Research Projects — The Management Game: design note

A single self-contained HTML page (index.html), no dependencies, no network, state in localStorage (rpm_v1). Sister of the Research Pitfalls game and built on the same stylesheet and engine helpers (tooltips, term chips, verdict boxes, profile spectra).

It shares that page’s visual identity (its design note states the tokens under Visual elements): system serif headings over Helvetica Neue, hairline rules instead of cards, navy ink #0E1B2E on paper #FBFAF7, blue #2B63AD for sound, coral #C23B22 for the trap, ochre #8F6100 for defensible, and a navy dark mode. On the run screen the month strip sits at the top, and the six dials and progress sit in one row as 1 px tracks with an ink fill that turns coral below 30. The debrief draws the dials as grey lines named at their right end, with progress as the one ink line and coral marks for the months a problem fired; its timeline marks a fired problem with a filled square and a prevented one with a hollow square. The cast on the setting screens is a two-column grid with each personality as a bold run-in.

What it is for

The year a PhD student first leads a project: a supervisor with no time, a senior collaborator with many projects and few time pressures, a master’s student the player is responsible for, and a deadline in month eight. The soft-skill problems that follow (time pressure, authorship fights, silent collaborators, dropped balls, late feedback, burnout) are not taught and are expected. The game makes them consequences of decisions, so the player can see which decision seeded which problem and what would have prevented it.

Sources: lab-culture discussions (roles only, no names: the principal investigator (PI), a student, a postdoc, a senior collaborator), the crisis-management and feedback notes behind them (three components of feedback, situation-behavior-impact, “can I give you some feedback?”, start/stop/continue, kind versus nice, ask for advice not feedback, “can this conflict deepen the relationship?”), two collaboration postmortems (a project stalled across two fields and mismatched time horizons; leading projects as the most junior member, where the coaches’ answer was “you told them how you work; have you asked how they work?”), and the Path to PhD posts on expectations, deadlines, planning, contingency, open-loop projects, meetings, rest, failure, conflict and delegated speed.

The people

The four characters have names and plain-language roles, because a junior player may not know what a PI is or how a master’s student ends up on a paper: Nadia (your supervisor, the PI), Tomasz (a tenured professor at another institute who joined after a chance conversation), Lea (a PhD student in your lab, a year ahead), Ben (the master’s student you co-supervise). Every card and message names them, with the role glossed on first mention.

A run opens with an intro screen (“Before month 1”): who you are (a second-year PhD student running a project for the first time), what the project is, the deadline in month 8, and how the game works, followed by the four people with a one-line role under each name (“your supervisor, the PI”); the same role line sits under the month counter throughout the run, so a junior player never has to remember who Nadia is. Each run also gives each of them one of two personalities, shown on that intro screen before month one: Nadia wants to be asked or trusts you to decide; Tomasz keeps opening doors or wants the paper out; Lea says it straight or goes quiet; Ben asks early or hides being stuck. This is the answer to a fair objection, that the “sound” move often depends on who the people are. On a few cards the personality changes which option is sound (with a supervisor who trusts you to decide, deciding and telling her beats reminding her; with a student who asks early, a Friday note is sound as well as the hour of pairing; with a colleague who goes quiet, asking in private whether the plan still makes sense to her comes before any request), on more cards it changes the reason, and the reveal on every people-dependent card says so and says what the other kind of person would need. A few problems are more or less likely with some people. The screen says plainly that a game can carry only two versions of a person and that reading them is part of the job.

Voice

Every situation is a short scene (who, what happened, what you are looking at, what is at stake) rather than a prompt, and every drill message reads like an email or a Slack message a person would send: greetings, natural sentences, no meta-labels such as “one question, fifteen seconds”. The better message still carries context, a concrete ask, a date and a way to answer briefly, in the way a considerate person writes, without announcing the technique: a good message does not say “you can answer in a line”, “a one-word answer is fine” or “one disagreement, with the evidence”; it asks the question, states the disagreement and points at the table. The weaker message in each pair is one a well-meaning person actually sends (polite, a little vague, missing the date or the ask), not a caricature of apologies, so the choice teaches something. Reveals refer to “the first message” and “the second”; the engine swaps the ordinals when the pair is shown in the other order.

Four parts

1. Run a project. Nine months, two decisions per month drawn from a pool of about three (the kickoff, the deadline week, the submission form and the postmortem are always drawn), plus the problems that fire. Six dials, 0 to 100: alignment, trust, buffer, credit, energy, loop; a progress bar; and a relationship bar per character. Every option is sound, the trap, or defensible-but-costly; the reveal names the effect on the dials, the reason, the sound move if missed, and the problem the option seeds. A card may carry two sound options (the reveal then says “a sound move” rather than “the sound move”), and a defensible option carries costIf, the condition under which its cost starts (“it runs past a week, or someone is writing toward the claim”), printed in the reveal’s header. A defensible option without a costIf carries a standing gloss instead: what turns a defensible call into the trap is usually how long you leave it. This is the answer to the objection that the game scores ordinary diligence as a trap: checking a vanished effect for a few days is sound when the team hears the same day that the claim is under review and by when the check ends, and defensible-with-a-date when it is done in silence; only the open-ended silent rescue is the trap. The same shape applies to deferring to the supervisor on framing after the doubt has been raised once, and to waiting on a supervisor when nobody downstream is blocked.
   • A senior person who does not yield. A card can require a seed and a personality to be drawn (req, reqTrait). With a Tomasz who keeps opening doors, a player who asked the scoping question in month four draws, in month five, a Tomasz who answered it his way and still wants the direction in. The sound move is to take the trade-off to Nadia and let the seniors settle the scope; the reveal says in so many words that the month-four message was right and the game does not charge for it. This answers the objection that the game rewards every well-phrased request as if persuasion always lands.
   • Seeds. Options set named seeds (auth_unspoken, expect_written, hero, …). At the start of each month the game evaluates the problems due that month: each has a base risk and a risk(state) function that adds for the seeds that cause it and subtracts for the ones that prevent it, plus a few dial thresholds (energy below 35 raises burnout). A roll under the risk fires the problem as an incident card with a “why now” banner that names the decisions behind it; a roll above it when the risk was pushed below the base is logged as prevented, with the decision that did it. Some problems fire at low odds whatever the player did, because some cannot be avoided, only handled.
   • Progress grows with every decision by a base amount that the loop and alignment dials raise and an empty buffer or an empty tank halve; a few options add or remove progress directly. Submission at the deadline needs progress above 55 and buffer above zero.
   • Outcome. Submitted; paper quality (progress and loop); team health (trust, alignment, relationships); energy; credit. Verdict bands: Ran it well (75), Shipped at a cost (58), Survived it (40), Learned the hard way. The submission is worth 30 points, so a paper shipped over a broken team or an empty tank does not read as a success.
2. Debrief (the analytics module). A line chart of the six dials and progress over the year with the months a problem fired marked; a timeline that lists every problem that fired with its odds and the decisions it traces back to, every problem prevented with the decision that prevented it, and every trap or costly call with the problem it seeded; a management profile on five spectra (controller–servant leader, assumes–writes it down, keeps the peace–says the hard thing, hero–delegates, sprinter–pacer), this run hollow and all runs filled, each pole with a virtue and a vice; three things to do differently (the problems that fired first, then the lowest dials); a reading list; a copyable summary. A cumulative profile screen shows, across runs, which problems were hit, prevented and repaired.
3. Say it better (the drills). Fifteen pairs of messages for the hard moments: the collaborator who went quiet, the authorship opener, the reminder to a silent PI, feedback on a late section, asking for advice rather than feedback, declining a request, the weekly update, reporting a dead result, a meeting invite, disagreeing with the PI’s framing, a cold email, checking the target venue, inviting a postmortem, telling the PI about burnout, and one written from Ben’s side (a stuck master’s student asking for help), so the junior is a person with his own messages and not only someone the player manages. In some pairs the weaker message is the longer one (apologies, context pasted in, a six-point agenda), in others the shorter, so length does not predict the answer; the better one carries context, a concrete ask, a date, and a way for the other person to answer in one line. A weaker pick is shown as the defensible kind (“The other one does more”), not as a trap, because it is a message a considerate person sends. Rounds of eight, unseen items first.
4. Field manual. Sixteen problems in three groups (alignment and communication; time and load; credit and trust), each with what it looks like, what causes it, how to prevent it, how to repair it once it has fired, and what to read; hit and prevented counts from this device; a glossary with hover tooltips on every card (a tap opens the same tooltip on a phone, where there is no hover; the “terms on this card” strip under each card is the second route); the readings in five groups. The manual uses roles (the PI, the collaborator, the master’s student) rather than the run’s names, so it reads without a run; its subtitle says so.

Didactic choices

• Delayed consequences with a visible trace. The point of the game is that management failures are seeded months before they fire. Every incident banner and every debrief entry names the seeding decision, in the player’s own words, so the lesson is causal rather than moral.
• Prevented problems are shown. Good management is invisible; the debrief makes it visible by listing what did not fire and why.
• Repair is scored separately from prevention. A fired problem has a sound repair, a trap, and a costly option, so the game teaches recovery as well as avoidance.
• Both directions of subjectivity. Which move is sound is one researcher’s judgment; the home screen says so and every reveal gives its reasons. The profile is descriptive, and each pole has a virtue.
• Voice. Peer address, sentence case, American English, typographic quotes, no emoji, no first person outside quoted messages. Terms get a tooltip on first use per card.
• No vocabulary assumed. The audience is a student who has never run anything, so the glossary covers the words the game itself uses without ceremony — the principal investigator (PI), postdoc, tenured, master’s student, authorship and author order, target venue, submission form, reviewer, lemma, cluster, noise, group meeting — and each of them carries a tooltip and a chip wherever it appears. The home screen, the run’s intro and the manual’s subtitle each say that the tooltips are there, because a definition nobody knows to hover over is not a definition.
• Every mode opens on its setting. The drills and the manual are reachable from the top bar and from a deep link without ever playing a run, so each says who the people are before it asks anything. “Before you write” carries the cast with their roles, what the pairs are testing, and the warning that length does not predict the answer; a player arriving from the debrief or asking for another round already has all of it and goes straight to a message.

Deep links

?mode=run|drills|manual|profile. drills opens on the cast, not on the first message.

Prose gate

tools/.limpid/check.sh research-projects lifts the player-facing prose out of index.html and runs the limpid CLI over it; tools/.limpid/trace.py maps a finding back to the line that produced it. The quoted drill messages are excluded, because one of each pair is written to be the weaker one. See tools/.limpid/README.md.