bokken · design thinking run report

iterate

Lanzadera is an on-demand commuter shuttle for Spanish suburbs whose core covenant is a +/-6-minute pickup window.

session gallery · mode dojo · status complete

evidence items (16 synthetic, labeled)

options generated with full lineage

4 supported · 2 contradicted · 4 untested

$10.36

63 model calls, illustrative — charged $0.00

executive summary

What this run concluded

Every number on this page is derived from the append-only Journal — nothing was written by hand.

Daily pass commuters (46%→38% of riders) plan their mornings around a ±6-minute promise the product no longer keeps (71% compliance) nor admits to - the trust gap, not the delay itself, is driving the 345/month churn

problem statement · define

'Ventana honesta': at 21:00, with the route already optimized, push the real pickup point and an honest window for tomorrow - including 'wide window (12 min) on your street tomorrow' - with a one-tap switch to an earlier van when the window is wide [serves: O0, O1, O3]

concept advanced · ideate

The verdict rests on simulated evidence and requires validation with real users before it is acted on.

how this run worked

The sequence, the cast, the machinery

Bokken executed the Design Thinking loop autonomously. Every actor below is journaled; persona contributions are simulated and labeled as such.

Empathize01

13 evidence items
4 features UI-tested
7 outcomes Ulwick-ranked

26 model calls

Define02

41 interpretations
problem statement selected
losers preserved with reasons

3 model calls

Ideate03

9 options, full lineage
3 firewalled lenses vote
skeptic on record

16 model calls

Prototype04

web research on the concept (sourced)
10 assumptions registered
2 artifacts hashed

6 model calls

Test05

fresh firewalled panel
4/2/4 sup/con/untested
verdict: iterate

11 model calls

Complete06

dossier (A/B/C)
handoff specs
this report (pptx + html)

deterministic

Interview panel (6 personas)

skepticskeptic-agent

cross-segment

feasibilityfeasibility-agent

cross-segment

viabilityviability-agent

cross-segment

segmentTomas (64, Badajoz)

daily pass commuters

segmentTomas (60, Zaragoza)

flexible pay-per-ride commuters

segmentJavier (44, Zaragoza)

daily pass commuters

Ideation panel (6 personas)

skepticskeptic-agent

cross-segment

feasibilityfeasibility-agent

cross-segment

viabilityviability-agent

cross-segment

segmentMarta (72, Zaragoza)

daily pass commuters

segmentPablo (48, Zaragoza)

flexible pay-per-ride commuters

segmentJavier (24, Murcia)

daily pass commuters

Test panel (6 personas — firewalled from the other panels)

skepticskeptic-agent

cross-segment

feasibilityfeasibility-agent

cross-segment

viabilityviability-agent

cross-segment

segmentCarmen (28, Zaragoza)

daily pass commuters

segmentRamon (36, Madrid)

flexible pay-per-ride commuters

segmentCarmen (52, Zaragoza)

daily pass commuters

System agents and models

Agent	Did	Model calls
`facilitator`	ran the stage machinery: programs, clustering, selection, register, fidelity, verdicts
`ui-walker`	walked the running app with a real browser; journaled observed facts and screenshots
`concept-researcher`	authorized deep web research on the selected concept, sources cited
`convergence lenses`	adversarial feasibility vs the codebase, independent RICE, outcome desirability
`skeptic`	mandatory on-record challenge before convergence closed
`claude-fable-5`	routing classes served	33
`claude-haiku-4-5`	routing classes served	9
`claude-opus-5`	routing classes served	12
`claude-sonnet-5`	routing classes served	9

inputs

What the run was grounded in

Brief

Segment: daily pass commuters
Segment: flexible pay-per-ride commuters
Constraint: the nightly route optimization stays (it is the unit economics)
Constraint: Spanish-first UI, plain language
Constraint: no new hardware or driver-side changes

Tangible corpus

code · /Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/repo
running app · file:///Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/app/index.html
metrics · /Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/kpis.csv
discussion · /Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/interview_marta.md
discussion · /Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/interview_diego.md

stage · empathize

Evidence, on the record

28 evidence items9 observed7 ranked outcomes16 research calls0 abstentions

Agents & activity

Javier (44, Zaragoza), Tomas (60, Zaragoza), Tomas (64, Badajoz), ui-tester, ui-walker, facilitator

1 cognition, 16 research, 9 sidekick model call(s) on claude-fable-5, claude-opus-5, claude-sonnet-5. Kata moves: stage_contract.

Process

Corpus-calibrated interview program per segment; grounded persona interviews with citation-validated answers or honest abstention; functional UI walkthrough of the running app; JTBD desired-outcome derivation and per-persona Importance/Satisfaction scoring into a deterministic Ulwick opportunity ranking.

Output

28 evidence items, 0 abstentions, 7 ranked outcomes, UI walkthrough done.

The window is the whole product for me. The README calls plus-minus six minutes 'the product's core covenant' - and my own months say the covenant is slipping: compliance fell from 87% to 71% between March and August while delay tickets quadrupled from 96 to 402. I plan childcare around that number. When it stretches to fifteen real minutes, I miss the 7:40 and the next seat is 9:05.

Tomas (64, Badajoz)simulated · 1 citation(s)

What makes me angry is not the delay - it is the green tick. The app says 'ventana cumplida' even when the van was nine minutes late, because it counts from the dawn re-plan, not from what it promised me the night before. I can plan around honesty. I cannot plan around a green tick that lies.

Javier (44, Zaragoza)simulated · 1 citation(s)

The 400-metre pickup rule reads reasonable in the README, but my corner moved behind the market one Monday after a 'route optimization' and nobody told me. Six minutes walking, uphill, on top of a 3.20-euro ride, while the Cercanias costs 1.85. Three days out of five the train now wins.

Tomas (64, Badajoz)simulated · 1 citation(s)

The numbers say riders like me are already voting: average rides per rider fell from 31.2 to 23.1 in six months and the monthly-pass share slid from 46% to 38%. The app still pushes the 89-euro pass at me like I commute ten trips a week. Charge me for what I use and I stay.

Javier (44, Zaragoza)simulated · 1 citation(s)

Booking changes close at 21:00 the night before - that is exactly when the app should tell me tomorrow's truth: my pickup point and an honest window. If tomorrow will be wide, say so and let me take the 7:20 van instead. The information exists; the route is re-optimized nightly.

Tomas (60, Zaragoza)simulated · 1 citation(s)

Screen: Lanzadera — tu van de mañana (file:///Users/maglionejuanmartin/code/bokken/src/bokken/demo/fixtures/app/index.html) Loaded in 512 ms. Headings: Lanzadera; Tu van de mañana; Reservar asiento; Ajustes. Primary actions visible: Mañana; Reservar; Ajustes; Cambiar a la van de las 07:20; Confirmar asiento. Forms: 0. Accessibility: 1 input(s) without a label, 0 image(s) without alt text. Console errors: none.

ui_walkthroughobserved

Opportunity ranking (Ulwick: Opp = Importance + max(Importance − Satisfaction, 0))

Opportunity score per outcome (≥15 severely underserved, 12–15 underserved)

O3: Increase trust that the app's on-time indicator reflects the promise made to the rider, not the re-planned route - opportunity 16.0 (severely underserved)

O0: Increase the likelihood that the pickup window promised at 21:00 is the window that actually happens the next morning - opportunity 14.0 (underserved)

O1: Minimize the surprise of a pickup point that moved overnight - the rider knows the point and the walk before going to bed - opportunity 13.0 (underserved)

O2: Minimize the cost of a missed pickup - a same-morning recovery option instead of a 90-minute gap - opportunity 11.3 (moderate)

O5: Increase the perceived fairness of route optimization - riders understand why their pickup changed - opportunity 10.3 (moderate)

O6: Minimize the door-to-van walk variance across weeks - opportunity 9.7 (served)

O4: Minimize the money a 2-3 day rider wastes versus pay-per-use reality - opportunity 8.3 (served)

observed · functional ui review

The product, exercised first-hand

A browser walkthrough of the running app — screenshots and facts are observed evidence, not simulation.

4 features exercised1 broken2 works1 unclear

Cambiar a una van más puntualworks

The switch responds instantly and confirms inline ('mañana te recoge la van de las 07:20') without leaving the screen - exactly the one-tap recovery the wide-window warning needs.

click [3] -> ok

Reservar un asientoworks

Day and van selectors plus one confirm tap; the confirmation promises the 21:00 notice, which is the right moment to promise.

click [1] -> ok

click [6] -> ok

Indicador de puntualidad de ayerbroken

The card shows a green 'Ventana cumplida' tick directly above a notice that the arrival alert went out 9 minutes after the promised window - the indicator measures the dawn re-plan, not the 21:00 promise. This is the exact 'green tick that lies' from the Marta interview, reproduced in the product's own UI.

click [0] -> ok

Aviso de las 21:00 (ajustes)unclear

The opt-in checkbox exists, but its state does not visibly persist when switching views, so whether the preference sticks cannot be confirmed from the UI alone.

click [2] -> ok

Full heuristic review (verbatim)

**Coverage:** the bundled Lanzadera mock (3 views: Mañana / Reservar / Ajustes), walked in a real browser at desktop and mobile viewports; 4 features functionally tested.

## What works at first contact
- Tomorrow's card leads with the two facts riders plan around - the window (07:38-07:46) and the walk (3 min) - and the one-tap van switch confirms inline without a page change.
- Booking is three taps end to end, and the confirmation promises the 21:00 notice: the product already knows its moment of truth.

## Findings
1. **Ayer card -> green 'Ventana cumplida' beside a 9-minutes-late notice -> this is the trust-breaking contradiction driving churn (Marta: 'a green tick that lies') -> measure the tick against the 21:00 promise, never the dawn re-plan.** The fine print admits the re-plan baseline; honesty is one comparison swap away.
2. Ajustes -> the 21:00 opt-in checkbox does not visibly persist across views -> riders who opt in may silently stay unnotified -> persist and echo the state ('te avisaremos a las 21:00').
3. Ajustes -> the 89-euro pass banner shows regardless of riding pattern -> 2-3 day riders read it as mispricing (Diego) -> gate the banner on trips/week.

Verdicts: 2 works · 1 broken · 1 unclear. The broken finding is the same insight the interviews surfaced - the UI reproduces the dishonest indicator faithfully.

stage · define

The problem statement, and why the losers lost

Agents & activity

facilitator

3 cognition model call(s) on claude-opus-5. Kata moves: hmw_reframe, stage_contract.

Process

Evidence clustered into insights tied to underserved outcomes; point-of-view candidates drafted and reframed (HMW) when solution-shaped; winner selected on evidence + opportunity coverage with losers preserved.

Output

One problem statement selected; losers preserved with reasons; opportunity coverage among the criteria.

selected

How might we give riders an honest picture of tomorrow's commute before the booking deadline?

Lost: solution-shaped; reframed and folded into the winner's opportunity space

Flexible riders lack a fair price, pushing them to rail on marginal days

Lost: real but narrower: pricing pain affects 2-3 day riders; the trust gap affects every segment and compounds monthly

The Kata reframed solution-shaped statements 1 time(s) before selection.

stage · ideate

9 options in, one concept out

Agents & activity

skeptic-agent, facilitator

4 challenge, 3 cognition, 9 extraction model call(s) on claude-fable-5, claude-haiku-4-5, claude-opus-5. Kata moves: stage_contract, timebox_pivot.

Process

Quota-driven work-alone divergence tied to outcome IDs with novelty monitoring; skeptic challenge on record; convergence through three firewalled lenses - adversarial feasibility vs the codebase (green/amber/red + first honest slice), independent RICE, outcome desirability.

Output

9 options with full lineage; one concept advanced; lens verdicts and dissent on record.

advanced to prototype

Dissent on record (feasibility)

stage · prototype

Artifacts against the riskiest assumptions

Agents & activity

concept-researcher, ui-walker, facilitator

2 cognition, 2 generation, 2 research model call(s) on claude-fable-5, claude-opus-5. Kata moves: stage_contract.

Process

Assumption register built and risk-classified (impact x uncertainty); cheapest artifact set chosen against the riskiest assumption; artifacts generated and hash-journaled with assumption linkage.

Output

10 assumptions registered; 2 artifacts generated and hash-journaled.

Fidelity decision — why these artifacts

The riskiest cluster is behavioral: will a warned rider switch vans instead of churning (assumptions 1 and 3)? A wireframe of the 21:00 push on the product's own design tokens makes the moment concrete enough to test the switch behavior; landing copy tests whether 'honesty as a feature' is a message pass holders buy. Both are days, not weeks.

Artifact	Kind	sha256	Assumptions
artifacts/prototype/wireframe_html.html	wireframe_html	`7842679043e0`	3
artifacts/prototype/landing_copy.md	landing_copy	`e3ffb3898c75`	2

reported · concept research

The web, on the record

Authorized deep research on the selected concept. Every signal carries its source; findings are reported evidence, not observation.

Competitors and prior art

Who	What	Overlap
BusUp (demo data)	B2B commuter shuttles with contractual SLAs	partial - no consumer-facing window promise
Renfe Cercanias (demo data)	publishes real-time punctuality	the substitute riders defect to at 1.85 EUR

Market signals

DEMO: 62% of shuttle churners cite reliability communication, not reliability itself

https://demo.example/mobility-surveyreported

Differentiation risks

DEMO: honesty-forward messaging is easily copied; the moat is the nightly engine commitment, not the copy

Open questions

Does a warned rider switch vans or cancel?

stage · test

The assumption register, scored

Agents & activity

Carmen (28, Zaragoza), Carmen (52, Zaragoza), Ramon (36, Madrid), facilitator

11 challenge model call(s) on claude-fable-5. Kata moves: loopback_proposal, stage_contract.

Process

Fresh firewalled panel evaluates the prototype against every register entry; quantified kill/iterate/proceed recommendation with loop-back proposal on contradiction.

Output

4 supported, 2 contradicted, 4 untested; recommendation: iterate.

Register outcome

An honest 'wide window' warning reduces anger more than it reduces bookings

SUPPORTED

The route engine can commit a next-morning window at 21:00 within +/-2 min accuracy

SUPPORTED

Riders on wide-window days will switch vans rather than churn

CONTRADICTED

Riders will read a 21:00 notification the night before commuting

SUPPORTED

Marta's segment (daily pass holders) values honesty over raw punctuality

UNTESTED

Publishing real compliance numbers will not be weaponized by competitors

SUPPORTED

Support ticket volume drops when windows are honest (402/mo baseline)

UNTESTED

Diego's segment will not abuse a free missed-pickup rescue

UNTESTED

The 21:00 booking-change deadline is the right moment for the truth push

CONTRADICTED

The 'promise vs actual' tick can be computed from data already stored

UNTESTED

how the agents argued

Votes, challenge, dissent, iteration

6 lens votes8 kata moves executed0 suppressed1 dissents on record

Convergence lens votes

feasibility — 2 vote(s)

feasibility
verdict green, effort S, first slice: push tomorrow's point+window at 21:00, read-only (no van switching yet) - The route engine already commits a plan nightly; surfacing it at 21:00 is exposure, not invention. The README confirms changes close at 21:00 - the data and the moment already align.

feasibility
verdict red, effort L - A rider-voted pickup map fights the optimizer head-on; every vote becomes a constraint the engine must not break. Not honestly buildable as scoped.

viability — 2 vote(s)

viability
RICE 4.2: reach 9 (every active rider gets the 21:00 push), impact 3 (attacks the top-ranked outcome, opp 16.0), confidence 0.7, effort 4.5 pw

viability
RICE 0.6: monthly governance overhead for a fairness perception the receipts idea buys cheaper

desirability — 2 vote(s)

desirability
Directly serves the 16.0 outcome and Marta's exact words: 'I can plan around honesty'

desirability
Council governance is nobody's morning problem

The skeptic, verbatim

The weakest claim on the table is that honesty reduces churn: the record proves riders are angry about the lying tick, but no evidence shows a warned rider stays. A 'wide window' push at 21:00 might simply move the churn decision to bedtime. The cheapest test: measure switch-vs-cancel behavior on the one-tap van change before believing the retention story.

skeptic · on the record before convergence closed

Dissent preserved on the decision

red verdict on 4e51e915d49843be82430d6bb9293a77: verdict red, effort L - A rider-voted pickup map fights the optimizer head-on; every vote becomes a constraint the engine must not break. Not honestly buildable as scoped.

feasibility

Facilitation (Kata) — every intervention journaled

Move	Stage	Fired	Trigger / note
`stage_contract`	empathize	yes	Opening empathize. Goal: understand the people in the problem space. Method: structured interviews with ladder
`stage_contract`	define	yes	Opening define. Goal: frame the problem worth solving. Method: cluster evidence into insights, reframe, select
`hmw_reframe`	define	yes	The current problem statement embeds a solution: "Build a night-before notification with tomorrow's window". R
`stage_contract`	ideate	yes	Opening ideate. Goal: generate genuinely different options, then converge deliberately. Method: quota-driven d
`timebox_pivot`	ideate	yes	Idea novelty has decayed (17% new against a floor of 20%). Proposing we move from divergence to convergence.
`stage_contract`	prototype	yes	Opening prototype. Goal: build the cheapest artifact that tests the riskiest assumption. Method: register assu
`stage_contract`	test	yes	Opening test. Goal: score the assumption register against honest reactions. Method: fresh-panel evaluation (Do
`loopback_proposal`	test	yes	A test result contradicts earlier work: the 'honesty reduces churn' assumption is contradicted by the same evi

continuous discovery

The Opportunity Solution Tree

Teresa Torres's discovery structure, read straight from the journal: outcome → opportunities (Ulwick-ranked) → the solution that advanced → its assumption tests.

Desired outcome (problem framed)

O3: Increase trust that the app's on-time indicator reflects the promise made to the rider, not the re-planned route - opportunity 16.0 (severely underserved)

Solution advanced: 'Ventana honesta': at 21:00, with the route already optimized, push the real pickup point and an honest window for tomorrow - including 'wide window (12 min) on your street tomorrow' - with a one-tap switch to an earlier van when the window is wide [serves: O0

supported An honest 'wide window' warning reduces anger more than it reduces bookings

supported The route engine can commit a next-morning window at 21:00 within +/-2 min accuracy

contradicted Riders on wide-window days will switch vans rather than churn

supported Riders will read a 21:00 notification the night before commuting

untested Marta's segment (daily pass holders) values honesty over raw punctuality

supported Publishing real compliance numbers will not be weaponized by competitors