Findings from the planning lab on branch claude/planning-lab. It prototypes the planning hand's card mechanics as data, plays them by hand, and simulates them with bots against four lab contracts. This is a prototype, not a PR.
How to play it: scripts/microslice.sh planning-table, then serve output/microslices/. On the setup screen, pick the contract, runner, scars, trait, preparation and Rules (the ruleset variants below). The headless tools are cargo run --release -p protokill_microslice_planning_table --example play -- office a 7 scars=3 rules=refined hint and --example sim -- seeds=100 rules=refined.
Seed 7, office retrieval, Runner A, stock rules. Leave the pry bar at home and take a favour, peek the street, and swap Coffee for a stim pack. Success goes from 47% to 55%.
What the lab is #
- Cards are data. 89 records: 7 runner kit cards, 6 scars, 24 job entries (8 of them two-faced), 27 street cards, 12 planning traits from #255, and rulesets. Each card is site requirements, generic effects and terms. Its rules line is generated from its effects. No card has Rust of its own.
- The effect vocabulary:
stage ±n,security ±n,route ±n,open,close,fit,tag,start,prep,objective(side job),condition,pay,trigger(a mid-mission complication),flip(a gamble),reroll,tilt,scar_risk,notoriety. - A mission sketch scores a plan: three stages (in, job, out), each served by routes that need tags (badge, climb, deck, pry), with integer risk. Lab tooling only. Spare cash buys preparation, and costs must be payable from cash or advances.
- Bots: greedy weighs each move by expected value over the sources' public odds; oracle searches three moves deep with full sight of the outcomes (the ceiling).
Mechanic families (one example each) #
| Family | Effects | Example | Verdict |
|---|---|---|---|
| Route | open, close, route ±n, tag, start, fit | Vent access: Vent −2. Its other face, Sealed vents: closes Vent | the keeper. The only cards that depend on the job, and the ones that tell its story |
| Pressure | stage ±n, security ±n | Skeleton crew: all −1 | readable, but the same on every job. Keep few |
| Hook | objective, condition, trigger | Fence's tip: side job 16 at risk 5 · Out +1 | right idea, under-tuned (below) |
| Gamble | flip, face-down entries | Inside contact: 80%: opens Staff · In −1, else all +2 | good tension at a glance |
| Deal | terms: cost, advance, cut, favour, time, scar risk | Loan shark: advance 14 @ 60% | a price every card can carry, not a family of its own |
| Hand rule | slots, floor, move costs and limits, source weights, tilt by tag | Improviser: swaps are free; you can't peek | traits only |
Measurements #
1,600 hands per ruleset (stock, refined and scouted). Each is 100 seeds × 4 runners × 4 contracts. Tight and scarce ran at 960 hands. Half the hands are scarless; the rest carry a light and a severe scar. Values are in lab cash.
| Ruleset | Dealt | Greedy | Oracle | Moves | Peek share | Close call | No job card | Grew past deal | Owes after | Default | Success |
|---|---|---|---|---|---|---|---|---|---|---|---|
| stock | 26.7 | 40.5 | 45.5 | 4.7 | 6% | 54% | 51% | 21% | 39% | 2% | 61% |
| tight: one push, one-shot holds, swaps 10m | 26.9 | 37.3 | 44.0 | 4.0 | 3% | 53% | – | – | 35% | 3% | 56% |
| scarce: tight, 1 swap per source, swaps tilt | 26.9 | 36.8 | 42.9 | 3.9 | 3% | 53% | – | – | 36% | 3% | 55% |
| refined: tight, hold 1 card, a mulligan can't empty a source | 26.7 | 36.2 | 43.5 | 3.5 | 14% | 84% | 6% | 31% | 43% | 3% | 57% |
| scouted: refined, two-faced job entries face down until a job peek | 26.9 | 36.1 | 43.3 | 3.4 | 14% | 83% | 6% | 31% | 42% | 3% | 57% |
Close call is the share of first decisions where the best two moves are within 1 cash. Greedy's first move matches the oracle's in only 10–18% of hands on every ruleset, so the best play is rarely the obvious one. Backing out was never the right call (0%). Triggers fired in 4–10% of runs. Taken side jobs failed 38–48% of the time. A light scar from scar risk was dealt in 3–4% of runs.
What the bots and the hands showed #
The stock moves have three exploits:
- Hold, then push. A held card can't be shed, so holding every card makes a push a free card. Greedy builds 8-card hands this way and beats the oracle. One-shot holds alone don't fix it; a one-card hold limit does.
- Mulligan the job card. With one job slot, a mulligan redeals zero cards, so it is a free discard of the complication. 51% of stock hands go in with no job card. With a source floor of 1 it's 6%.
- Peek is pointless while swaps are free. Sources never run dry, so a blind swap beats looking first (6% of moves). Once swaps cost time, peeks rise to 14%.
Main has a peek bug. A job peek showed the next cards and then tilted the job source, so the next draw could differ from what the peek showed. The lab tilts first, then shows.
Cards and context. Of 59 job and street cards, only 5 are good in some contract-and-runner cells and bad in others: Borrowed kit, Bribable supervisor, Hot gear, Maintenance backdoor and Second chance, all route, kit or deal cards. 24 are always good, 23 always bad, and 7 never matter. "Every street card cuts both ways" doesn't hold when the downside is a flat number.
- Dead: Coffee, Media blackout (fame only), Long window (its tilt is invisible in the plan), Traffic jam (its two shifts cancel), Crowdfunded, Incomplete brief (its side job cancels its cost).
- Always bad: Loan shark, Bodyguard, Rival's offer, Side job, Insurance fraud. Always good: Power outage, Job interview, Ghost map.
- Redundant: Contact A (Job −1, owes a favour) is strictly worse than Favour A (free prep). Borrowed kit is dead when the hand already climbs.
- Saturation: at risk 0, further help is worth nothing but still costs.
Readability. Every generated rules line is 12 words or fewer; the longest are gambles and triggers at 8–11. A bare tag ("+climb") didn't say what it unlocks until the screen showed "+climb → Roof".
What was fun: runner-deck swaps, because you know your remaining deck ("swap Contact A: half the time I get the badge"). Complications that re-route the plan: Double posting sends you over the roof. A job card that targets your strengths: Daemons on the doors against a runner with a deck and a badge. A poor runner taking a cash advance to buy prep. What wasn't fun: street cards with an obvious answer (keep the free good one, swap the bad one), free discards of homework, and flat number shifts.
Traits. On stock rules all twelve traits are within about 2 cash of none, except Overcommitted (−4). Traits that bend moves come alive once moves cost something: Improviser goes to +4.3 under refined. Lucky is negative, because "better of two" ranks by rarity, and rare is not the same as good.
Scouted rules, seed 1, office, Runner B. "Party upstairs?" is face down, at 34% odds of its opening.
The job peek turns it over (the complication) and shows the next two job cards.
Stock rules, seed 7: the plan the runner walks in with, its terms and the mission score.
Recommended core #
- Families: Route (the core), Pressure (few), Hook, Gamble. Deal is the price on any card. Hand rules belong to traits.
- Moves:
- Swap costs time, and a job swap tilts the source back.
- Peek (scout) turns face-down job cards over and shows the next two.
- One push per session, with one held card protected through it.
- A mulligan can't empty a source, so on 1-slot sources it is a runner-deck move only.
- Back out has its small cost.
- Double-edged means context, not a flat penalty. A street card's downside should land on a route, a tag or a stage the job uses, so it is good on one job and bad on another.
- Hooks pay on the spot or carry their own risk, and never add a flat "Out +1". A trigger names its stage and what changes.
How #254 and #255 collapse into it #
- #254 job deck:
- Building and tech, 16–50, become Route over site features: vents, roof, dock, net, staff door, lifts.
- Security and staffing, 1–15, become Pressure. In the real mission they should be roster, patrol and alert changes, not abstract risk.
- Client and brief, 75–86, become Hook (objective, condition, fit). Time, 66–74, becomes Hook triggers.
- The many paired entries (16/17, 58/59, 71/72, 87/88, 89/90, 94/95, 96/97) become two-faced entries.
- Contact 51–53 becomes one Gamble entry.
- #254 street deck:
- Money becomes Deal. Gear, 11–20, becomes Route (tag) plus a Deal. Ways in, 21–30, become
startplus a constraint. - Info becomes Route or Pressure plus a Deal or trigger. People become Pressure plus a favour or a Gamble.
- Heat and fame, 51–60, move to debrief and standing unless they change security.
- Body becomes Pressure plus scar risk. City becomes Route (close or open a kind of route), not symmetric shifts.
- Weird becomes Gamble or Hook.
- Money becomes Deal. Gear, 11–20, becomes Route (tag) plus a Deal. Ways in, 21–30, become
- #255:
- The hand-size, luck, homework and move traits map onto the trait delta (slots, floor, disable, free, weights, tilt by tag, term scaling).
- Category traits become tilt or weight by tag.
- Scar traits (Resilient, Stoic, Troubled) need a scar-aware move rule, not built in the lab.
- "Better of two" needs a meaning of "better" that isn't rarity.