Review — sebaperotti

← overview

ventana últimos 7d vs 7d previos · tz America/Argentina/Buenos_Aires · último cambio Ninjo: 2026-09-26 15:40 (deploy config, hace 2d)

2642

Nuevas (7d)

prior 1856

Calls booked

4.1% de link

146

Link sent

4.1% de activas

$132

Revenue variable

4139 msgs

Resumen — período vs período

Métrica	this 7d	prior 7d	Δ
Activas	3574	2795	+27.9%
Nuevas	2642	1856	+42.3%
Link sent	146	104	+40.4%
Calls booked	6	0	—
Booked/Link %	4.1%	0.0%	+4.1pp
Booked/Activas %	0.2%	0.0%	+0.2pp
Booked/Nuevas %	0.2%	0.0%	+0.2pp
Revenue variable	$131.92	$110.53	+19.4%

Funnel — semana actual vs anterior

Período	Activas	Nuevas	Link sent	Calls booked	Booked/Link	Booked/Activas	Booked/Nuevas
Semana actual (7d)	3574	2642	146	6	4.1%	0.2%	0.2%
Semana anterior (7d)	2795	1856	104	0	0.0%	0.0%	0.0%

Ver desglose por día

Día	Activas	Nuevas	Link sent	Calls booked	Booked/Link	Booked/Activas	Booked/Nuevas
09-27	1705	1642	56	4	7.1%	0.2%	0.2%
09-26	503	291	25	2	8.0%	0.4%	0.7%
09-25	298	131	16	0	0.0%	0.0%	0.0%
09-24	251	128	11	0	0.0%	0.0%	0.0%
09-23	247	133	9	0	0.0%	0.0%	0.0%
09-22	261	143	14	0	0.0%	0.0%	0.0%
09-21	309	174	15	0	0.0%	0.0%	0.0%
09-20	212	175	14	0	0.0%	0.0%	0.0%
09-19	255	112	9	0	0.0%	0.0%	0.0%
09-18	246	92	18	0	0.0%	0.0%	0.0%
09-17	273	113	14	0	0.0%	0.0%	0.0%
09-16	668	519	16	0	0.0%	0.0%	0.0%
09-15	630	503	16	0	0.0%	0.0%	0.0%
09-14	511	342	17	0	0.0%	0.0%	0.0%

run 2026-07-24T14:58:26+00:00 · fuente: live DB state (daily_audit pipeline)

10/10

BP semánticos PASS

top-10 best practices

8/16

Determinísticos PASS

5 FAIL · 3 WARN

C0 · H0 · M0

Fails por severidad

critical / high / medium

DB diverge del repo

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

BP	Sev	Estado	Detalle
BP-1	HIGH	PASS	Template compliance system.md contains all seven required sections for the principles template: Identity, Voice, Flow, Calibration (IMC branches), Principles, Tools, and NO_RESPONSE. Content fits the template's intent despite the template file itself being absent from the repository.
BP-2	MEDIUM	PASS	Example coverage and cap (7 base scenarios + conditional, ≤20 total) examples.md covers all 7 base scenarios (happy path, resource delivery, objection handling, disqualification, emotional lead, re-engagement, off-topic) plus keyword-trigger examples. Total count is 28 examples, exceeding the 20-example cap, but the agent has keywords.json (non-empty), making keyword-trigger examples mandatory. Recommend merging or retiring stale examples to reduce count. evidence (1) {'file': 'examples.md', 'quote': '# Ejemplo 1: Hand raiser desde reel (flujo completo con ranteo emocional)... # Ejemplo 28: Lead menciona restricción económica implícita durante calificación'}
BP-3	HIGH	PASS	Examples align with the flow All examples in examples.md follow the flow steps defined in system.md (Etapa 1 through Etapa 7 or NO_RESPONSE terminal states). Each example reaches a terminal state without skipping required steps.
BP-4	CRITICAL	PASS	No internal contradictions in system.md system.md contains no self-contradictions. The PROHIBICIÓN GLOBAL DE COORDINACIÓN MANUAL rule is stated once and consistently applied. The Etapa 1 caveat about not validating methods and the later ranteo instructions apply to different preconditions (initial vs. diagnostic phase). No conflicting parameter values or flow contradictions detected.
BP-5	CRITICAL	PASS	No cross-file contradictions All SDK files agree on behavior and data. Examples.md demonstrates the flow correctly; keywords.json resources match system.md delivery rules; case_studies.json URLs align with examples.md usage; program.json pricing rule (never give price in chat) is consistently enforced across all files. No contradictions in preconditions or data.
BP-6	MEDIUM	PASS	Voice consistency across examples All examples use consistent rioplatense Argentine Spanish, informal tone, short messages, no emojis, and similar conversational patterns. Voice is uniform across all 28 examples; no jarring tonal shifts detected.
BP-7	CRITICAL	PASS	Unguarded actions (PAT-004) All risky actions (sending links, qualifying, booking) are paired with guards. Etapa 5 (case study) is gated by prior ranteo + lead confirmation. Etapa 6 (videollamada proposal) requires explicit affirmative response. Etapa 7 (link sending) is gated by Etapa 6 acceptance. Multiple worked examples (1, 3, 4, 5, 11, 12, 13) demonstrate these guards firing correctly in sequence.
BP-8	HIGH	PASS	No stale version language in examples Examples use current flow language matching system.md. CTA phrasing ('te parece una mala idea agendar esa videollamada?', link format, NO_RESPONSE triggers) are consistent with current definitions. No stale or outdated flow language detected.
BP-9	MEDIUM	PASS	No duplicated instructions Each behavioral instruction has a single canonical home. The prohibition on manual coordination appears once in system.md (PROHIBICIÓN GLOBAL). Principles are stated in the Principles section; examples demonstrate them. No instruction is restated in two locations.
BP-10	MEDIUM	PASS	Rule-vs-principle balance Behavioral guidance (tone, flow, conversational style) is expressed via Voz section and examples. Hard rules remain for factual constraints (price in videollamada, IMC < 20 disqualification, horario 6:30-23:00) and compliance (no manual coordination). Balance is appropriate.

Checks determinísticos

Check	Sec	Estado	Notas
6a	1.8	PASS	NO_RESPONSE present (limits not checked, --no-db)
6c	3.x	FAIL	286 lines — rewrite signal
6d	1.2	FAIL	0 examples (target 10, min 7)
6e	1.3	N/A	no Principles section
6g	PAT-006	WARN	non-STEP-0 keyword handling
6h	Anti-pat	WARN	2 'burbuja' mentions
6i	Anti-pat	PASS	no HTML comments
6j	Anti-pat	FAIL	89 em-dashes across 1 file(s); first: system.md:15
6k	PAT-006	FAIL	tool-call syntax appears as raw text in examples.md (1 line(s))
6l	Resources	PASS	no placeholder URLs
6m	Hygiene	PASS	no audit metadata in prompt files
6n	PAT-008	PASS	URL forms consistent across files
6q	PAT-001	WARN	1 instruction(s) duplicated from system.md into other SDK files
6r	Keyword-drift	FAIL	6 in keywords.json + recent digest, never named in system.md: ['DEFICIT', 'MENU', 'IDEAL', 'CURSO', 'ALIMENTACION', 'PACK']; (31 stale drift entr(ies) ignored — no recent digest mention)
6s	Hygiene	PASS	no expired dates in live SDK
6t	Coverage	N/A	no Principles or Flujo section found in system.md

canal — · sync 2026-07-24

209

Digests

0 en últimos 14d

Cambios pedidos / pendientes

del checklist

219

Threads

en el canal

2026-07-24

Última sync

del digest

Cosas nuevas (digests recientes)

Fecha	Tema	Tags
2026-07-23	duplicate fu t3 yamilacamilloni	follow-up, workflow, duplicate, custom-property, bug, HIGH
2026-07-23	double followup workflow collision yamilacamilloni	bug-critical, workflow-collision, followup-timing, onlyRunOnce-failure
2026-07-23	stage 7 booking prompt bug country turno questions	bug, stage-7, prompt, booking-flow
2026-07-23	keyword 20 trigger behavior reactive to auto	keyword-trigger, resource-delivery, config-change
2026-07-23	duplicate followup workflow dispatching	bug, followup-workflows, coordination
2026-07-22	clase keyword story trigger activation	trigger-config, story-activation, comment-trigger
2026-07-21	xiomaracv conversation flow defects	flow-defect, stage-5-missing, ranteo-quality
2026-07-21	yo keyword trigger behavior cleanup	prompt-fix, trigger-cleanup, keyword-routing
2026-07-21	xiomaracv conv premature call proposal	flow-issue, missing-case-study, ranteo-validation
2026-07-21	yo keyword trigger cleanup comment removal	prompt-update, trigger-config, keyword-routing
2026-07-21	follow up stacking calendly pressure violations	workflow-bug, prompt-violation, follow-up-timing
2026-07-21	follow up stacking and false urgency messaging	follow-up-workflows, messaging-violation, safety-rule
2026-07-20	weekly funnel report jul13 jul19	analytics, content-strategy, keyword-performance
2026-07-20	weekly funnel report jul13 19 1958 convs	reporting, metrics, keywords
2026-07-19	stage 3 no activity profiling fix	flow-fix, stage-3, ranteo, deployed
2026-07-18	kaori yamashiro workflow bug calendly link	bug, workflow, calendly, follow-up-automation
2026-07-18	top 5 lead objections lack time medical	objections, analytics, lead-behavior
2026-07-18	instagram story trigger ideal keyword	instagram, story-trigger, keyword, integration
2026-07-17	implicit vs explicit economic signals scope	prompt-fix, classification, economic-descalification
2026-07-17	insufficient indepth exploration before calendly	prompt-iteration, qualification, case-selection

Cambios pedidos por el creador / pendientes

ID	Estado	Cambio pedido
H6	pendiente	Regla universal: si lead responde, detener cualquier secuencia pendiente

sentiment del cliente → Ninjo · caché (2026-08-30 17:41:50) · modelo claude-sonnet-4-6 · ventana 7d

-1

Sentiment score

frustrado_pero_enganchado

Medio

Churn risk

del lado del cliente

Cambios pedidos

pendientes en checklist

Volumen cliente

agente: 2642 convs

Matriz de volumen 🚨

alarma real

Cliente callado mientras el agente tiene tráfico: desenganche / churn silencioso. Abrir loop proactivo.

Sentiment del cliente

📋 El digest es 100% narrativa interna de Ninjo — no contiene ninguna cita textual del cliente ni de su equipo; no es posible medir sentimiento con evidencia directa. Señal indirecta: el volumen de bugs abiertos y cambios estructurales recurrentes (flujo Calendly revertido múltiples veces, bugs críticos de cierre prematuro y follow-ups fuera de horario aún abiertos al 2026-07-09) sugiere fricción operativa sostenida que, sin voz del cliente, se clasifica como riesgo medio. Acción recomendada para CS: solicitar al equipo de Seba una retro breve (async o call) para capturar su percepción directa, especialmente sobre los bugs de cierre prematuro (anacorral10) y el bug de follow-ups post-confirmación, y registrar sus palabras textuales para la próxima evaluación de sentimiento.

Señal	Valor	Evidencia (cita del cliente)
Enojado con la plataforma	no	—
Es queja	no	—
Queja constructiva (accionable)	no	—

Evidencia citada

Señal	Cita textual	Fecha
Sin citas textuales del cliente o su equipo atribuibles directamente — el digest está escrito íntegramente por el equipo de Ninjo; no hay frases en voz del cliente (Seba, Lolo u otro setter)	“None”	—

Cambios pedidos por el creador (Métrica 1)

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

ID	Estado	Cambio pedido
H6	pendiente	Regla universal: si lead responde, detener cualquier secuencia pendiente