupoperations, reporting, agent-monitoring
2026-09-21weekly funnel report sep 14 20 emprendedor keywordmetrics, keywords, conversion, content-strategy
2026-09-21weekly funnel metrics sep 14 20 zero conversionsmetrics, funnel, keywords, performance, conversion-optimization
2026-09-18dual permission instagram messaging verifiedinstagram, messaging, manychat, ninjo, permissions, verified
2026-09-17instagram manychat ninjo routing conflict resolutioninstagram, manychat, ninjo, integration, bug-fix
2026-09-16manychat instagram messaging conflict fixbug, integration, instagram, manychat, ninjo, urgent
2026-09-14funnel analysis sep 07 13 volume spike keyword winsanalytics, funnel, keywords, conversion, agendas
2026-09-10say hi to new followers feature requestfeature-request, ninjo, manychat, automation
2026-09-07funnel analysis sep 01 07 keyword performanceanalytics, funnel, keywords, conversion, leads
2026-08-31weekly funnel report aug 25 31reporting, funnel-metrics, lead-qualification, performance
2026-08-31weekly funnel report aug 25 31 volume growthfunnel-metrics, lead-quality, agent-performance, icp-validation
2026-08-24weekly funnel report aug 18 24funnel, metrics, weekly-report, leads, conversion
2026-08-17weekly funnel report aug 11 17 low volumeweekly-report, funnel-metrics, content-strategy, lead-follow-up
2026-08-17weekly report low volume keyword signalweekly-report, funnel-analytics, content-strategy, keyword-emprendedor
2026-08-14instagram account restriction resolvedinstagram, account-restriction, agent-pause, resolved
2026-08-14mcp working notes claude project setupmcp, claude-project, agent-setup, documentation

Cambios pedidos por el creador / pendientes

IDEstadoCambio pedido
M2pendienteClarity Call (team)
sentiment del cliente → Ninjo · scoreado ahora (2026-09-28 20:13:09) · modelo claude-sonnet-4-6 · ventana 7d
+1
Sentiment score
conforme
Bajo
Churn risk
del lado del cliente
1
Cambios pedidos
pendientes en checklist
7
Volumen cliente
agente: 47 convs

Matriz de volumen

enganchado sano
Cliente activo con tono ok — buena señal de retención.

Sentiment del cliente

📋 Cliente operativamente activo y sin señales de frustración expresada hacia Ninjo; los bugs reportados (auto-activación, género) fueron resueltos el mismo día, lo que neutraliza el riesgo. No hay citas textuales del cliente disponibles en los digests para evidenciar sentimiento positivo o negativo explícito — todo el tono proviene del equipo de Ninjo (cortex, Benjamín, Benji). Acción CS: solicitar al equipo operativo que documente frases literales del cliente en futuros digests; programar check-in proactivo para validar satisfacción con la resolución de los bugs y el rendimiento del funnel (42% conversión a link).
SeñalValorEvidencia (cita del cliente)
Enojado con la plataformano—
Es quejasí—
Queja constructiva (accionable)sí—

Evidencia citada

SeñalCita textualFecha
bug reportado: agente se activa pese a estar deshabilitado (múltiples veces en la semana)“None”2026-09-23
bug reportado: error de detección de género en respuestas del agente“None”2026-09-23
engagement operativo activo: Benji monitorea y actualiza el agente diariamente, solicitó ser incluido en reportes“None”2026-09-22
historial extenso de iteraciones y mejoras al prompt (tono, reglas, ejemplos reales) — indica inversión sostenida en la plataforma“None”—
adopción de métricas y funnel tracking — cliente usa activamente los reportes para decisiones de contenido y conversión“None”2026-09-28

Cambios pedidos por el creador (Métrica 1)

Un creador que pide cambios concretos está enganchado (mejor señal que el silencio).

IDEstadoCambio pedido
M2pendienteClarity Call (team)