Review — avaperu.oficial

← overview

ventana últimos 7d vs 7d previos · tz America/Argentina/Buenos_Aires · último cambio Ninjo: 2026-06-01 16:31 (deploy config, hace 2d)
287
Nuevas (7d)
prior 408
18
Calls booked
40.9% de link
44
Link sent
5.9% de activas
$27
Revenue variable
777 msgs

Resumen — período vs período

Métricathis 7dprior 7dΔ
Activas747715+4.5%
Nuevas287408-29.7%
Link sent4465-32.3%
Calls booked1813+38.5%
Booked/Link %40.9%20.0%+20.9pp
Booked/Activas %2.4%1.8%+0.6pp
Booked/Nuevas %6.3%3.2%+3.1pp
Revenue variable$27.20$30.28-10.2%

Funnel — semana actual vs anterior

PeríodoActivasNuevasLink sentCalls bookedBooked/LinkBooked/ActivasBooked/Nuevas
Semana actual (7d)747287441840.9%2.4%6.3%
Semana anterior (7d)715408651320.0%1.8%3.2%
Ver desglose por día
DíaActivasNuevasLink sentCalls bookedBooked/LinkBooked/ActivasBooked/Nuevas
06-0289205480.0%4.5%20.0%
06-017506466.7%5.3%
05-31282200.0%0.0%0.0%
05-30103314125.0%1.0%3.2%
05-29134687342.9%2.2%4.4%
05-28145731417.1%0.7%1.4%
05-27173936583.3%2.9%5.4%
05-261909318422.2%2.1%4.3%
05-2525221518316.7%1.2%1.4%
05-2422163133.3%4.5%6.2%
05-2342165120.0%2.4%6.2%
05-2259206116.7%1.7%5.0%
05-2177197114.3%1.3%5.3%
05-2073298225.0%2.7%6.9%
run 2026-06-02T17:28:49+00:00 · fuente: live DB state (daily_audit pipeline)
10/10
BP semánticos PASS
top-10 best practices
9/16
Determinísticos PASS
2 FAIL · 5 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
meta.md declares template='scripted', which specifies a 14-section structure (identity_and_context, style_and_tone, engagement_signals, conversation_logic, happy_path_general_example, knowledge_base, prequalification_questions, trigger_keywords, lead_magnets, qualification_criteria, agent_instruction, instruction_priority, hard_rules, pre_send_checklist). The agent's system.md follows a 6-section 'principles' template structure (Identidad, Voz, Flujo, Calibración, Principios, Herramientas, TEMPLATE PRECIOS) with no engagement_signals, instruction_priority, or pre_send_checklist sections.
evidence (2)
  • {'file': 'meta.md', 'quote': '**Template**: scripted'}
  • {'file': 'system.md', 'quote': '## Identidad\n...\n## Voz\n...\n## Flujo\n...\n## Calibración\n...\n## Principios\n...\n## Herramientas'}
BP-2MEDIUMPASSExample coverage and cap (7 base scenarios + conditional, ≤20 total)
examples.md contains 23 examples, exceeding the 20-example cap. However, the agent has keywords.json (4 keywords) requiring keyword-trigger examples, and program.json declares pricing rules requiring price-objection examples. The examples cover all 7 base scenarios (happy path, resource delivery, objection handling, disqualification, emotional lead, re-engagement, off-topic) plus conditional scenarios (keyword triggers in Examples 4, 17, 18; price objections in Examples 5, 6, 19, 20). The excess is driven by legitimate conditional coverage and detailed branching scenarios (Bootcamp variants in Examples 21, 22, 23). While over 20, the coverage is complete and justified by the agent's configuration.
BP-3HIGHPASSExamples align with the flow
All examples in examples.md follow the flow defined in system.md (Verificar → Precalificar → Contexto → Nutrir → Cerrar) and reach terminal states (link sent, NO_RESPONSE, or graceful closure). Examples progress through steps in order without skipping required stages.
BP-4CRITICALPASSNo internal contradictions in system.md
system.md contains one potential tension: 'Regla universal: Si la verificación profesional NO se completó, verificar primero antes de cualquier acción' vs. 'Keyword triggers: Verificar vet PRIMERO, después continuar flujo principal.' However, these are not contradictory — both state that verification must occur first. The keyword rule explicitly names verification as the precondition, not an exception. No shared precondition produces conflicting outputs.
BP-5CRITICALPASSNo cross-file contradictions
All SDK files align on behavior and data. examples.md demonstrates the flow and pricing rules from system.md correctly. keywords.json notes state 'Verificar vet primero' (matching system.md's universal rule). program.json pricing rules match TEMPLATE PRECIOS in system.md. resources.json provides the single call_admision link used consistently across examples.md and system.md. No file declares behavior or data that contradicts another under the same precondition.
BP-6MEDIUMPASSVoice consistency across examples
All examples use the same persona (Francisco, warm and consultive), vocabulary (doc, chévere, qué bueno), formality level (casual, WhatsApp-style), and dialect (neutral Spanish, tuteo, no voseo). No jarring tonal shifts across the 23 examples.
BP-7CRITICALPASSUnguarded actions (PAT-004)
The primary risky action — sending the call link — is guarded in system.md: 'Nunca enviar el link directo, siempre proponer la reunión, explicar el valor y pedir permiso. Enviar solo después del sí del lead.' This guard is demonstrated consistently across Examples 1, 2, 5, 9, 19, 20, 21, 22, 23, where the agent asks 'te gustaría que te lo comparta?' and only sends the link after affirmative response ('si dale' / 'dale'). The guard is explicit and working examples confirm its intended behavior.
BP-8HIGHPASSNo stale version language in examples
All examples use current flow language and CTAs from system.md. Closing language ('me avisas si los horarios no te llegan a funcionar, estoy atento a cualquier cosa') matches the current flow definition. No stale phrases from prior versions detected.
BP-9MEDIUMPASSNo duplicated instructions
Each behavioral instruction has a single canonical home. The verification rule appears once in Flujo (step 1) and is reinforced by a principle ('Verificar siempre antes de avanzar') — this is principle + example reinforcement, not duplication. The pricing template is stated once in TEMPLATE PRECIOS and referenced in objections.md with a cross-reference, not duplicated. No instruction is restated in two locations.
BP-10MEDIUMPASSRule-vs-principle balance
Behavioral guidance (tone, flow, conversational style) is expressed via Voz section (principles + examples) and Flujo (flow steps with examples). Hard rules are reserved for factual constraints (programs, pricing approach, support email, link fallback). The balance is appropriate — behavior is shown through examples and principles; facts remain 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.xPASS130 lines
6d1.2PASS27 examples
6e1.3PASS7 principles
6gPAT-006WARNnon-STEP-0 keyword handling
6hAnti-patPASSno 'burbuja'
6iAnti-patPASSno HTML comments
6jAnti-patWARN3 em-dash(es) in 1 file(s); likely incidental
6kPAT-006FAILtool-call syntax appears as raw text in examples.md (9 line(s))
6lResourcesPASSno placeholder URLs
6mHygieneWARNaudit metadata (dates/slack refs) in 1 file(s) — adds noise to live prompt
6nPAT-008PASSURL forms consistent across files
6qPAT-001PASSno instruction duplication into examples.md, keywords.json, program.json, resources.json
6rKeyword-driftFAIL1 in keywords.json + recent digest, never named in system.md: ['METODO']; (2 stale drift entr(ies) ignored — no recent digest mention)
6sHygieneWARN2 expired date reference(s) in SDK files
6tCoverageWARN3/7 principle(s) without example: #2 Un bloque = una, #6 Al presentar el programa,, #7 Si no tienes la; 3/5 flow step(s) without example: #2 Precalificar, #3 Contexto, #4 Nutrir
canal · sync 2026-06-03
261
Digests
67 en últimos 14d
5
Cambios pedidos / pendientes
del checklist
250
Threads
en el canal
2026-06-03
Última sync
del digest

Cosas nuevas (digests recientes)

FechaTemaTags
2026-06-03zero booking links conversion failure patternbooking, conversion, critical-issue, zero-bookings
2026-06-03high intent leads detected require immediate outreachleads, high-intent, outreach, opportunity
2026-06-02opportunities detected high intent leadslead-qualification, opportunity, instagram
2026-06-02daily funnel report 2026 06 02funnel-metrics, leads, booking-conversion, daily-report
2026-06-02high intent leads detected scoring frameworkleads, scoring, qualification
2026-06-01keyword performance metodo avaperu 11pct conversionkeyword-analysis, conversion-rate, content-strategy
2026-06-01high volume week funnel analysisreporting, funnel-metrics, conversion, quality-concern
2026-06-01manual outreach warm pool 30 leads expansionoutreach, warm-pool, content-strategy, manual-action
2026-06-01nahiara cordoba basic incursion 4 5 profilelead-opportunity, qualification
2026-06-01scheduled call anesthesia strategies lead validationcall-confirmed, lead-validation, anesthesia
2026-06-01daily funnel snapshot 70 convos 3 links 1 agendametrics, daily-report
2026-06-01daily funnel snapshot 70 new convos 80 activedaily-report, funnel-metrics
2026-06-01whatsapp agent habla argentino voseovoice-tone, whatsapp, voseo, rioplatense, high-priority, francisco-request
2026-06-01agent quality decline reported whatsapp discriminationbug, quality, discrimination, pending-investigation
2026-06-01weekly performance record 1452 leads 21 confirmedmetrics, performance, weekly-report
2026-06-01daily funnel report jun 01reporting, funnel-metrics, booking-distribution
2026-06-01agent tone dialect correction requestbug-report, agent-behavior, linguistic-issue
2026-06-01fix 031 neutral spanish tone deployment statusdeployment, agent-tone, fix, whatsapp
2026-06-01weekly funnel performance record 21 confirmationsmetrics, conversion, performance, decision
2026-05-31daily funnel report may 31reporting, funnel-metrics, leads

Cambios pedidos por el creador / pendientes

IDEstadoCambio pedido
B5pendienteNo usar jerga rioplatense (voseo, "bacan", etc.) — neutralizar al peruano/latinoamericano. **⚠️ REGRESIÓN 2026-06-01 (ver X29)**: el agente WA volvió a usar voseo (`avisás`, `manej
N3.8pendiente**Bot detection — filtro de auto-respuestas**: si el inbound parece auto-respuesta de otro sistema ("en este momento no estamos atendiendo", "fuera de horario", "déjanos tu consult
X22pendiente**Unsubscribe en Messenger sigue ocurriendo** esporádicamente — plan B automático pendiente (redirect a IG DM). Victor quiere fix de raíz. Francisco confirma que no está 100% resue
X24pendiente~~**Booking proposal actual no convence a Victor**~~ — IMPLEMENTADO 2026-04-29 en v23/v1.4/v1.5 (ver F12, F13). Pendiente validación con synthetics + 24-48h convs reales antes de c
AA13pendiente**Política de tono — amable y neutral SIEMPRE**, independientemente de cómo responda el lead. AVA busca neutralidad para que las respuestas se lean bien en todos los países LATAM.
sentiment del cliente → Ninjo · scoreado ahora (2026-06-03 15:24:32) · modelo claude-sonnet-4-6 · ventana 7d
-1
Sentiment score
frustrado_pero_engachado
Medio
Churn risk
del lado del cliente
5
Cambios pedidos
pendientes en checklist
36
Volumen cliente
agente: 287 convs

Matriz de volumen

enganchado pero en distres
Cliente activo (volumen alto) pero con tono negativo: enganchado, riesgo está en el TONO, no en el silencio.

Sentiment del cliente

📋 El cliente está activo y comprometido (récord histórico de leads, decisiones estratégicas coordinadas con Ninjo), pero hay una queja directa de Francisco sobre el tono argentino del agente y una falla operativa crítica silenciosa: decisiones de booking link del 2026-06-01 nunca se ejecutaron, generando cero conversiones en días con decenas de leads calientes — riesgo de erosión de confianza si el cliente lo detecta. Acción inmediata: (1) confirmar a Francisco que el fix de voseo fue deployado en los 3 mirrors, (2) escalar la falla del booking link flow ante el cliente con diagnóstico claro y fecha de resolución antes de que lo note solo.
SeñalValorEvidencia (cita del cliente)
Enojado con la plataformano
Es queja“el agente de WhatsApp está empezando a hablar como argentino en algunas conversaciones”
Queja constructiva (accionable)“Pide que se corrija para mantener un tono más neutro”

Evidencia citada

SeñalCita textualFecha
Queja directa del cliente (Francisco vía Victor) sobre tono rioplatense del agente WhatsApp“el agente de WhatsApp está empezando a hablar como argentino en algunas conversaciones”2026-06-01
Pedido explícito de corrección — el cliente identificó el problema y comunicó la solución esperada“Francisco quiere tono neutro — no argentino. Es pedido directo del cliente, no de Victor.”2026-06-01
Compromiso activo del cliente con métricas y funnel — señal de enganche positivo“None”2026-06-01

Cambios pedidos por el creador (Métrica 1)

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

IDEstadoCambio pedido
B5pendienteNo usar jerga rioplatense (voseo, "bacan", etc.) — neutralizar al peruano/latinoamericano. **⚠️ REGRESIÓN 2026-06-01 (ver X29)**: el agente WA volvió a usar voseo (`avisás`, `manej
N3.8pendiente**Bot detection — filtro de auto-respuestas**: si el inbound parece auto-respuesta de otro sistema ("en este momento no estamos atendiendo", "fuera de horario", "déjanos tu consult
X22pendiente**Unsubscribe en Messenger sigue ocurriendo** esporádicamente — plan B automático pendiente (redirect a IG DM). Victor quiere fix de raíz. Francisco confirma que no está 100% resue
X24pendiente~~**Booking proposal actual no convence a Victor**~~ — IMPLEMENTADO 2026-04-29 en v23/v1.4/v1.5 (ver F12, F13). Pendiente validación con synthetics + 24-48h convs reales antes de c
AA13pendiente**Política de tono — amable y neutral SIEMPRE**, independientemente de cómo responda el lead. AVA busca neutralidad para que las respuestas se lean bien en todos los países LATAM.