A four-role, live-then-async decision simulation — how it runs, its fairness and safety rules, and a fully worked public example.

The Alignment Lab#

The Alignment Lab is the course’s central exercise: a small group is given a real-shaped decision to make together, each person holding one of four roles and a private brief the others cannot see in full. It is marked as The Alignment Lab assessment, worth 45% of the course.

The scenario used for the graded Lab is confidential and distributed by the teaching team before your group’s session — publishing it here would let later groups prepare against roles they haven’t been assigned yet, which defeats the exercise. This page instead runs you through the process using a small, fully public worked example, so you can see exactly what a role card and a decision record look like before you are handed the real ones.

The four roles#

Every Lab scenario is built around the same four functional archetypes, even though the fictional person and organisation change each time:

• Executive Sponsor — owns the outcome the organisation wants, and answers for it upward.
• Delivery Lead — has to make whatever gets decided actually work.
• Finance and Risk Lead — controls whether the decision can be funded, and what it might cost if it goes wrong.
• Staff Representative — speaks for the people the decision is done to, not just the people making it.

Each role’s private brief carries the same six things: a public responsibility everyone can see, one piece of evidence only that role holds, one constraint they cannot trade away, one form of leverage or dependency over the others, the minimum outcome they can live with, and the specific condition under which they would change their position.

Ground rules, before anyone is assigned a role#

These apply to every run of the Lab, and are agreed before role cards are distributed — not introduced partway through, and not something you discover only in Week 11’s material on the ethics of alignment.

• No real disputes. Every scenario is fictional. You will never be asked to re-litigate an actual disagreement from your workplace, household or student group.
• Opt-out without penalty. If a role or scenario would put you in a position you are not comfortable with, you can request a swap, or the alternative pathway below, by messaging your facilitator any time before role cards are distributed — no explanation required. The same pathway is also where a student who missed role assignment and the contact window entirely is routed — see “Make-up and async participation” below for why that case cannot use the ordinary async make-up.
  • The alternative / structured-async pathway, concretely: you work from this page’s public worked example instead of the group’s live, confidential scenario, but you do not simply read and comment on the example materials already on this page. Playing one role through the Contact Simulator in Section 2 above (or, as a written equivalent, the same role cards’ frozen contact transcripts your facilitator can provide), you choose whom to contact within the same self-initiated contact limit everyone else has, and at every contact log its purpose, the new information it surfaced, any commitment made, and whether your judgment changed — the same four things a live contact log records. That produces your own structured negotiation log, not a summary of the page’s example one in Section 4. From your own negotiation log and the fully revealed role cards, you then independently produce your own decision record, in the same format as Section 5’s, and write your own individual autopsy and partial rerun of it. Every artefact you submit is one you produced, not an analysis of one the page already gives you a finished answer for.
  • It is marked against the same five criteria as the group-based submission (see the assessment), with one adjustment. Four of the five — initial judgment and strategy memo (20%), pre-meeting negotiation (20%), individual decision autopsy (35%), partial rerun (10%) — already mark individual work on a record each person produces or works through themselves, so nothing about them requires a live group. The fifth, formal decision record (15%), is normally the one artefact a live group produces together and is marked once, identically, for everyone named in it — with no group to produce one, this pathway marks it instead on the decision record you independently produce from your own negotiation log, treated as an individual submission standing in for the shared artefact, not on a commentary about the page’s example record. See the criterion table under “What you submit” near the bottom of this page for exactly which pathway each criterion draws its evidence from, and why each is held to the same evidentiary standard.
• In-role, not in-character. You are arguing a position, not performing a persona. Personal attacks, real names, and anything outside the brief are out of bounds, and the session facilitator can pause the room if the line is crossed.
• Nothing said in-role is evidence of what you personally believe. The decision autopsy afterwards is marked on your analysis, not on the position your role required you to argue.
• The record is the record. Whatever the group’s decision record says is what gets marked as the group’s outcome, even if you privately disagreed — which is precisely the situation the decision autopsy asks you to examine.
• Absence doesn’t erase information. From the moment role cards are distributed, each person keeps a continuity memo alongside their brief — the same five fields every time: current position, commitments made or received, evidence surfaced, what’s still unresolved, and the single most important thing for a replacement to know first. It is updated after every contact you log in your negotiation log, and again immediately before the meeting starts, so it is never more than one contact out of date.
  • How this reaches the group if you’re absent: if the person holding a role cannot attend, the group reads that role’s continuity memo aloud at the start of the meeting, and plays that position from the memo as written — not from guesswork about what the absent person “probably” would have argued beyond what the memo actually says.
  • What it produces, and how it’s marked: the memo is submitted alongside your negotiation log. It is not marked as a separate criterion; it feeds the “Pre-meeting negotiation” criterion (20%) as the evidence that negotiation continuity survived your own possible absence, and it protects your own “Individual decision autopsy” mark (35%) by giving you, the absent person, a documented position of your own to autopsy from even though you were not in the room.
• Make-up and async participation are always available, and split into two genuinely different cases depending on what you actually missed.
  • (a) You were assigned a role and had the contact window, but missed only the live meeting itself. Your initial judgment, strategy memo and negotiation log are already your own genuine work from before and during the real contact window — you submit exactly those, on the same clock as everyone else, and do not reconstruct or rewrite them from the group’s materials after the fact; the point of this case is that you were present for everything except the meeting, so nothing about your own artefacts up to that point needs redoing. What you didn’t get to do is help the group produce its decision record live, so once your group does meet and produce one, you are given everything it had at the reveal, not a summary of it: the full reveal pack (every role’s brief, not only your own), the group’s own negotiation log and contact records, the decision record itself, and the continuity memo of the role you’re standing in for. Using those materials, within one week of receiving them, you audit the group’s actual decision record against the full reveal pack and submit your own corrected or reconstructed decision record: where you agree with a field, say so and cite what in the reveal pack supports it; where a field is incomplete or wrong against the full pack, name the specific correction and the evidence for it. This is scored against the same ten fields and the same quality standard as a live record — not against whether it matches the group’s original, since a well-evidenced disagreement with the original record is not itself a mark against you. From your own negotiation log, the group’s decision record and your corrected version of it, you then submit the same individual autopsy and partial rerun everyone else does.
  • (b) You missed role assignment or the contact window itself, not only the meeting. This is not folded into case (a) above — there is no genuine negotiation log or strategy memo of your own to submit if you were never in the window that would have produced them, and pretending to reconstruct one after the fact from the group’s record would mean submitting someone else’s evidence as your own. Instead you are routed to the alternative / structured-async pathway described under “Opt-out without penalty” above, which is built from the start to produce every artefact independently rather than assuming a real contact window happened for you.
  • Why no one’s grade depends on a groupmate’s attendance or timeliness: the negotiation log, strategy memo and individual autopsy are all marked as individual work regardless of who else was present or how promptly they submitted, and the decision record is finalised at the live meeting itself, then marked once, identically, for everyone named in it — a groupmate’s absence or a late personal submission changes how they produced their own account of the record, not which artefacts exist for anyone else, including you, to be marked on.
  • Deadlines, and what actually governs them: the relative windows above (48–72 hours for contact, one week for the async make-up) describe the normal sequencing of a run, but the fixed date and time on the Alignment Lab assessment page is the actual deadline for every artefact — none of the relative language above extends a submission past it. If your group’s meeting or reveal lands late enough that a relative window would run past that fixed date, the fixed date still governs; raise the timing with your facilitator rather than assuming the relative window buys extra time.
• Peer feedback checks contributions, not popularity. Any peer input into marking asks what a specific person actually contributed — a piece of evidence surfaced, a commitment logged, a position changed — never a general vote on who did well.
• You exit the role afterwards. Once the reveal has happened, the group explicitly closes the exercise before moving on: what a role did in the simulation is discussed as the role’s behaviour, not as a judgement of the classmate who played it.

How a run of the Lab works#

0. Rehearsal. A small, ungraded run of this whole process — using the public worked example below — before anyone sits the assessed Lab.
1. Public materials and initial judgment. Before any role is assigned, everyone reads the same public materials and submits an individual initial judgment: what they think should happen, on the record, before anyone has a stake in a role.
2. Role assignment and strategy memo. Each person receives one confidential role brief and, individually, submits a one-page strategy memo: their responsibilities, their evidence, what they don’t know, who they plan to contact, and the ethical boundaries they won’t cross.
3. Pre-meeting contact window (48–72 hours). By default, each person may make at most two self-initiated contacts — but can reply to as many requests from others as they like. Every contact, sent or received, is logged by both parties: its purpose, any new information, any commitments made, and whether either person’s judgment changed. The simulator below shows the shape of this window using the public worked example.
4. The meeting. Live or synchronous, one sitting of around 25 minutes, no extensions on the clock once it starts.
5. Decision record. The group produces one record — the decision or a reasoned deferral, who holds final decision authority over it and separately who holds spending and implementation authority (and whether each has actually been granted, not just implied), evidence versus unverified assumption, unresolved dissent, who owns the resulting actions, and what would reopen it — using the decision record template. Consensus is not required.
6. Reveal and consequence test. Every role brief is shared with the whole group, then a choice-based consequence test runs against the decision record. Its outcomes come from explicit, traceable case rules, not from chance or dramatic licence — it is a teaching model of one plausible consequence, not a claim that a real decision like it would play out the same way.
7. Individual autopsy and partial rerun. Individually: a decision autopsy reaching one of three conclusions about what actually decided the outcome — a specific distortion, no significant distortion, or indeterminate on the available record (naming the missing evidence, the alternative you could not rule out, and what would change your judgment) — then a short rerun of one specific juncture to test whether a named process change would have produced a different one.

Worked example: should the Reading Room move to hot-desking?#

This is a small, entirely public scenario used only to demonstrate the mechanics above — it is not the graded Lab scenario, and none of the documents below are marked on how convincing they sound, only on the evidence and reasoning inside them. It walks through every artefact a real run produces, end to end, with an example filled in at each step.

Facilities wants to convert 40 assigned desks in the Reading Room into unassigned hot-desks, to fit a larger cohort. Staff were not consulted before the proposal was drafted. Four people, one for each archetype, have to reach a decision.

1. Public materials and an example initial judgment#

Everyone reads the paragraph above — the same materials, before any role is assigned — and records what they think should happen. Here is one example of a filled-in initial judgment:

Converting to hot-desking might be defensible if utilisation really is low, but a proposal drafted with no staff input is a bad way to get there even if the destination is right. Before I’d support it, I would want to see the actual utilisation data behind the proposal, and confirmation that anyone with a documented accessibility need has been asked what they require. My initial position: support investigating the change, do not support committing to a date yet.

This is submitted on the record before role assignment specifically so it can later be compared with what the same person argued once they were holding a role — the gap between the two is often the most interesting thing in the decision autopsy.

2. The four role cards, fully revealed#

Because this is the public example, every field of every role card is shown below — which is exactly what would not happen before a real, graded run of the Lab, where each person sees only their own card until the reveal.

3. A filled example strategy memo#

Individually, each person turns their role card into a one-page memo before contacting anyone. Here is the Staff Representative’s:

My responsibilities: speak for the 12 staff who currently have an assigned desk, and specifically flag anything that affects the 3 with fitted accessibility equipment.

My evidence: three staff have accessibility equipment permanently fitted to their current desk that a hot-desk cannot accommodate without notice. I don’t yet have a cost estimate for reserved-desk hardware, or know whether Facilities has already budgeted for it.

What I don’t know: whether the booking software has any bearing on cost at all, whether the fit-out has been costed by anyone yet, and whether Finance would treat these as one proposal or two.

Who I plan to contact: the Delivery Lead first, since they own whatever the room’s technical side involves, to ask what it would actually take to guarantee reserved desks and whether that raises a cost I don’t already know about; the Finance and Risk Lead only if that conversation raises a number I need confirmed.

Ethical boundaries I won’t cross: I will not agree to any timeline that has not been taken back to affected staff for comment, regardless of what it would cost the group to wait.

4. An example negotiation log#

The log is not capped at two entries — it is the two self-initiated contacts that are capped; replies to requests from others, and any follow-up on the same contact, are logged too. Purpose, new information, any commitment made, and whether either side’s judgment changed are all recorded for every entry:

1. Staff Representative → Delivery Lead (self-initiated). Purpose: find out if the software upgrade and accessibility fit-out could be one proposal. New information: drawing on the licence terms only they had read (their stated leverage on the role card), the Delivery Lead confirms the booking software already supports named, reserved desks at no extra cost — only genuine unassigned hot-desking needs the paid upgrade. Commitment: Delivery Lead agrees to draft a combined line item. Judgment change: Staff Representative’s minimum acceptable outcome now looks achievable without a fight — a material change from the initial judgment above.
2. Finance and Risk Lead → Staff Representative (self-initiated). Purpose: Finance asked what the accessibility fit-out would cost, to fold it into the same costed proposal as the software upgrade. New information: Staff Representative did not have a number, but named the Facilities staff member who does. Commitment: Staff Representative will chase that number before the meeting. Judgment change: none recorded either side.
3. Staff Representative → Executive Sponsor (self-initiated, second and final self-initiated contact). Purpose: check whether a slower phase-in would still count toward this year’s target. New information: Executive Sponsor confirms they would accept a slower phase-in if it comes with a firm date. Commitment: none made either way. Judgment change: Executive Sponsor’s position moves from “any plan, fast” toward “a firm date matters more than speed” — a change worth bringing into the meeting.

5. A formal decision record example#

The record the group actually produces does not have to be a clean agreement — dissent, conditional support and deferral are all legitimate outcomes, provided the record says so honestly:

Decision: proceed with a combined proposal — booking-software upgrade and reserved accessibility desks — as one line item, with the accessibility fit-out still awaiting full costing, phased in over two terms with reserved desks guaranteed from day one, on condition that the phase-in timeline is taken back to the 12 affected staff for comment before it is finalised. That condition is not the group’s addition — it is the Staff Representative’s stated non-tradeable constraint, so the decision is written as conditional on it rather than as settled and separately hoping the constraint gets honoured later.

Who agreed: Executive Sponsor, Delivery Lead and Finance and Risk Lead agree to the proposal as drafted, including the consultation condition above. The Staff Representative’s position is recorded as final agreement pending consultation, not agreement to a settled plan: she has not yet agreed to anything final, because her own non-tradeable constraint makes the plan’s status contingent on a consultation that has not happened yet.

Evidence relied on: the booking-software licence terms (Delivery Lead); the accessibility equipment list (Staff Representative); the utilisation report (Executive Sponsor).

Unverified assumption: that the accessibility fit-out cost, still unconfirmed at meeting time, will not exceed the software upgrade saving Finance identified; and that returning the timeline to the 12 affected staff for comment will not surface an objection serious enough to reopen the phase-in itself.

Dissent: Finance and Risk Lead supports the plan conditionally — on record as not approving final spend until the fit-out figure is confirmed in writing. The Staff Representative’s final-agreement-pending- consultation position above is recorded separately from dissent: she is not opposing the plan, she is withholding final agreement until her non-tradeable constraint is actually met, which this record treats as a different thing from disagreeing with the plan’s content.

Action owner: Delivery Lead drafts the combined proposal; Staff Representative confirms the fit-out figure within one week, and separately runs the staff-consultation step before the phase-in date is fixed — her constraint was that the timeline goes back to staff before it holds, not merely before it is announced.

Final decision authority: Executive Sponsor signs off the proposal going forward once the fit-out figure lands and the staff consultation above has actually happened — not the group by majority, and not a completed decision on either point alone. This sign-off is a licence to keep drafting and costing the combined line item; it is not, by itself, either of the two authorities below.

Spending authority: Finance and Risk Lead, under their standing authority over spend above the university’s threshold — not yet granted. The Executive Sponsor’s final decision authority above does not stand in for this; any actual spend still requires the Finance and Risk Lead’s own sign-off once the fit-out figure is confirmed, which this record does not and cannot override.

Implementation authority: Delivery Lead, for the software configuration, and Facilities, for the physical fit-out — not yet granted. Purchasing or installing the accessibility equipment, physically reassigning or relabelling any of the 40 desks, and telling the 12 affected staff a fixed go-live date are all implementation actions that depend on spending authority above being granted first; none of them may happen before then.

What would reopen this record: a fit-out figure that exceeds the software-upgrade saving, Facilities confirming the “no extra cost” licence claim was wrong, or the staff consultation surfacing an objection to the phase-in that the group has not addressed.

6. Consequences, from stated rules#

The consequence test is not a dramatised outcome — it applies explicit, ordered rules to whatever the decision record actually says, so the same record always produces the same consequence. As of this page, those rules are not only argued in prose: they are a small, typed, pure function — evaluateCostAndAuthority and evaluateRoleRule, defined in src/data/consequence-rules.ts — and the worked result below is that function’s actual output on the record above, not a hand-written paraphrase of what it should say.

Be precise about what that function does and does not do. It does not read the prose record above and parse it — no software here extracts “the fit-out cost is unconfirmed” from a paragraph of English. A human reads the record and codes its relevant facts into the typed inputs the function actually takes, hotDeskingCostAuthorityInput and hotDeskingRoleRuleInputs in src/data/consequence-rules.ts (for instance, costFullyKnown: false and missingFigure: { name: "the accessibility fit-out cost", owner: "Staff Representative", deadline: "within one week" }, coded from the record’s own “Unverified assumption” and “Action owner” fields above). That coding step is a human judgement call, not a mechanical extraction, which is exactly why the walkthrough below cites the specific record text behind each field it uses — so the coding is auditable, not asserted. What the function then guarantees is only that its rules are applied to that coded input consistently and deterministically, every time, rather than being re-argued from scratch in prose on each page that needs the result.

The rules turn on three kinds of authority a record can name, which are not the same thing and a record naming one does not establish the others:

• Final decision authority — who signs off on the decision going forward.
• Spending authority — who can actually authorise the money moving. Holding final decision authority over a record does not, on its own, mean you hold spending authority over what it costs.
• Implementation authority — who can act on the decision in the world (buy the equipment, move the desks, announce the date). Nothing implementation-shaped may proceed until spending authority has actually been obtained, however clearly the decision itself has been signed off.

The rules are checked in this order, because a record can honestly satisfy more than one trigger at once — “no named final decision authority” and “defers spend pending a missing figure” are independent facts about a record, not opposite ends of one switch, so a record missing both can trip both at the same time unless something decides which one governs:

1. Is there a named final decision authority? If no → the Space Committee sends the record back for one revision before it will consider anything else about it. This is checked first: without someone named to hold the record’s decisions, there is no one for a deferred figure to come back to, so an authority gap always takes priority over a costing gap, including when both are missing at once.
2. Does the record treat that final decision authority’s sign-off as spending authority? If yes → sent back for revision regardless of how complete the costing is. This is the mistake the worked example’s own record is careful to avoid: naming a final decision authority is not the same fact as that person, or anyone else, having spending authority — a record that quietly conflates the two has made an authority error, not a costing one, and rule 3 below never gets to evaluate a record that makes it.
3. If a final decision authority is named and not mistaken for spending authority, is the cost fully costed and validly authorised? This branch is exhaustive — every remaining record lands in exactly one of three buckets, and the test for which one is not just “does the record name a missing figure” but “does the record actually withhold spend until that figure lands, and has spending authority actually been obtained”:
   • Fully costed and validly authorised → the figure is complete and the spend has been approved by someone who actually holds spending authority → the Space Committee accepts the plan as submitted, no delay.
   • Specific cost explicitly deferred — the record says exactly which number is missing, who owns getting it, and the deadline, and explicitly prohibits any dependent spend until that figure is confirmed and validly approved → a one-week delay while that figure is confirmed, and any other commitment already made in the record (an accessibility guarantee, for instance) still holds regardless of the delay.
   • Uncosted, prematurely authorised, or authorised by the wrong authority — this branch covers four distinct ways a record can fail to earn acceptance or a genuine deferral: a vague “cost to be confirmed” that names no figure, owner or deadline; a record that does name the missing figure and its owner but still authorises the dependent spend as if the figure were already confirmed; a record where the cost is fully known but no one who actually holds spending authority has approved it; and (caught earlier, at rule 2) a record where final decision authority is wrongly treated as spending authority. Naming the gap does not move a record out of this branch; only actually withholding spend until the gap is closed, and getting the spend approved by someone with real spending authority, does. This is a harder stop than the middle branch: a genuine deferral has a concrete number to wait for and nothing spend-related moves until it lands, so the rest of the record can proceed around it; a record that authorises spend regardless — or that was never going to be approved by anyone with the authority to approve it — has nothing left to wait for by the time the Committee sees it, so nothing in it is treated as settled until it is re-drafted around an actual, honoured deferral and a valid approval.
4. Independently of 1–3, for each role in turn: does the record satisfy both that role’s stated minimum acceptable outcome and their non-tradeable constraint from their role card, checked together rather than as alternatives? Meeting the minimum outcome does not excuse silently overriding the constraint, and honouring the constraint does not excuse falling short of the minimum outcome — either gap on its own flags the record as contested for that role going into the reveal, regardless of whether they spoke up in the room, and regardless of which outcome 1–3 produced. A record can be accepted or delayed by rules 1–3 and still be contested under rule 4 for one or more roles. A record the text genuinely doesn’t confirm either way for a given role — on the outcome, the constraint, or both — is left unresolved rather than guessed at, exactly as an evidence-limited autopsy would treat it — see the walkthrough below for a case where this happens.

Applied to the record above, rule by rule — each bullet names what the function actually returned, computed live from hotDeskingCostAuthorityInput and hotDeskingRoleRuleInputs:

• Rule 1 — named final decision authority? Yes: “Executive Sponsor signs off the proposal going forward.” Proceed to rule 2.
• Rule 2 — final authority mistaken for spending authority? No: the record is explicit that the sponsor’s sign-off “authorises drafting and costing the combined line item; it does not itself authorise spending.” Proceed to rule 3.
• Rule 3 — which of the three cost branches? The function’s computed result: consequence delayed (the record names the missing figure, its owner and a deadline, and explicitly withholds dependent spend until it is confirmed and validly approved). This is the middle branch, not the third one: the fit-out figure is an explicitly named, specific gap (“Unverified assumption: that the accessibility fit-out cost… will not exceed the software upgrade saving”) with an owner (Staff Representative) and a (within one week) deadline, so there is a concrete number and date to wait for rather than an open-ended “cost to be confirmed.” Just as important, the record actually withholds spend on it, with any real spend requiring the Finance and Risk Lead’s own sign-off first — naming the gap alone would not be enough to land here; it is the combination of naming it and not spending through it that keeps this record out of the third branch. → One-week delay, accessibility commitment holds.
• Rule 4 — each role’s minimum outcome and constraint, checked together:
  • Delivery Lead — minimum outcome (“a decision that does not commit to a go-live date before the software question is answered”) met: the software question was answered in the negotiation log before the meeting, and the record’s phase-in follows from that answer, not ahead of it. Constraint (“unassigned hot-desking cannot proceed until the paid software upgrade it requires has an approved budget and valid spending authority behind it”) honoured, not bypassed: the record’s own Implementation authority field states plainly that the Delivery Lead’s software configuration work is “not yet granted” and depends on spending authority being granted first — nothing implementation-shaped, including configuring the upgrade, may proceed before then. The upgrade now has a known price and sits inside the same line item Finance is reviewing, but a known price is not an approved budget, and the constraint does not ask for a price — it asks that nothing proceed without one. → clear: the constraint is met because the record explicitly withholds implementation until budget approval and spending authority both exist, not because the upgrade is now funded, which the record itself does not claim.
  • Staff Representative — minimum outcome (“named, reserved desks for anyone with a documented accessibility need”) met: the record names reserved desks “guaranteed from day one.” Constraint (“cannot agree to anything not first taken back to staff for comment”) honoured, not bypassed: the record makes her agreement explicitly conditional on that consultation happening, rather than recording full agreement and hoping the consultation follows. → clear.
  • Finance and Risk Lead — minimum outcome (“a named, costed proposal… rather than an in-principle agreement to spend later”) not met: the record’s own Dissent field already says so directly — conditional support, “not approving final spend until the fit-out figure is confirmed in writing.” Constraint (“cannot approve new spend without a proposal that names a specific line item, not a category”) met: the proposal is one named, combined line item, not a category. → contested flag for this role (from the unmet minimum outcome, not the constraint), on top of the rule-3 delay.
  • Executive Sponsor — minimum outcome (“any plan that raises reported utilisation above 50% within two terms”) undetermined from the record: the record commits to a two-term phase-in but never states a resulting utilisation figure, so whether this specific minimum is met cannot be read off the text either way. Constraint (“must report a credible plan to the Space Committee within the month”) is also undetermined, not met: nothing in the worked example states when the Executive Sponsor’s month started, so there is no calculable date to check this record’s timing against. This record is a plan for the Space Committee, but whether it lands inside or outside that unstated window cannot be read off the case materials either — marking it “met” would require inventing a start date nothing in the example provides. → both fields undetermined from the record — reported as fully unresolved for this role, not rounded up to “clear” or down to “contested” on either half. That is a limit of what the record and the wider worked example say, not a finding that either half is unmet — the honest call for both halves is “cannot be determined from this record,” not a guess in either direction.

Final result, computed: cost-and-authority consequence delayed (rule 3 — one-week delay); role rule (rule 4) clear for the Delivery Lead (their non-tradeable constraint honoured by the record’s deferred implementation authority, not by the upgrade being funded) and the Staff Representative (her non-tradeable constraint honoured rather than bypassed); contested for the Finance and Risk Lead (the minimum outcome is not met by the record); and undetermined for the Executive Sponsor (the record does not state enough to determine this role's outcome and/or constraint either way) — four separate, individually traceable rule-4 outputs, one per role, alongside the rule-3 delay as a fifth, distinct result — never collapsed into a single “one-week delay” headline that quietly drops the other four.

7. An example retrospective and partial rerun#

After the reveal, each person writes a short individual autopsy. One example, from whoever played the Staff Representative:

What actually shifted my position wasn’t the meeting — it was contacting the Delivery Lead beforehand and learning the booking software was already free under the existing licence. By the time we sat down, I was arguing from “this is basically free,” not “please don’t take our desks” — and looking back, that shorthand overstated what I actually knew. Only the software licence was confirmed free; the accessibility fit-out itself was never costed at that point, and “the software part is free” is not the same claim as “the whole thing is free,” even though that’s the version I let the room hear. It’s still what changed how much the Executive Sponsor was willing to concede, which is worth being honest about: it worked partly because it sounded more settled than the evidence actually supported. If I hadn’t made that contact at all, I think the meeting would have produced a straight yes/no vote instead of a combined proposal — but the contact earning that outcome doesn’t excuse overstating what it proved.

The partial rerun makes this same claim testable by labelling exactly what it changes and what it assumes, rather than leaving “I think” as the final word:

Changed variable: the Staff Representative’s two self-initiated contacts go to the Finance and Risk Lead and the Executive Sponsor instead of the Delivery Lead.

Held constant: every other role’s private brief, the meeting length, and the order in which items were raised in the room.

Resulting difference: without contacting the Delivery Lead, the Staff Representative never learns that the booking software is already free under the existing licence, and walks into the meeting able to argue only “this costs something” — rather than the narrower, accurate claim that contact actually supports: “the software side of this costs nothing; the fit-out is still an open figure.” (The Staff Representative’s real account above overstated even that narrower claim as “basically free” — this rerun states the accurate version instead of repeating the overstatement.)

Unverified assumption: that the Finance and Risk Lead or Executive Sponsor would not have independently surfaced the same licence fact if asked directly. Nothing in the record confirms or rules this out — the rerun’s conclusion holds only as strongly as this assumption does.

This is the format the partial rerun expects: one named change, everything else pinned down explicitly, the concrete difference that follows from it, and the one assumption still doing unverified work — not a general impression of what “probably” would have happened.

What you submit#

See The Alignment Lab for the full marking breakdown. In short: an initial judgment, a strategy memo, a negotiation log, a decision record — produced with your live group, or individually under the async make-up or alternative pathway, per the breakdown below — an individual decision autopsy, and a partial rerun of one key juncture — each modelled above using the public example, and each available as a starting-point format on the templates page.

Each criterion’s evidence comes from a different place depending on which of the three pathways you’re on — live, async make-up (you had a role and a contact window, but missed only the live meeting — case (a) above), or the alternative / structured-async pathway (an opt-out, or case (b) above: you missed role assignment or the contact window itself). The weights and total (100% of the Lab’s own mark) are the same across all three; only where the evidence for each one comes from, and what counts as meeting the standard, changes:

Why each row is equivalent, not merely parallel#

Four of the five rows are individual work under every pathway; only the decision record briefly reads as “shared” in the live case, which is exactly why it gets its own, longer reasoning below rather than the others’ shorter one.

• Initial judgment & strategy memo. What’s marked is reasoning produced before any negotiated information existed — every pathway enforces that same ordering (before role cards / contact for the live and async make-up cases, before touching the simulator for the alternative pathway), so the task tests the same skill regardless of which materials supplied it.
• Pre-meeting negotiation. What’s marked is whether each logged contact records its purpose, the new information it surfaced, any commitment made, and whether judgment changed — not whether the contact happened inside a real-time group. The Contact Simulator enforces the same self-initiated contact cap and reveals the same shape of fact-plus-constraint a live contact would, so a log built by playing it through is evidence of the same skill — using a limited number of contact opportunities to gather and act on evidence — under a worked rather than live counterpart, not a smaller version of the task.
• Formal decision record. In every pathway this is marked against the same ten fields on the decision record template (the decision, evidence relied on, unverified assumptions, who agreed, who dissented, action owners, final decision authority, spending authority, implementation authority, and what would reopen the record) to the same standard of evidence-use. What differs is only who supplies the underlying negotiation the record is built from — a live group, an audited correction of one, or a record you produce yourself — never whether you personally sat in a room. A defensible, well-evidenced correction to the group’s own record (async make-up) is treated exactly as favourably as a defensible original one (alternative), because the criterion asks whether the record meets the standard, not whether it agrees with someone else’s draft.
• Individual decision autopsy. The autopsy is marked on the reasoning applied to a decision record and the role briefs behind it, not on whether the record was produced live in front of the writer. A record read carefully in full, alongside the complete reveal pack, offers the same evidentiary base as one watched being produced — the task was never “recall what you personally witnessed,” it was always “reason from a documented record together with the evidence behind it,” which every pathway supplies.
• Partial rerun. A rerun only ever needs one concrete juncture, one named process change, and a defended prediction of the concrete difference that follows. A documented juncture — from written make-up materials or a logged Contact Simulator run — supplies exactly those same three things a live juncture would; nothing about the task requires having lived the juncture rather than read it precisely.