Skip to main content
Compendium · Compositions · Hat Orchestration

Hat Orchestration

Items
16
of 1002 total

Full-featured composed scenes — one per row so each design has room to breathe. Click for the live render, the spec, and the source.

№ 1341

Four-Hat Overview

Scene STA do
Hats, not titles 4 roles · worn by moment

Four hats.
Worn — not held.

A hat is a role you wield at a specific moment. The same person can wear all four; an enterprise can spread them across four sub-teams. The work doesn't change — only who is wearing what.

Intent Architect Hat 1/4
Pillar DO
Responsibility
Owns the spec. Decides what we are building and why before any code is generated.
Worn
When a problem is framed but the artefact is not.
Composition Lead Hat 2/4
Pillar CO
Responsibility
Directs AI labour. Translates spec into prompts, harnesses, and verification gates.
Worn
When the spec is sealed and the build is live.
Flow Steward Hat 3/4
Pillar DE
Responsibility
Keeps convergence honest. Small batches, fast feedback, no merge backlog.
Worn
Between merge and customer reach.
Governor Hat 4/4
Pillar GO
Responsibility
Holds the boundaries. Signs, waives, or kills. Evidence-bound, not relationship-bound.
Worn
At every gate. Especially the ones that look routine.
// The hat is the work. The person is the carrier. Distribution scales — responsibility does not.
Compositions / hat orchestration

The four DoCoDeGo hats — Intent Architect, Composition Lead, Flow Steward, Governor — as a four-card hero strip.

open ↗
№ 1342

Solo Developer Hat-Swap Timeline

Scene STA co
Solo · one carrier 8 hat-swaps in 9h

@author wears all four — in turn.

2026-05-21
Time on hat 8 moments
Intent Architect
2 / 8
Composition Lead
2 / 8
Flow Steward
2 / 8
Governor
2 / 8
  1. 08:30 IA · Intent Architect

    Frame the day’s problem

    → SPEC-204 v0.1 draft

  2. 09:15 GV · Governor

    Self-review the spec, check threats

    → Acceptance criteria sealed

  3. 09:45 CL · Composition Lead

    Prompt the agent, set verification

    → PR #1182 opened, build green

  4. 11:00 FS · Flow Steward

    Smoke test in staging, watch metrics

    → Convergence telemetry +0.4 ICS

  5. 13:30 CL · Composition Lead

    Second loop — negative criteria

    → PR #1183 opened

  6. 15:00 GV · Governor

    External eyes on the PR (peer ping)

    → Two-person rule waiver logged

  7. 16:30 FS · Flow Steward

    Ship to canary, watch drift

    → Customer-reaching in 12 min

  8. 17:45 IA · Intent Architect

    Capture lessons for tomorrow’s spec

    → AAR note appended

// Solo doesn't mean wearing every hat at once. It means knowing which hat you're under right now.
Compositions / hat orchestration

One developer, four hats, one day — a vertical timeline showing which hat is worn at each hour and what it produces.

open ↗
№ 1343

Small Team Hat Distribution

Scene STA co
Team size 3-10 Soft assignments

Every hat has a primary and a secondary.

No one carries a hat alone. No hat goes uncarried. Overlap is the safety net.

Role IACLFSGV
@author-a PRIMARYSECOND·LEARN
@author-b SECONDPRIMARYLEARN·
@author-c LEARNSECONDPRIMARY·
@governor ··SECONDPRIMARY
@gatekeeper LEARN··SECOND
@steward ·LEARNSECONDLEARN
coverage 1P · 1S · 2L 1P · 2S · 1L 1P · 2S · 1L 1P · 1S · 2L
Primary — owns the hat in this moment Secondary — backup, can sub in Learning — apprenticing into the hat
Compositions / hat orchestration

A 6-person small team shares the four hats with deliberate overlap — primary and secondary carriers per hat.

open ↗
№ 1344

Larger Team Hat Specialisation

Scene STA co
Team size 10+ 14 roles · 4 columns

Hats become careers. Duties stay rotational.

DO · spec ownership
Intent Architect
  • @author-pm-a primary
  • @author-pm-b primary
  • @author-staff primary
  • @author-domain primary
depth · 4
CO · agent direction
Composition Lead
  • @tech-lead-a primary
  • @tech-lead-b primary
  • @platform-eng primary
  • @ai-eng primary
depth · 4
DE · convergence
Flow Steward
  • @steward-a primary
  • @release-eng primary
  • @steward-b primary
depth · 3
GO · gates
Governor
  • @governor-prime primary
  • @gatekeeper-sec primary
  • @gatekeeper-rel primary
depth · 3
Weekly rotation — on-call hats 3-week cadence
Week IACLFSGV
Wk 18 @author-pm-a @tech-lead-a @steward-a @governor-prime
Wk 19 @author-pm-b @tech-lead-b @release-eng @gatekeeper-sec
Wk 20 @author-staff @platform-eng @steward-b @gatekeeper-rel
// Specialisation is depth. Rotation is freshness. The team needs both.
Compositions / hat orchestration

A 14-person team specialises — most roles wear one hat dominantly, with rotating governors and a shared review pool.

open ↗
№ 1345

Enterprise Hat Orchestration

Scene STA go
Enterprise · 42 carriers 4 sub-teams · 1 protocol

Hats become org-charts. Arrows become contracts.

Sub-team · IA
Intent Architect
Carriers
8
Lead
@spec-guild-lead
Intake
Customer signal, OKRs
Output
Sealed specs → CL
Sub-team · CL
Composition Lead
Carriers
24
Lead
@platform-eng-lead
Intake
Sealed specs
Output
Built artefacts → FS
Sub-team · FS
Flow Steward
Carriers
6
Lead
@release-eng-lead
Intake
Built artefacts
Output
Customer-reaching → GV
Sub-team · GV
Governor
Carriers
4
Lead
@chief-governor
Intake
Evidence at every gate
Output
Sign / waive / kill → IA
  1. IA CL

    sealed spec + acceptance + threats

  2. CL FS

    green build + verification trace

  3. FS GV

    convergence telemetry + drift

  4. GV IA

    sign / waive / kill + lessons

// Governance ratio · 9.5% of headcount // 42 carriers · 4 sub-teams · 4 handoffs
Compositions / hat orchestration

Four sub-teams, one per hat, connected by named handoff arrows — the enterprise topology for hat distribution.

open ↗
№ 1346

Intent Architect Handbook

Scene STA do
DO · Intent Architect

Spec is the artefact. Everything else is a derivative.

Handbook · v1
Practices
What to do while wearing the hat
  1. 01 Write the spec before the code — every time.
  2. 02 Name the negative criteria (what it must NOT do).
  3. 03 Capture failure modes alongside acceptance.
  4. 04 Link upstream signal — issue, conversation, customer.
  5. 05 Version the spec — v0.1 draft, v1.0 sealed.
Signals
What tells you the hat is working
  1. 01 Spec compiles to ICS ≥ 60 without coaching.
  2. 02 AI agent does not need clarification on first run.
  3. 03 Acceptance criteria translate 1:1 to test cases.
  4. 04 Threats are enumerable, not vague.
  5. 05 Reviewers can quote the spec in their comments.
Gotchas
How the hat fails — watch for these
  1. 01 Writing implementation language ("call /api/x").
  2. 02 Skipping the failure-modes section "to save time".
  3. 03 Approving your own spec without external eyes.
  4. 04 Letting deadline pressure shrink acceptance count.
  5. 05 Treating "negative criteria" as optional.
Mantra · "If it's not in the spec, it doesn't exist." hat retired at sealing
Compositions / hat orchestration

Field handbook for the Intent Architect hat — practices, signals, gotchas, and the day-one rituals that make specs trustworthy.

open ↗
№ 1347

Composition Lead Handbook

Scene STA co
CO · Composition Lead

Humans architect. AI labours. The hat directs.

Handbook · v1
Practices
What to do while wearing the hat
  1. 01 Read the sealed spec verbatim — no paraphrasing.
  2. 02 Pin model + tools + harness; record the trio.
  3. 03 Add verification gates before agent ever runs.
  4. 04 Capture prompt + output diff in the PR.
  5. 05 Re-prompt over editing — escalate only if both fail.
Signals
What tells you the hat is working
  1. 01 First-pass build green on ≥ 70% of specs.
  2. 02 Verification trace is reviewable, not opaque.
  3. 03 Agent edits track 1:1 to acceptance criteria.
  4. 04 Token spend trends down per shipped artefact.
  5. 05 Human edits to AI output stay under 10% lines.
Gotchas
How the hat fails — watch for these
  1. 01 Editing what should have been re-prompted.
  2. 02 "Just fix it in code" — invisibly drifting the spec.
  3. 03 Trusting a green build without verification trace.
  4. 04 Stacking prompts so context becomes incoherent.
  5. 05 Auto-merging when CL and IA hats are the same person.
Mantra · "Direct the agent. Don't become it." hat retired at merge
Compositions / hat orchestration

Field handbook for the Composition Lead hat — prompt practice, verification gates, and gotchas in AI-driven builds.

open ↗
№ 1348

Flow Steward Handbook

Scene STA de
DE · Flow Steward

Convergence is honest only when it's continuous.

Handbook · v1
Practices
What to do while wearing the hat
  1. 01 Ship in batches of one — never bundle to "save".
  2. 02 Watch convergence telemetry, not vanity metrics.
  3. 03 Roll forward when safe, roll back when drift trips.
  4. 04 Keep the canary cohort live and observable.
  5. 05 Surface flow blockers within an hour of detection.
Signals
What tells you the hat is working
  1. 01 Lead time from merge to customer < 30 min p95.
  2. 02 Drift detector trips < 5% of deploys.
  3. 03 No PR waits in queue longer than its build time.
  4. 04 AAR / GTR ratio trending toward green.
  5. 05 Rollback is a routine event, not a crisis.
Gotchas
How the hat fails — watch for these
  1. 01 Letting "release pressure" silence drift signals.
  2. 02 Bundling small batches into a "release train".
  3. 03 Treating canary as ceremony — no real cohort.
  4. 04 Hiding rollbacks to keep dashboards green.
  5. 05 Talking to GV only when something explodes.
Mantra · "Small batches, honest signals, no exceptions." hat retired at customer reach
Compositions / hat orchestration

Field handbook for the Flow Steward hat — small batches, convergence telemetry, and the gotchas of release pressure.

open ↗
№ 1349

Governor Handbook

Scene STA go
GO · Governor

Autonomy is the gift. The gate is the boundary.

Handbook · v1
Practices
What to do while wearing the hat
  1. 01 Cite evidence in every sign / waive / kill.
  2. 02 Time-box waivers — every waiver needs an expiry.
  3. 03 Refuse same-person sign-off — escalate or recuse.
  4. 04 Log decisions to the audit ledger immediately.
  5. 05 Review threats before acceptance, not after.
Signals
What tells you the hat is working
  1. 01 Waiver expiry rate < 5% (most converted, not lapsed).
  2. 02 Audit ledger queryable, no missing decisions.
  3. 03 Reviewers cite spec sections, not personal taste.
  4. 04 Kill events drop over time as upstream tightens.
  5. 05 Same-person sign-offs at zero for the quarter.
Gotchas
How the hat fails — watch for these
  1. 01 Signing because the relationship is warm.
  2. 02 Letting the same person hold CL and GV on a PR.
  3. 03 Vague waivers without an evidence pointer.
  4. 04 Permanent waivers ("temporary" for 9 months).
  5. 05 Allowing post-hoc retroactive sign-offs.
Mantra · "Evidence over relationship. Always." hat present at every gate
Compositions / hat orchestration

Field handbook for the Governor hat — sign, waive, kill — evidence-bound, never relationship-bound.

open ↗
№ 1350

Hat Conflict Mitigation Matrix

Scene STA go
Two-hat collision risk 4x4 · with mitigations

When two hats land on one head — and what to do.

also wearing → IACLFSGV
IA
·
LOW
LOW
HIGH

External validation — never self-judge own spec.

CL
LOW
·
MED

Insert verification trace before steward signs flow.

HIGH

Temporal separation — different day for build vs sign.

FS
LOW
MED

Surface drift telemetry to a second pair of eyes.

·
HIGH

Evidence override — pressure cannot bend the gate.

GV
HIGH

External validation — never self-judge own spec.

HIGH

Temporal separation — different day for build vs sign.

HIGH

Evidence override — pressure cannot bend the gate.

·
Low — pairing is normal, no mitigation needed Med — observable mitigation required High — structural mitigation required
Compositions / hat orchestration

When two hats land on the same person — the 4x4 risk matrix and the mitigations that make solo wearable.

open ↗
№ 1351

Composition Lead × Governor Conflict

Scene STA go
CL × GV · high risk

If you built it, you can't be the one who signs it.

When build-hat and gate-hat are the same person on the same spec, the gate ceases to exist. Temporal separation makes it real again.

Broken · same-spec sign-off verdict · CORRUPT
  1. @author wearing CL
    Prompts agent, opens PR, build green.
    10:47
  2. @author wearing GV
    Reviews own PR, signs it, merges.
    10:48

Failure — the carrier evaluates their own work under one minute. No external evidence. Gate is decorative.

Mitigated · temporal separation verdict · SAFE
  1. @author wearing CL
    Build complete. Hat off. Walks away.
    10:47 · day 1
  2. @peer-reviewer
    External pair of eyes. Comments captured in ledger.
    08:14 · day 2
  3. @author wearing GV
    Re-reads spec cold. Cites peer review. Signs.
    08:42 · day 2
Compositions / hat orchestration

Same person built it AND signed it — the CL×GV conflict, why it corrupts gates, and the temporal-separation mitigation.

open ↗
№ 1352

Intent Architect × Governor Conflict

Scene STA do
IA × GV · high risk

You can't judge your own intent.

The author always thinks the spec is clear. External validation is the only way to find out otherwise.

Broken · self-judgement verdict · CORRUPT
  1. @author wearing IA
    Drafts spec. Captures intent. Lists acceptance.
  2. @author wearing GV
    Reads own draft. Finds it clear. Seals it.

Failure — the author's tacit knowledge fills in the gaps the spec leaves. The agent will not.

Mitigated · external validation verdict · SAFE
  1. @author wearing IA
    Drafts. Submits for external validation.
  2. @peer-validator
    Restates each AC. Flags 2 hidden assumptions.
  3. @author · IA again
    Adds the 2 missing ACs. v1.1.
  4. @governor
    Seals on the validated, revised spec.
Compositions / hat orchestration

You wrote the spec and approve the spec — the IA×GV self-judgement trap and how external validation rescues it.

open ↗
№ 1353

Flow Steward × Governor Under Pressure

Scene STA de
FS × GV · pressure-amplified

The same person, three pressures, three failures.

Evidence-bound governance has to survive the bad day, not the good one.

Routine SAFE
pressure
Scenario
Normal day. Drift trip on a canary cohort.
Decision
Rolls back. Files AAR.
Consequence
No customer reach. Evidence honoured.
Heated COMPROMISED
pressure
Scenario
Launch week. Same drift signal, harder schedule.
Decision
"Watch it for an hour" — sign with verbal waiver.
Consequence
Customers reach a flaky build. Tomorrow’s incident lurks.
Crisis CORRUPT
pressure
Scenario
Board demo at 09:00. Drift trips at 08:42.
Decision
Self-overrides. Says nothing. Ships.
Consequence
Demo passes. Customer learns of failure first.
Mitigation pattern
  1. 01 · Separation of duty — GV cannot be the same person as FS during pressure windows (launch, on-call).
  2. 02 · Evidence override — drift trip auto-blocks merge until a non-FS governor signs.
Compositions / hat orchestration

Delivery pressure vs evidence — when the steward is also the governor, who breaks first? Three release scenarios show the failure curve.

open ↗
№ 1354

Hat Coverage Sprint Heatmap

Scene STA co
Sprint heatmap 10 working days · 4 hats

Who held which hat — across the sprint.

Hat / Day Mon
d1
Tue
d2
Wed
d3
Thu
d4
Fri
d5
Mon
d6
Tue
d7
Wed
d8
Thu
d9
Fri
d10
Intent Arch. author-a author-a author-b author-b author-a author-c author-c author-a author-b author-b
Comp. Lead lead-a lead-a lead-a lead-b lead-b lead-a lead-a lead-b lead-a lead-a
Flow Steward steward-a steward-a steward-b steward-b steward-a steward-b steward-a steward-a steward-b steward-a
Governor governor governor governor gatekeeper governor governor gatekeeper governor governor governor
Concentration alert · @governor carried 8 cells of 40 (20%). rebalance next sprint
Compositions / hat orchestration

A two-week sprint heatmap — 4 hats × 10 working days — coloured by who carried each hat each day.

open ↗
№ 1355

Hat Handoff Protocol Card

Scene STA co
Handoff protocol v1 · 4 handoffs

A hat hands off an artefact, not a feeling.

  1. IA CL
    Payload

    Sealed spec + threat model + acceptance criteria

    Format

    Markdown · YAML frontmatter (id, version, status: sealed)

    Acceptance check

    CL refuses if status ≠ sealed or threat section missing.

  2. CL FS
    Payload

    Built artefact + verification trace + agent prompt log

    Format

    PR with passing CI + linked verification artefact

    Acceptance check

    FS refuses if any acceptance criterion lacks a passing test.

  3. FS GV
    Payload

    Convergence telemetry + drift report + customer-reach signal

    Format

    Dashboard link + 7-day telemetry CSV + canary cohort proof

    Acceptance check

    GV refuses if drift > threshold or canary cohort < spec.

  4. GV IA
    Payload

    Sign / waive / kill + lessons + spec deltas requested

    Format

    Audit ledger entry + signed AAR + delta list

    Acceptance check

    IA refuses next-cycle spec without prior cycle ledger entry.

Contract · the receiver refuses on missing format. No verbal handoffs. audit-recorded
Compositions / hat orchestration

The contract between adjacent hats — what is handed off, in what format, and what the receiver checks before accepting.

open ↗
№ 1356

Hat Accountability Ledger

Scene STA go
Hat Accountability Ledger SPEC-204 · lifecycle
12 entries
Timestamp Hat Carrier Decision Evidence ref
05-18 08:14 IA @author-a
SPEC Drafted v0.1
SPEC-204@a8f1
05-18 11:02 IA @peer-b
SPEC Validated, 2 ACs added
SPEC-204@c2e9
05-18 14:30 GV @governor
SIGN Sealed v1.0
AUD-1001
05-19 09:01 CL @lead-a
BUILD Agent prompt + harness pinned
PR-1182
05-19 11:47 CL @lead-a
BUILD Build green, verification passed
PR-1182@v3
05-19 13:51 GV @gatekeeper
WAIVE Two-person rule waiver · 24h
AUD-1002
05-19 14:08 FS @steward-a
SHIP Canary cohort 5%
REL-447
05-19 14:42 FS @steward-a
KILL Drift threshold · auto-revert
INC-1842
05-19 15:30 GV @governor
SIGN Rollback exception authorised
AUD-1003
05-19 17:45 IA @author-a
SPEC AAR captured, v1.1 delta filed
SPEC-204@d8
05-20 08:42 GV @governor
SIGN v1.1 sealed for re-run
AUD-1004
05-20 14:18 FS @steward-b
SHIP Customer reach · drift normal
REL-448
total 12
IA 3
CL 2
FS 3
GV 4
Compositions / hat orchestration

An auditable ledger of which hat was active at each decision — timestamp, hat, carrier, decision, evidence pointer.

open ↗
Alpha · Honest about it

The framework is real.
The community is forming now.

DoCoDeGo is in Alpha. The framework is documented, the practices are battle-tested at small scale, and the next release is being shaped in public.

If it produces anything, it should produce engineers and teams who think more clearly about what they are building and why.

Two doors in
Join the Discord

Discord is where specs are debated, the framework gets sharper, and decisions land in writing. The conversation is the artefact.