System trace

Vacancy → outreach brief

One signal. Five layers. Every step is a versioned artefact in git. Follow a single job posting through the full Next Move Engine pipeline — from raw file to a ready-to-send message with stop conditions.

inputs/market-state/actions/ 2 model skills · 1 human decision

Scenario

Input: a Head of Engineering vacancy at Ve##ra — a fictional B2B fintech (~80 engineers, Series B, Q3 regulatory milestone).
Output: a typed outreach action: who, channel, exact message angle, CTA, KPI, stop condition.
Names and companies are synthetic. The file structure, skill contracts, and decision rules are real.

1

Layer 1 — Input

File arrives. No processing yet.

A vacancy is dropped into inputs/vacancies/. This is the boundary of the system — raw text, no classification.

# inputs/vacancies/ve##ra-head-of-engineering.md

Ve##ra — Head of Engineering

We are a B2B payment infrastructure company (~80 engineers, Series B).
CTO K**er B**vik is looking for a Head of Engineering to own delivery
across three product squads and one platform team.

Requirements: 5+ years engineering leadership, experience scaling orgs
past 50 engineers, strong on process and technical standards.

Context: Q3 2026 regulatory certification milestone.
Delivery pace and squad autonomy are top of mind.
Skill invoked: none — file waits for the next run of vacancy-intel
2

Layer 2 — Signal

vacancy-intel classifies the input.

First output: a signal card in market-state/signals/. Type and strength assigned against a fixed vocabulary — no free-form inference.

# market-state/signals/ve##ra-leadership-gap-2026-03.md

---
id: S-41
company: Ve##ra
signal_type: leadership-gap
strength: strong
source: inputs/vacancies/ve##ra-head-of-engineering.md
date: 2026-03-14
linked_person: K**er B**vik
---

CTO posting a Head of Engineering role = explicit delegation play.
Strength: strong — public source + structural indicator + named deadline.
One signal. Not yet a pattern.
Decision rule:  strong requires: public source + structural indicator + time pressure. All three present here. A vacancy without a named deadline would be moderate.
Skill invoked: vacancy-intel — pass 1 of 2
3

Layer 3 — Intelligence

vacancy-intel produces the deep profile.

Second output: a structured intel report in market-state/intels/. Company profile, role analysis, hidden agenda, two entry-point lenses.

# market-state/intels/ve##ra-head-of-engineering.md

---
skill: vacancy-intel
date: 2026-03-14
company: Ve##ra
source: inputs/vacancies/ve##ra-head-of-engineering.md
signals: [S-41]
---

## Company profile
Series B fintech, payment infrastructure, ~80 engineers.
Regulatory certification Q3 2026 — delivery predictability is existential.
Growth stage: scaling from startup to structured engineering org.

## Role analysis
HoE under existing CTO = delegation play, not co-founder hire.
CTO wants delivery ownership off their plate before the deadline.
Squad structure already decided (3 product + 1 platform).

## Hidden agenda
CTO is measured on the Q3 milestone.
Hiring for delivery acceleration — 60–90 day decision window.

## Entry point: consulting lens
High fit: engineering leadership setup + delivery acceleration.
Angle: delivery ownership at scale, not a recruitment pitch.

## Entry point: hiring lens
Fit: strong on process and delivery. Risk: HoE under CTO = limited scope.
Probe: squad autonomy level, tech debt situation, board pressure.
Constraint:  every sentence in the intel file must trace back to a line in the source. No speculation beyond what the public input contains.
Skill invoked: vacancy-intel — pass 2 of 2. Skill complete.
4

Layer 4 — Hypothesis

market-analyst synthesizes across the full state.

Runs periodically. Reads new signals and intel, cross-references the full market-state/. Output: a dated, confidence-scored hypothesis.

# market-state/hypotheses/ve##ra-delivery-pressure.md

---
id: H-22
level: micro
confidence: high
date: 2026-03-14
companies: [Ve##ra]
signals: [S-41]
intel: market-state/intels/ve##ra-head-of-engineering.md
expires: 2026-06-30
---

Ve##ra CTO is under delivery pressure ahead of Q3 regulatory deadline.
Delegating engineering execution now — looking for someone to own delivery.

This is a narrow, time-bound window: 60–90 days.
Receptive to delivery-acceleration conversation.
Not receptive to a generic EM pitch.

Confidence: high
Why: structural signal (public vacancy) + named deadline + delegation pattern.
Decision rule:  high confidence requires two independent signal sources or one structural + named deadline. Both are present. A low-confidence hypothesis produces a PI-* positioning item — not an outreach action.
Skill invoked: market-analyst
5

Layer 5 — Action

action-planner drafts the outreach brief.

Reads the high-confidence hypothesis. Produces a typed AI-* action: concrete, executable, self-contained.

# market-state/actions/AI-28.md

---
id: AI-28
kind: outreach
hypothesis: ve##ra-delivery-pressure
created: 2026-03-14
who: K**er B**vik, CTO, Ve##ra
channel: LinkedIn DM
status: not started
dependency: none
---

Angle: noticed the HoE opening — curious what the current bottleneck looks
like at 80 engineers when delivery ownership is being split out.

Message draft:
  "K**er, saw you're building out delivery ownership at Ve##ra.
   I've set up similar structures at 50–200 engineer orgs — the hard part
   is usually not the org chart but the squad contract model.
   What's your biggest constraint right now — pace, autonomy, or something else?"

CTA: one open question. No CV. No services list. No call request.
KPI: any reply within 7 days.
Stop: silence → log in market-state/people/b**vik.md, no retry for 30 days.
Human-in-the-loop:  the action file is a draft. The human reads it, edits the message, and sends — or drops it. The system writes the brief. The human keeps the send button.
Skill invoked: action-planner — pipeline complete.

What this trace produced

Signal

S-41

leadership-gap · strong

Intel

ve##ra-hoe.md

company + role + entry points

Hypothesis

H-22

micro · confidence: high

Action

AI-28

outreach draft · ready to send

2 model skills. 1 human decision. Every artefact versioned in git. No guessing.

Architecture decisions

Why files, not API calls?

Each layer writes a file to market-state/. The next layer reads files. No direct skill-to-skill calls. No shared memory.

Every state transition is versioned in git. You can reconstruct what the system knew on any given day.

Why typed actions?

An AI-* action has fixed fields: who, channel, angle, CTA, KPI, stop condition. A generic "reach out to X" has none of those.

Typed actions are immediately executable. They also enable dependency graphs: AI-9 unlocks AI-10 only after AI-9 is logged.

Why confidence threshold?

Low-confidence hypotheses produce PI-* positioning items — fix the narrative, write content. Not outreach.

Outreach on weak signal is noise. The system treats your attention as expensive.

Want to talk systems?

I can build this for your business.

The same approach — signals, intelligence, hypotheses, typed actions — works for any domain where the input is public data and the output needs to be a concrete next move.