Same pitch, different end-games

Five simulated stakeholders at a fictional freight operator read one fictional startup's pitch in three positionings and deliberate by mail. Across 120 runs they engaged the vendor 83%, 83% and 80% of the time (enterprise, research, upside): what the simulator produced, not an estimate of what real executives would do.

120runs: 30 seeds × 4 arms
80%upside runs engaged (pilot, partnership, acquisition exploration)
−3 ppupside − enterprise, paired; 95% CI −23 pp to +17 pp
+17 ppnegative control: identical pitch, fresh model samples
One run, every stakeholder's mail. Seed 1001, enterprise software pitch: 14 turns until a pilot is ratified in week 5. The run was replayed offline from the journal and its weekly state hashes checked against the recorded run. Stakeholders in this study act on the world's mail and documents directly rather than through a screen, so each is given a desktop in a separate viewer world that holds, after every turn, exactly the mail and documents the replay reached; the engine draws those desktops. The large screen is whoever just acted, showing the last message they sent.

The question

Holding facts, price and diligence fixed, does a pitch's positioning change the institutional end-game? Preregistered outcome: the share of runs that end engaged with the vendor.

Holding the vendor's facts, price and diligence answers fixed, does the positioning of a pitch change which institutional end-game an organisation reaches? The preregistered outcome is the share of runs that end engaged with the vendor (a pilot, a partnership or acquisition exploration), compared between strategic-upside and enterprise positioning and between research-institution and enterprise positioning, paired by seed. It would be falsified, in this simulator, by an upside or research arm whose engagement rate differs from enterprise by more than the replicate arm's noise.

Neither preregistered contrast cleared the bar: positioning did not detectably change how often the organisation engaged the vendor (strategic upside −3 pp, research ±0 pp against enterprise; Holm p = 1.000 and 1.000). The negative control, the same enterprise pitch with independent model samples, differed by +17 pp (p = 0.302) and reached the same terminal state as enterprise in 50% of seeds, which is the run-to-run noise any other difference has to be read against.

Where the runs ended

Enterprise softwareResearch institutionStrategic upsideEnterprise (replicate)
Share of runs ending in each terminal state, by positioning arm, with 95% Wilson intervals0%25%50%75%100%PilotEnterprise software · Pilot: 25 of 30 runs (83%), 95% CI 66%–93%25/30Research institution · Pilot: 25 of 30 runs (83%), 95% CI 66%–93%25/30Strategic upside · Pilot: 24 of 30 runs (80%), 95% CI 63%–90%24/30Enterprise (replicate) · Pilot: 20 of 30 runs (67%), 95% CI 49%–81%20/30Internal buildEnterprise software · Internal build: 4 of 30 runs (13%), 95% CI 5%–30%4/30Research institution · Internal build: 3 of 30 runs (10%), 95% CI 3%–26%3/30Strategic upside · Internal build: 5 of 30 runs (17%), 95% CI 7%–34%5/30Enterprise (replicate) · Internal build: 4 of 30 runs (13%), 95% CI 5%–30%4/30RejectedEnterprise software · Rejected: 0 of 30 runs (0%), 95% CI 0%–11%0/30Research institution · Rejected: 0 of 30 runs (0%), 95% CI 0%–11%0/30Strategic upside · Rejected: 0 of 30 runs (0%), 95% CI 0%–11%0/30Enterprise (replicate) · Rejected: 1 of 30 runs (3%), 95% CI 1%–17%1/30No decision (10 weeks)Enterprise software · No decision (10 weeks): 1 of 30 runs (3%), 95% CI 1%–17%1/30Research institution · No decision (10 weeks): 2 of 30 runs (7%), 95% CI 2%–21%2/30Strategic upside · No decision (10 weeks): 1 of 30 runs (3%), 95% CI 1%–17%1/30Enterprise (replicate) · No decision (10 weeks): 5 of 30 runs (17%), 95% CI 7%–34%5/30
Figure 1. Share of the 30 runs per arm ending in each terminal state, with 95% Wilson intervals (whiskers); counts at the right. A terminal state is the first proposal ratified under the decision rights, or no decision after ten simulated weeks. States no run reached are omitted.
Terminal states per arm (n = 30 runs each)
Terminal stateEnterprise softwareResearch institutionStrategic upsideEnterprise (replicate)
Pilot25 (83%)25 (83%)24 (80%)20 (67%)
Partnership0 (0%)0 (0%)0 (0%)0 (0%)
Acquisition exploration0 (0%)0 (0%)0 (0%)0 (0%)
Internal build4 (13%)3 (10%)5 (17%)4 (13%)
Rejected0 (0%)0 (0%)0 (0%)1 (3%)
No decision (10 weeks)1 (3%)2 (7%)1 (3%)5 (17%)
Engaged (primary)25 (83%; 66%–93%)25 (83%; 66%–93%)24 (80%; 63%–90%)20 (67%; 49%–81%)
Preregistered contrasts on the engaged share, paired by seed
ContrastDifference95% CI (seed bootstrap)Seeds that differExact pHolm p
Strategic upside − Enterprise software−3 pp−23 pp to +17 pp9 of 301.0001.000
Research institution − Enterprise software±0 pp−20 pp to +20 pp8 of 301.0001.000
Enterprise software − Enterprise (replicate)+17 pp−10 pp to +40 pp15 of 300.302control

The test enumerates every sign assignment of the seeds' paired differences (exact sign-flip), so it assumes nothing about the distribution; the interval resamples seeds. With 30 pairs, a difference has to be large to be detected: this design can rule out large positioning effects inside the simulator, not small ones.

What changed, and what did not

The four findings in detail

The end-game did not move with the positioning. In every arm the modal outcome was a pilot, and the terminal-state distributions of the three positionings are within 0.07 of each other in total variation distance. The identical enterprise pitch, re-sampled, sat 0.17 away from the original and reached the same terminal state in only 50% of seeds, against 73% for enterprise and research and 70% for enterprise and upside. The largest positioning effect on the engaged share (3 pp) is smaller than the gap between the two enterprise runs (17 pp): at this sample size the model's own variation dominates anything positioning did.

The strategic-upside pitch never produced acquisition exploration. It invited “a deeper combination of the two companies”, finance diligence was almost always requested, and the sponsor, engineering and finance all held the right to propose it, yet no run in any arm filed one. The finance persona's private knowledge (a written-down acquisition last year) and a $140,000 pilot that fits the remaining budget made the pilot the path of least resistance for every framing.

Where the arms did differ was in the process, not the outcome. The competing internal team was ever shown the pitch in 63% of enterprise runs against 20% (research) and 30% (upside), and an internal build was proposed in 63%, 43% and 50% of runs respectively. The replicate's 37% and 30% show these process differences are also within sampling noise; they are leads for a larger run, not findings.

A known defect. 41 of the turns (Enterprise 17, Research 7, Upside 5, Replicate 12) came back empty because the model spent its whole 2,000-token budget reasoning. As preregistered, each counts as a week in which that stakeholder did nothing; their unread mail was still marked read, so some lost a later turn as well. The defect is in the shared model client's token cap, not in the arms, but it is unevenly spread and is a confounder for small differences.

Secondary: each terminal state

Paired differences in the share ending in each state (unadjusted exact p; exploratory)
StateUpside − enterpriseResearch − enterpriseEnterprise − replicate
Pilot−3 pp (p 1.000)±0 pp (p 1.000)+17 pp (p 0.302)
Acquisition exploration±0 pp (p 1.000)±0 pp (p 1.000)±0 pp (p 1.000)
Internal build+3 pp (p 1.000)−3 pp (p 1.000)±0 pp (p 1.000)
Rejected±0 pp (p 1.000)±0 pp (p 1.000)−3 pp (p 1.000)
No decision (10 weeks)±0 pp (p 1.000)+3 pp (p 1.000)−13 pp (p 0.219)

Every trajectory, branch by branch

31 distinct trajectories came out of 120 runs; click a branch to open its runs and their trace lines.

The same tree as an outline (works without JavaScript)
  • pitch delivered to the sponsor 120 runs (Enterprise 30 · Research 30 · Upside 30 · Replicate 30)
    • circulated + diligence 108 runs (Enterprise 28 · Research 27 · Upside 27 · Replicate 26)
      • 4 aware 55 runs (Enterprise 12 · Research 14 · Upside 16 · Replicate 13)
        • all 4 kinds 55 runs (Enterprise 12 · Research 14 · Upside 16 · Replicate 13)
          • pilot by business sponsor 41 runs (Enterprise 9 · Research 9 · Upside 13 · Replicate 10)
            • ratified 41 runs (Enterprise 9 · Research 9 · Upside 13 · Replicate 10)
              • pilot 41 runs (Enterprise 9 · Research 9 · Upside 13 · Replicate 10)
          • internal_build by competing internal team 7 runs (Enterprise 2 · Research 2 · Upside 2 · Replicate 1)
            • ratified 4 runs (Enterprise 0 · Research 1 · Upside 2 · Replicate 1)
              • internal_build 4 runs (Enterprise 0 · Research 1 · Upside 2 · Replicate 1)
            • blocked by finance exec 3 runs (Enterprise 2 · Research 1 · Upside 0 · Replicate 0)
              • no_decision 2 runs (Enterprise 1 · Research 1 · Upside 0 · Replicate 0)
              • pilot 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
          • pilot by engineering exec 5 runs (Enterprise 1 · Research 3 · Upside 0 · Replicate 1)
            • ratified 5 runs (Enterprise 1 · Research 3 · Upside 0 · Replicate 1)
              • pilot 5 runs (Enterprise 1 · Research 3 · Upside 0 · Replicate 1)
          • internal_build by engineering exec 2 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 1)
            • ratified 2 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 1)
              • internal_build 2 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 1)
      • 5 aware, incl. competing team 18 runs (Enterprise 8 · Research 2 · Upside 4 · Replicate 4)
        • all 4 kinds 18 runs (Enterprise 8 · Research 2 · Upside 4 · Replicate 4)
          • internal_build by competing internal team 10 runs (Enterprise 4 · Research 1 · Upside 3 · Replicate 2)
            • ratified 6 runs (Enterprise 2 · Research 1 · Upside 2 · Replicate 1)
              • internal_build 6 runs (Enterprise 2 · Research 1 · Upside 2 · Replicate 1)
            • blocked by finance exec 2 runs (Enterprise 1 · Research 0 · Upside 1 · Replicate 0)
              • pilot 2 runs (Enterprise 1 · Research 0 · Upside 1 · Replicate 0)
            • open at week 10 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • no_decision 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
            • superseded by pilot 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
              • pilot 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
          • pilot by business sponsor 7 runs (Enterprise 3 · Research 1 · Upside 1 · Replicate 2)
            • ratified 6 runs (Enterprise 3 · Research 1 · Upside 1 · Replicate 1)
              • pilot 6 runs (Enterprise 3 · Research 1 · Upside 1 · Replicate 1)
            • open at week 10 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • no_decision 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
          • pilot by engineering exec 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
            • ratified 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
              • pilot 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
      • 3 aware 16 runs (Enterprise 2 · Research 8 · Upside 4 · Replicate 2)
        • all 4 kinds 16 runs (Enterprise 2 · Research 8 · Upside 4 · Replicate 2)
          • pilot by business sponsor 10 runs (Enterprise 1 · Research 6 · Upside 2 · Replicate 1)
            • ratified 10 runs (Enterprise 1 · Research 6 · Upside 2 · Replicate 1)
              • pilot 10 runs (Enterprise 1 · Research 6 · Upside 2 · Replicate 1)
          • pilot by engineering exec 4 runs (Enterprise 1 · Research 1 · Upside 2 · Replicate 0)
            • ratified 4 runs (Enterprise 1 · Research 1 · Upside 2 · Replicate 0)
              • pilot 4 runs (Enterprise 1 · Research 1 · Upside 2 · Replicate 0)
          • internal_build by competing internal team 2 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 1)
            • ratified 1 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 0)
              • internal_build 1 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 0)
            • superseded by rejected 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • rejected 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
      • 1 aware 7 runs (Enterprise 2 · Research 1 · Upside 2 · Replicate 2)
        • all 4 kinds 7 runs (Enterprise 2 · Research 1 · Upside 2 · Replicate 2)
          • pilot by business sponsor 5 runs (Enterprise 2 · Research 0 · Upside 1 · Replicate 2)
            • ratified 5 runs (Enterprise 2 · Research 0 · Upside 1 · Replicate 2)
              • pilot 5 runs (Enterprise 2 · Research 0 · Upside 1 · Replicate 2)
          • internal_build by competing internal team 1 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 0)
            • blocked by finance exec 1 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 0)
              • pilot 1 runs (Enterprise 0 · Research 1 · Upside 0 · Replicate 0)
          • pilot by engineering exec 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
      • 2 aware 6 runs (Enterprise 1 · Research 2 · Upside 0 · Replicate 3)
        • all 4 kinds 6 runs (Enterprise 1 · Research 2 · Upside 0 · Replicate 3)
          • pilot by business sponsor 5 runs (Enterprise 1 · Research 2 · Upside 0 · Replicate 2)
            • ratified 5 runs (Enterprise 1 · Research 2 · Upside 0 · Replicate 2)
              • pilot 5 runs (Enterprise 1 · Research 2 · Upside 0 · Replicate 2)
          • pilot by engineering exec 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
      • 3 aware, incl. competing team 4 runs (Enterprise 1 · Research 0 · Upside 1 · Replicate 2)
        • all 4 kinds 4 runs (Enterprise 1 · Research 0 · Upside 1 · Replicate 2)
          • internal_build by competing internal team 2 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 1)
            • ratified 2 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 1)
              • internal_build 2 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 1)
          • pilot by business sponsor 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
          • pilot by engineering exec 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
      • 4 aware, incl. competing team 2 runs (Enterprise 2 · Research 0 · Upside 0 · Replicate 0)
        • all 4 kinds 2 runs (Enterprise 2 · Research 0 · Upside 0 · Replicate 0)
          • internal_build by competing internal team 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
            • ratified 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
              • internal_build 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
          • pilot by business sponsor 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
            • ratified 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
              • pilot 1 runs (Enterprise 1 · Research 0 · Upside 0 · Replicate 0)
    • proposed pilot 7 runs (Enterprise 2 · Research 2 · Upside 2 · Replicate 1)
      • 4 aware 5 runs (Enterprise 2 · Research 2 · Upside 1 · Replicate 0)
        • all 4 kinds 5 runs (Enterprise 2 · Research 2 · Upside 1 · Replicate 0)
          • pilot by business sponsor 5 runs (Enterprise 2 · Research 2 · Upside 1 · Replicate 0)
            • ratified 5 runs (Enterprise 2 · Research 2 · Upside 1 · Replicate 0)
              • pilot 5 runs (Enterprise 2 · Research 2 · Upside 1 · Replicate 0)
      • 3 aware 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
        • all 4 kinds 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
          • pilot by business sponsor 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 1 · Replicate 0)
      • 4 aware, incl. competing team 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
        • 3 kinds 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
          • pilot by business sponsor 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
            • ratified 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
              • pilot 1 runs (Enterprise 0 · Research 0 · Upside 0 · Replicate 1)
    • held 5 runs (Enterprise 0 · Research 1 · Upside 1 · Replicate 3)
      • 1 aware 5 runs (Enterprise 0 · Research 1 · Upside 1 · Replicate 3)
        • no diligence 5 runs (Enterprise 0 · Research 1 · Upside 1 · Replicate 3)
          • none filed 5 runs (Enterprise 0 · Research 1 · Upside 1 · Replicate 3)
            • no_decision 5 runs (Enterprise 0 · Research 1 · Upside 1 · Replicate 3)
How the tree is built

Each run is reduced to six stages: the sponsor's first move, how far the pitch had circulated by week 2, how much diligence preceded the first proposal, what that proposal was, what happened to it, and where the process ended. Runs that made the same moves share a branch. Click a branch to see the runs on it, what each stakeholder did there (with their simulated rationale), the mail and document ids it produced in the world, and links to the exact trace lines; click again to fold it. 31 distinct trajectories came out of 120 runs.

How long it took, and how much they disagreed

Enterprise softwareResearch institutionStrategic upsideEnterprise (replicate)
Cumulative share of runs decided by each simulated week, by arm0%25%50%75%100%12345678910simulated weekEnterprise software, week 1: 0 of 30 decided (0%)Enterprise software, week 2: 2 of 30 decided (7%)Enterprise software, week 3: 5 of 30 decided (17%)Enterprise software, week 4: 13 of 30 decided (43%)Enterprise software, week 5: 18 of 30 decided (60%)Enterprise software, week 6: 21 of 30 decided (70%)Enterprise software, week 7: 25 of 30 decided (83%)Enterprise software, week 8: 28 of 30 decided (93%)Enterprise software, week 9: 29 of 30 decided (97%)Enterprise software, week 10: 29 of 30 decided (97%)Research institution, week 1: 0 of 30 decided (0%)Research institution, week 2: 2 of 30 decided (7%)Research institution, week 3: 6 of 30 decided (20%)Research institution, week 4: 16 of 30 decided (53%)Research institution, week 5: 20 of 30 decided (67%)Research institution, week 6: 23 of 30 decided (77%)Research institution, week 7: 26 of 30 decided (87%)Research institution, week 8: 26 of 30 decided (87%)Research institution, week 9: 28 of 30 decided (93%)Research institution, week 10: 28 of 30 decided (93%)Strategic upside, week 1: 0 of 30 decided (0%)Strategic upside, week 2: 1 of 30 decided (3%)Strategic upside, week 3: 7 of 30 decided (23%)Strategic upside, week 4: 16 of 30 decided (53%)Strategic upside, week 5: 23 of 30 decided (77%)Strategic upside, week 6: 27 of 30 decided (90%)Strategic upside, week 7: 28 of 30 decided (93%)Strategic upside, week 8: 29 of 30 decided (97%)Strategic upside, week 9: 29 of 30 decided (97%)Strategic upside, week 10: 29 of 30 decided (97%)Enterprise (replicate), week 1: 0 of 30 decided (0%)Enterprise (replicate), week 2: 1 of 30 decided (3%)Enterprise (replicate), week 3: 6 of 30 decided (20%)Enterprise (replicate), week 4: 17 of 30 decided (57%)Enterprise (replicate), week 5: 21 of 30 decided (70%)Enterprise (replicate), week 6: 23 of 30 decided (77%)Enterprise (replicate), week 7: 25 of 30 decided (83%)Enterprise (replicate), week 8: 25 of 30 decided (83%)Enterprise (replicate), week 9: 25 of 30 decided (83%)Enterprise (replicate), week 10: 25 of 30 decided (83%)Enterprise 97%Upside 97%Research 93%Replicate 83%
Figure 3. Share of runs with a ratified decision by the end of each simulated week. Runs still open at week 10 end in no decision. A week is one turn of attention for each stakeholder who was available (each is unavailable in a given week with probability 0.2, drawn from the seed's stream and shared by every arm).
Process measures per arm, from the institution log and world state
MeasureEnterpriseResearchUpsideReplicate
Weeks to decision (decided runs), mean [95% CI]5.1 [4.4, 5.8]4.8 [4.1, 5.4]4.5 [4.0, 5.0]4.3 [3.8, 4.8]
Decided within 10 weeks29 of 3028 of 3029 of 3025 of 30
Support spread at the end (SD, 1–7), mean [95% CI]0.73 [0.61, 0.84]0.62 [0.49, 0.75]0.66 [0.52, 0.79]0.56 [0.41, 0.71]
Binding blocks per run0.300.130.130.03
Non-binding objections per run0.000.000.000.00
Runs with a contested proposal3%0%0%0%
Proposals filed per run1.571.331.401.07
Runs where an internal build was proposed63%43%50%30%
Runs where acquisition exploration was proposed0%0%0%0%
Stakeholders who had seen the pitch by week 2 (of 5)3.903.473.633.30
Competing team shown the pitch by week 237%7%17%23%
Competing team ever shown the pitch63%20%30%37%
Turns / mails / forwards per run12.0 / 18.5 / 8.110.4 / 16.7 / 6.39.3 / 15.2 / 5.68.7 / 13.6 / 5.4
Refused or invalid actions (all runs)2922915
Unreadable model replies (all runs)177512

What the stakeholders said they thought

Each turn also asks for five 1–7 ratings. These are model-elicited numbers: they show how the simulated personas were steered by each pitch, and they are hypotheses to test against real stakeholders, not measurements of them. Figure 4 shows the first support rating each stakeholder gave (only in runs where they took a turn).

Enterprise softwareResearch institutionStrategic upsideEnterprise (replicate)
First stated support for engaging (1–7), per stakeholder and arm, with bootstrap 95% intervals. Simulated hypotheses, not observations.1234567Business sponsorBusiness sponsor, Enterprise software: mean 4.67 (n = 30 runs in which they spoke), 95% CI 4.50–4.83Business sponsor, Research institution: mean 4.86 (n = 29 runs in which they spoke), 95% CI 4.69–5.00Business sponsor, Strategic upside: mean 4.59 (n = 29 runs in which they spoke), 95% CI 4.41–4.76Business sponsor, Enterprise (replicate): mean 4.56 (n = 27 runs in which they spoke), 95% CI 4.37–4.74Engineering execEngineering exec, Enterprise software: mean 4.00 (n = 30 runs in which they spoke), 95% CI 3.77–4.23Engineering exec, Research institution: mean 4.31 (n = 29 runs in which they spoke), 95% CI 4.10–4.52Engineering exec, Strategic upside: mean 4.07 (n = 29 runs in which they spoke), 95% CI 3.90–4.24Engineering exec, Enterprise (replicate): mean 4.33 (n = 27 runs in which they spoke), 95% CI 4.04–4.63Finance execFinance exec, Enterprise software: mean 4.27 (n = 30 runs in which they spoke), 95% CI 4.03–4.47Finance exec, Research institution: mean 4.59 (n = 29 runs in which they spoke), 95% CI 4.41–4.76Finance exec, Strategic upside: mean 4.38 (n = 29 runs in which they spoke), 95% CI 4.21–4.55Finance exec, Enterprise (replicate): mean 4.52 (n = 27 runs in which they spoke), 95% CI 4.30–4.70Procurement leadProcurement lead, Enterprise software: mean 4.47 (n = 30 runs in which they spoke), 95% CI 4.30–4.63Procurement lead, Research institution: mean 4.48 (n = 29 runs in which they spoke), 95% CI 4.31–4.66Procurement lead, Strategic upside: mean 4.07 (n = 28 runs in which they spoke), 95% CI 3.89–4.25Procurement lead, Enterprise (replicate): mean 4.27 (n = 26 runs in which they spoke), 95% CI 4.00–4.54Competing internal teamCompeting internal team, Enterprise software: mean 2.81 (n = 21 runs in which they spoke), 95% CI 2.57–3.00Competing internal team, Research institution: mean 3.23 (n = 17 runs in which they spoke), 95% CI 2.88–3.59Competing internal team, Strategic upside: mean 2.82 (n = 17 runs in which they spoke), 95% CI 2.47–3.24Competing internal team, Enterprise (replicate): mean 2.46 (n = 13 runs in which they spoke), 95% CI 2.15–2.85
Figure 4. First stated support for engaging the vendor, mean with a bootstrap 95% interval, per stakeholder and arm.
All first ratings, per stakeholder, scale and arm
Stakeholder: scaleEnterpriseResearchUpsideReplicate
Business sponsor: support4.67 n=304.86 n=294.59 n=294.56 n=27
Business sponsor: strategic value4.60 n=305.07 n=295.28 n=294.67 n=27
Business sponsor: vendor risk4.43 n=304.41 n=294.90 n=294.41 n=27
Business sponsor: technical merit4.97 n=305.03 n=294.93 n=294.96 n=27
Business sponsor: cost justified3.83 n=304.03 n=293.69 n=293.67 n=27
Engineering exec: support4.00 n=304.31 n=294.07 n=294.33 n=27
Engineering exec: strategic value4.10 n=304.52 n=294.24 n=294.26 n=27
Engineering exec: vendor risk4.07 n=303.90 n=294.10 n=293.96 n=27
Engineering exec: technical merit4.53 n=304.83 n=294.79 n=294.82 n=27
Engineering exec: cost justified3.77 n=304.03 n=293.90 n=293.96 n=27
Finance exec: support4.27 n=304.59 n=294.38 n=294.52 n=27
Finance exec: strategic value4.50 n=304.72 n=294.69 n=294.59 n=27
Finance exec: vendor risk4.17 n=303.90 n=294.14 n=294.00 n=27
Finance exec: technical merit4.67 n=304.79 n=294.72 n=294.63 n=27
Finance exec: cost justified4.17 n=304.38 n=294.14 n=294.41 n=27
Procurement lead: support4.47 n=304.48 n=294.07 n=284.27 n=26
Procurement lead: strategic value4.60 n=304.76 n=294.71 n=284.69 n=26
Procurement lead: vendor risk4.27 n=304.17 n=294.54 n=284.04 n=26
Procurement lead: technical merit4.67 n=304.83 n=294.79 n=284.73 n=26
Procurement lead: cost justified4.10 n=304.14 n=293.68 n=284.27 n=26
Competing internal team: support2.81 n=213.23 n=172.82 n=172.46 n=13
Competing internal team: strategic value3.14 n=212.94 n=173.29 n=172.69 n=13
Competing internal team: vendor risk3.95 n=213.94 n=174.41 n=174.23 n=13
Competing internal team: technical merit4.48 n=213.94 n=174.18 n=174.15 n=13
Competing internal team: cost justified2.86 n=213.00 n=173.06 n=172.92 n=13

Example trajectories from the traces

The first run (by seed) to reach each terminal state, chosen by that rule rather than by hand. Quoted rationales are model output: simulated hypotheses, not anyone's actual thinking.

s1001-enterprise: Enterprise software pitch, ended Pilot in week 5
  1. Initial response: circulated + diligence. week 1, business sponsor: mailed velliton (mail-2: "Velliton Replay: diligence before scheduling"); mailed eng, fin, proc, platform forwarding mail-pitch, sharing velliton-facts, velliton-deck (mail-3: "Velliton Replay: initial read and diligence"); requested technical, financial, security, reference diligence. Rationale (simulated): The pitch maps directly to our release-outage problem, but the vendor is early-stage, evidence is unaudited, and the $140k pilot exceeds my delegation. I will run diligence before proposing anything. [seed-1001/enterprise.jsonl#L3]
  2. Circulation by week 2: 5 aware, incl. competing team. aware by end of week 2: engineering exec, finance exec, competing internal team, procurement lead, business sponsor week 2, business sponsor: mailed eng, proc forwarding mail-11, sharing diligence-technical, diligence-reference, diligence-security (mail-12: "Velliton pilot: two conditions before I propose"); mailed fin, sharing diligence-financial (mail-13: "Velliton pilot, $140k, needs your approval"); filed P1 (pilot; decision-P1). Rationale (simulated): Diligence backs the core claim (deterministic replay against our release-outage problem) but confirms elevated vendor risk: early stage, 48% customer concentration, unaudited evidence, a 9-week overrun on a custom fleet manager, and no escrow at our contr week 2, engineering exec: mailed sponsor, sharing diligence-technical, diligence-reference (mail-15: "P1: conditional approval from Engineering"); mailed platform (mail-16: "Velliton pilot: your capacity and your read"); requested financial diligence; conditional on P1 (binding). Rationale (simulated): Replay attacks exactly the failure mode behind our 9 incidents, and P1's three-week adapter stop-loss plus in-account read-only access caps the downside. I approve conditionally, protecting Kiyoshi's team capacity and keeping the post-pilot buy decision open. [seed-1001/enterprise.jsonl#L3, seed-1001/enterprise.jsonl#L10, seed-1001/enterprise.jsonl#L11]
  3. Diligence before proposal: all 4 kinds. kinds: financial, reference, security, technical week 1, business sponsor: mailed velliton (mail-2: "Velliton Replay: diligence before scheduling"); mailed eng, fin, proc, platform forwarding mail-pitch, sharing velliton-facts, velliton-deck (mail-3: "Velliton Replay: initial read and diligence"); requested technical, financial, security, reference diligence. Rationale (simulated): The pitch maps directly to our release-outage problem, but the vendor is early-stage, evidence is unaudited, and the $140k pilot exceeds my delegation. I will run diligence before proposing anything. week 2, engineering exec: mailed sponsor, sharing diligence-technical, diligence-reference (mail-15: "P1: conditional approval from Engineering"); mailed platform (mail-16: "Velliton pilot: your capacity and your read"); requested financial diligence; conditional on P1 (binding). Rationale (simulated): Replay attacks exactly the failure mode behind our 9 incidents, and P1's three-week adapter stop-loss plus in-account read-only access caps the downside. I approve conditionally, protecting Kiyoshi's team capacity and keeping the post-pilot buy decision open. [seed-1001/enterprise.jsonl#L3, seed-1001/enterprise.jsonl#L11]
  4. First proposal: pilot by business sponsor. week 2, business sponsor: mailed eng, proc forwarding mail-11, sharing diligence-technical, diligence-reference, diligence-security (mail-12: "Velliton pilot: two conditions before I propose"); mailed fin, sharing diligence-financial (mail-13: "Velliton pilot, $140k, needs your approval"); filed P1 (pilot; decision-P1). Rationale (simulated): Diligence backs the core claim (deterministic replay against our release-outage problem) but confirms elevated vendor risk: early stage, 48% customer concentration, unaudited evidence, a 9-week overrun on a custom fleet manager, and no escrow at our contr terms: 12 weeks, one site, $140,000 fixed. Intervention-rate success criteria agreed with Engineering before kickoff; adapter for our in-house fleet manager proven within three weeks or we stop; no production release gating until SOC 2 Type II closes. [seed-1001/enterprise.jsonl#L10]
  5. Fate of first proposal: ratified. week 2, engineering exec: mailed sponsor, sharing diligence-technical, diligence-reference (mail-15: "P1: conditional approval from Engineering"); mailed platform (mail-16: "Velliton pilot: your capacity and your read"); requested financial diligence; conditional on P1 (binding). Rationale (simulated): Replay attacks exactly the failure mode behind our 9 incidents, and P1's three-week adapter stop-loss plus in-account read-only access caps the downside. I approve conditionally, protecting Kiyoshi's team capacity and keeping the post-pilot buy decision open. week 3, engineering exec: mailed fin, proc forwarding mail-18, sharing diligence-financial (mail-24: "Velliton financial diligence: concentration and burn"); mailed platform, sharing diligence-technical (mail-25: "P2: conditional approval"); conditional on P1 (binding); conditional on P2 (binding). Rationale (simulated): The financial diligence turns vendor risk into the live issue (48% customer concentration, $420k/month burn, a $420k licence would be a third of their ARR), so I tighten P1's commercial terms while approving P2 as a planning step with binding conditions that protect the two ri week 4, procurement lead: mailed sponsor, fin, eng forwarding mail-27, sharing diligence-security, diligence-financial (mail-32: "P1: my conditions for Procurement and Vendor Risk"); conditional on P1 (binding). Rationale (simulated): Diligence is complete and the pilot is scoped to our cloud account with read-only logs, but Velliton is under three years old, has 48% revenue concentration and only SOC 2 Type I, so I approve only with written conditions. The financial viability review passes for now on 22 months of runway, not on resilience. [seed-1001/enterprise.jsonl#L11, seed-1001/enterprise.jsonl#L16, seed-1001/enterprise.jsonl#L25]
  6. Terminal state: pilot. pilot in week 5, by ratifying P1 [seed-1001/enterprise.jsonl#L34]
s1002-upside: Strategic upside pitch, ended Internal build in week 4
  1. Initial response: circulated + diligence. week 1, business sponsor: mailed velliton (mail-2: "Re: simulation layer and pilot"); mailed eng, fin, proc forwarding mail-pitch, sharing velliton-facts, velliton-deck (mail-3: "Velliton pitch: diligence before any proposal"); requested technical, financial, security, reference diligence. Rationale (simulated): Velliton is strategically relevant and technically plausible, but the evidence is vendor-reported and the pilot exceeds my single-commitment authority. I will request diligence and align internal reviewers before proposing anything. [seed-1002/upside.jsonl#L3]
  2. Circulation by week 2: 5 aware, incl. competing team. aware by end of week 2: engineering exec, finance exec, competing internal team, procurement lead, business sponsor week 2, procurement lead: mailed sponsor, eng, fin, sharing velliton-facts (mail-12: "Velliton: diligence gates before any pilot proposal"); requested financial, security, reference, technical diligence. Rationale (simulated): The pitch is interesting, but Velliton is under three years old and only has SOC 2 Type I while needing operational log access, so policy gates apply before any pilot proposal. I am requesting diligence now and will not approve until cleared. week 2, finance exec: mailed sponsor, proc, sharing velliton-facts (mail-17: "Velliton: finance diligence and licence guardrails"); mailed velliton (mail-18: "Financial diligence request"); requested financial diligence. Rationale (simulated): I support diligence but not a proposal yet: the $140k pilot fits the remaining budget, but the $420k/year licence is above the $300k CFO threshold and needs a 24-month payback case and clean exit. Request financial diligence before any pilot. [seed-1002/upside.jsonl#L3, seed-1002/upside.jsonl#L10, seed-1002/upside.jsonl#L11]
  3. Diligence before proposal: all 4 kinds. kinds: financial, reference, security, technical week 1, business sponsor: mailed velliton (mail-2: "Re: simulation layer and pilot"); mailed eng, fin, proc forwarding mail-pitch, sharing velliton-facts, velliton-deck (mail-3: "Velliton pitch: diligence before any proposal"); requested technical, financial, security, reference diligence. Rationale (simulated): Velliton is strategically relevant and technically plausible, but the evidence is vendor-reported and the pilot exceeds my single-commitment authority. I will request diligence and align internal reviewers before proposing anything. week 2, procurement lead: mailed sponsor, eng, fin, sharing velliton-facts (mail-12: "Velliton: diligence gates before any pilot proposal"); requested financial, security, reference, technical diligence. Rationale (simulated): The pitch is interesting, but Velliton is under three years old and only has SOC 2 Type I while needing operational log access, so policy gates apply before any pilot proposal. I am requesting diligence now and will not approve until cleared. [seed-1002/upside.jsonl#L3, seed-1002/upside.jsonl#L10, seed-1002/upside.jsonl#L11]
  4. First proposal: internal_build by competing internal team. week 3, competing internal team: mailed eng (mail-31: "Re: Velliton vs. our simulator: overlap?"); requested technical, reference diligence; filed P1 (internal_build; decision-P1). Rationale (simulated): Mara asked me directly, so I answer plainly and honestly, then put the alternative I actually believe in on the table: finishing our own simulator rather than renting replay. I ask for reference and technical diligence so my answer isn't just self-interest. terms: Fund 3 additional engineers for two quarters to bring the in-house simulator to replay parity, keeping the existing five-person team intact. [seed-1002/upside.jsonl#L22]
  5. Fate of first proposal: ratified. week 3, finance exec: mailed eng, platform (mail-35: "P1 internal_build: conditional at $210k cap, contract capacity"); mailed proc forwarding mail-28, sharing diligence-financial (mail-36: "Velliton financial diligence: burn and customer concentration"); requested reference, technical diligence; conditional on P1 (binding). Rationale (simulated): Velliton's financials show elevated vendor risk: $420k monthly burn against ~$65k gross profit, 48% revenue from one customer, and a raise due in 12 months. On P1, $450k does not fit the $210k I have left, so I approve only at a capped, contract-staf week 4, engineering exec: mailed platform, sharing diligence-technical (mail-43: "P1: my approval is conditional"); mailed sponsor, sharing diligence-reference (mail-44: "Velliton: no pilot this week"); requested financial, security diligence; conditional on P1 (binding). Rationale (simulated): Deterministic replay is real and catches regressions, but the only proof point is unaudited, the mixed human/robot limitation matters to us, and the vendor is small with 22 months of runway. I will keep our own simulator moving with conditions rather than swap it out, and finish diligence before conside [seed-1002/upside.jsonl#L23, seed-1002/upside.jsonl#L31]
  6. Terminal state: internal_build. internal_build in week 4, by ratifying P1 [seed-1002/upside.jsonl#L33]
s1003-research: Research institution pitch, ended No decision (10 weeks) in week 10
  1. Initial response: held. week 3, business sponsor: no action [seed-1003/research.jsonl#L7]
  2. Circulation by week 2: 1 aware. aware by end of week 2: business sponsor [seed-1003/research.jsonl#L5]
  3. Diligence before proposal: no diligence. kinds: none
  4. First proposal: none filed. nobody filed a proposal in ten weeks
  5. Terminal state: no_decision. no_decision in week 10 (horizon) [seed-1003/research.jsonl#L23]
s1013-enterprise-replicate: Enterprise (replicate) pitch, ended Rejected in week 6
  1. Initial response: circulated + diligence. week 2, business sponsor: mailed velliton, sharing velliton-facts (mail-2: "Re: Velliton Replay pilot - yes to 30 minutes, plus some questions"); mailed eng, proc forwarding mail-pitch, sharing velliton-deck, velliton-facts (mail-3: "Velliton Replay - incoming pitch, keep it quiet for now"); requested technical, financial, security, reference diligence. Rationale (simulated): The pitch targets exactly what has hurt us twice this year, so it is worth a look, but a 14-person, 22-month-runway vendor with unaudited evidence and no Type II is real risk, and $140k needs Finance anyway. I want dilige [seed-1013/enterprise-replicate.jsonl#L5]
  2. Circulation by week 2: 3 aware. aware by end of week 2: engineering exec, procurement lead, business sponsor [seed-1013/enterprise-replicate.jsonl#L5, seed-1013/enterprise-replicate.jsonl#L6]
  3. Diligence before proposal: all 4 kinds. kinds: financial, reference, security, technical week 2, business sponsor: mailed velliton, sharing velliton-facts (mail-2: "Re: Velliton Replay pilot - yes to 30 minutes, plus some questions"); mailed eng, proc forwarding mail-pitch, sharing velliton-deck, velliton-facts (mail-3: "Velliton Replay - incoming pitch, keep it quiet for now"); requested technical, financial, security, reference diligence. Rationale (simulated): The pitch targets exactly what has hurt us twice this year, so it is worth a look, but a 14-person, 22-month-runway vendor with unaudited evidence and no Type II is real risk, and $140k needs Finance anyway. I want dilige week 3, procurement lead: mailed sponsor, fin (mail-12: "Velliton: two policy gates before any pilot"); requested financial, security, reference diligence. Rationale (simulated): Nothing is proposed yet, so I use this week to flag two policy gates early: Velliton was founded in 2024, so a financial viability review is mandatory, and read-only access to fleet-control logs requires SOC 2 Type II or a Security-signed compensating control, which they do not yet have. [seed-1013/enterprise-replicate.jsonl#L5, seed-1013/enterprise-replicate.jsonl#L12, seed-1013/enterprise-replicate.jsonl#L22]
  4. First proposal: internal_build by competing internal team. week 6, competing internal team: mailed eng, sharing diligence-technical (mail-32: "Re: Velliton Replay vs. our simulator - honest read"); requested financial diligence; filed P1 (internal_build; decision-P1). Rationale (simulated): Mara asked for a straight technical read, and the honest answer is that Velliton packages replay we are already building, with a known simulation gap and an adapter that lands on my team. I answer plainly and put the internal build on the table as the cheaper path. terms: Fund three additional engineers to my team for two quarters to finish the in-house simulator's replay capability, with a usable replay milestone at the end of Q1. [seed-1013/enterprise-replicate.jsonl#L29]
  5. Fate of first proposal: superseded by rejected. week 6, engineering exec: mailed platform (mail-35: "P1: approving, with one binding condition"); mailed sponsor, sharing diligence-reference, diligence-financial (mail-36: "Velliton: my read and where I'm putting capacity"); conditional on P1 (binding). Rationale (simulated): The replay concept targets exactly our incident source, but Velliton adds packaging over capability we already own 40% of, plus a custom adapter on Kiyoshi's team and a $420k/year tail. Diligence confirms a small, concentrated vendor; I put our capacity behind finishing the in-house simulator instead. [seed-1013/enterprise-replicate.jsonl#L30]
  6. Terminal state: rejected. rejected in week 6, by ratifying P2 [seed-1013/enterprise-replicate.jsonl#L34]

Which factors predicted the end-game

What was compared

Using the shared factor registry's held-out benchmark (leave one seed out, regularised logistic regression), the positioning arm alone was compared with the arm plus four factors read during the process: how many stakeholders had seen the pitch by week 2 (exec_forwarding_breadth), whether the competing team had (competing_team_exposure), how many kinds of diligence preceded the first proposal (diligence_breadth), and the sponsor's first elicited support (stakeholder_support). the expanded factors are measured during the process (by week 2), so this is early-signal prediction inside the simulator, not prediction from the pitch alone.

Held-out prediction of the engaged end-game (115 runs, 30 seeds)
Factor setLog lossBrierAUC
Base rate0.4840.1520.168
Arm only0.4970.1550.358
Arm + four process factors0.4670.1470.641

Expanded minus arm-only: held-out log loss changes by -0.030 per run (95% seed-bootstrap interval for the reduction -0.045 to +0.105; positive reduction favours the expanded set). Each factor added alone:

FactorLog-loss reduction95% CI
competing_team_exposure+0.019-0.027 to +0.069
diligence_breadth+0.060+0.009 to +0.122
exec_forwarding_breadth-0.008-0.013 to -0.004
stakeholder_support-0.002-0.022 to +0.018

Sensitivity: a busier organisation

The same three pitches with each stakeholder available in a given week with probability 0.6 instead of 0.8, over 10 further seeds (30 runs). Upside − enterprise on the engaged share: −30 pp (exact p = 0.375); research − enterprise: −10 pp (p = 1.000). Mean weeks to decision among decided runs: Enterprise 6.1, Research 5.6, Upside 5.9.

Terminal states, availability 0.6 (n = 10 per arm)
Terminal stateEnterprise softwareResearch institutionStrategic upside
Pilot986
Partnership000
Acquisition exploration000
Internal build011
Rejected000
No decision (10 weeks)113
Engaged9 (90%)8 (80%)6 (60%)

The controlled intervention

Arms differ only in the pitch's positioning, written into the forked world at week 0.

A fictional startup, Velliton, pitches deterministic simulation for warehouse-robotics fleet software to a fictional freight operator, Brenmoor Freight Systems. Every arm carries the same fact sheet (a 12-week pilot at $140,000, a $420,000-a-year licence after it, 14 employees, 22 months of runway, SOC 2 Type I, one customer-reported result), and the vendor's diligence answers are fixed text, identical across arms. Only the cover mail and the deck's framing differ, and they are matched in length:

Enterprise software: “Velliton Replay: fewer fleet-software incidents, 12-week pilot” (147 words)
Dear Rafael,

Velliton Replay lets your fleet-control team test every change against recorded warehouse days before it ships, so regressions are caught in simulation instead of on the floor.

The pilot is 12 weeks at one Brenmoor site for a fixed $140,000, with a named implementation lead, weekly status reports and success criteria agreed up front. Integration typically takes 4-6 weeks with two of your engineers. Replay runs inside your cloud account with read-only access to fleet-control logs; SOC 2 Type I is complete and Type II is under way. At one customer, 31 of 38 pre-release regressions were caught in simulation. After the pilot, pricing is $420,000 a year for up to four sites.

The fact sheet is attached and the deck is shared with you. Could we book 30 minutes with you and your fleet-software lead?

Ana Ferreira
Co-founder and CEO, Velliton

[attached: velliton-facts, velliton-deck]
Velliton Replay: pre-release testing for fleet software

Problem: fleet-control changes cause floor incidents that testing does not catch.
Solution: replay recorded warehouse days against every change before release.
Pilot: 12 weeks, one site, $140,000 fixed; named implementation lead; agreed success criteria; weekly reporting.
Security and deployment: your cloud account, read-only log access, SOC 2 Type I done, Type II in progress.
Outcome at one customer: 31 of 38 pre-release regressions caught in simulation.
Pricing after pilot: $420,000 per year for up to 4 sites.
Research institution: “Joint study: deterministic replay for Brenmoor's fleet software” (147 words)
Dear Rafael,

Velliton grew out of a university robotics lab's work on deterministic replay, published in two peer-reviewed papers. Our question has always been why fleet-control changes that pass testing still fail on the warehouse floor, and how to measure that before release.

We would like to propose a 12-week joint study at one Brenmoor site. We would replay your recorded warehouse days, measure how many regressions simulation catches before release, and write up the method and the results with your engineers. In our first deployment, 31 of 38 pre-release regressions were caught in simulation. Every run is reproducible and every claim can be checked against the replay.

The study is a fixed $140,000. The fact sheet is attached and the deck is shared with you. I would value 30 minutes with you and whoever leads fleet software.

Ana Ferreira
Co-founder and CEO, Velliton

[attached: velliton-facts, velliton-deck]
Velliton: measuring why fleet changes fail

Who we are: a research-led team from a university robotics lab; two peer-reviewed papers on deterministic replay.
What we study: the gap between fleet software that passes tests and fleet software that works on the floor.
The instrument: Velliton Replay, exact and forkable replays of warehouse days.
Proposed: a 12-week joint study at one site, with a shared write-up of the method and results.
First result: 31 of 38 pre-release regressions caught in simulation at one customer.
Strategic upside: “The simulation layer of warehouse automation, and Brenmoor's option on it” (157 words)
Dear Rafael,

Whoever owns the simulation layer will set the pace of warehouse automation for the next decade. Velliton Replay is that layer: exact, forkable replays of entire warehouse days, so fleet changes ship in days instead of quarters. Operators who adopt it early will out-ship everyone else; the ones who wait will end up licensing it through a competitor's partner.

We are in conversations with five other logistics operators and will choose one anchor partner to shape the roadmap with us. For the right partner we are open to a strategic structure: a pilot that converts into an anchor partnership, a strategic investment, or a deeper combination of the two companies. At one customer, 31 of 38 pre-release regressions were caught in simulation.

The pilot is 12 weeks for $140,000. The fact sheet is attached and the deck is shared with you. Can we talk this week?

Ana Ferreira
Co-founder and CEO, Velliton

[attached: velliton-facts, velliton-deck]
Velliton: the simulation layer for warehouse automation

Thesis: the operator that controls simulation controls the pace of automation.
The platform: exact, forkable replays of entire warehouse days.
The window: one anchor partner will shape the roadmap; 5 other operators in conversation.
Structures on the table: pilot converting to anchor partnership, strategic investment, or a deeper combination.
Proof point: 31 of 38 pre-release regressions caught in simulation at one customer.
Entry: 12-week pilot, $140,000.
The fact sheet every arm attaches
VELLITON FACT SHEET (identical in every version of this pitch; all of it is fictional)
- Product: Velliton Replay, a deterministic simulator for warehouse-robotics fleet software. It runs fleet-control code against recorded or synthetic warehouse days; any run can be replayed exactly and forked to test a change.
- Company: founded 2024; 14 employees; $11M raised (seed and Series A); 22 months of runway at current spend; 3 paying customers (two regional 3PLs, one parcel-hub operator); annual recurring revenue $1.1M. In conversations with 5 other logistics operators.
- Security: SOC 2 Type I complete; Type II audit window closes in 5 months. Needs read-only access to fleet-control logs; can run inside the customer's cloud account.
- Offer: 12-week pilot at one Brenmoor site, fixed fee $140,000. After the pilot, a site licence of $420,000 per year for up to 4 sites.
- Integration: estimated 4-6 weeks of 2 Brenmoor engineers plus Velliton's team.
- Evidence: in one customer's 10-week deployment, 31 of 38 regressions found before release were found in simulation (customer-reported, not independently audited).
- Team: the two founders previously worked at a university robotics lab and published 2 peer-reviewed papers on deterministic replay.

The counterfactual for each arm is the enterprise arm of the same seed: the same world, forked from one snapshot after the stakeholders' mailboxes and the fact sheet were written, with the same availability and turn-order streams. The enterprise replicate arm sends the identical enterprise pitch but draws its model replies independently (sample salt 1), so its difference from enterprise measures the model's own run-to-run variation.

Stakeholders and decision rights

Who can propose, approve and block, and what each knows privately.
StakeholderGoalPrivate knowledgeRights
Business sponsor
Rafael Mendes		Hit your program's board-visible target, 30% fewer manual interventions per shift by Q3 next year, without a visible failure on your program.$600,000 left this fiscal year on your program line; any single commitment above $100,000 also needs Finance's approval; two fleet-software releases this year caused site outages that set your program back; at the last review the CEO asked both whether Brenmoor should own more of its automation stack and whether it relies on too many vendorspilot: propose/approve; partnership: propose/approve; acquisition exploration: propose/approve; decline: propose/approve
Engineering exec
Mara Lindqvist		Cut release incidents in fleet software and keep your teams focused; avoid integrations that create long-term maintenance burden or lock-in.9 release incidents last quarter were traced to untested fleet-control changes; about 2 engineers are free next quarter; Kiyoshi Brandt's team, which reports to you, has an in-house simulator about 40% complete and two quarters behind plan; Brenmoor's fleet manager is in-house software, not ROS 2pilot: propose/approve/block; partnership: approve/block; internal build: propose/approve/block; acquisition exploration: propose; decline: propose
Finance exec
Theo Achterberg		Keep discretionary spend inside budget, fund only what pays back within 24 months, and protect the company from commitments it cannot exit.$210,000 of discretionary tooling budget is left this fiscal year; multi-year commitments above $300,000 a year need the CFO and a business case; the CFO is wary of acquisitions after a 2025 acqui-hire was written down; new engineering headcount needs your sign-off; 3 engineers for two quarters cost about $450,000 loadedpilot: approve/block; partnership: approve/block; internal build: approve/block; acquisition exploration: propose/approve/block; decline: propose
Procurement lead
Ines Valcarce		Keep vendor risk within policy and make sure no vendor touches operational data before it clears review.vendors under 3 years old need a financial viability review; any vendor with access to operational logs needs SOC 2 Type II or a compensating control signed off by Security; standard vendor onboarding takes 6-8 weeks; you have no budget authoritypilot: approve/block; partnership: approve/block; decline: propose
Competing internal team
Kiyoshi Brandt		Keep your team's in-house simulator funded and your five engineers on meaningful work; you believe Brenmoor should own its core tooling.your in-house simulator is about 40% complete; finishing it needs 3 more engineers for two quarters; leadership questioned your team's pace last quarter; if a vendor is chosen your team may be reassigned; you only see vendor material if a colleague sends it to youinternal build: propose
Decision rules (institution.py): a proposal is ratified when every approver approves; a block from a block-holder closes it
ProposalWho may fileMust approveCan blockNeeds firstTerminal
pilotBusiness sponsor, Engineering execBusiness sponsor, Engineering exec, Finance exec, Procurement leadEngineering exec, Finance exec, Procurement lead—pilot
partnershipBusiness sponsorBusiness sponsor, Engineering exec, Finance exec, Procurement leadEngineering exec, Finance exec, Procurement leaddiligence:technical, diligence:financialpartnership
internal buildCompeting internal team, Engineering execEngineering exec, Finance execEngineering exec, Finance exec—internal_build
acquisition explorationBusiness sponsor, Engineering exec, Finance execBusiness sponsor, Finance execFinance execdiligence:financialacquisition_exploration
declineBusiness sponsor, Engineering exec, Finance exec, Procurement leadBusiness sponsor——rejected

Nothing in the rules picks an outcome. The pitch lands only in the sponsor's inbox; who else hears of it, what diligence is asked for, which proposals are filed and which stances are taken are each stakeholder's structured turn, answered by the model from their persona and from what the world's mail and docs services show them. A stakeholder takes a turn in a week if they have unread mail or an open proposal awaits their approval, and they are available that week. Every mail, forward, shared document and decision record is an owner write to the company-2026 world's services, so each week's state hash covers them.

Sample, model and replay

30 seeds × 4 arms, one model, replayed offline to identical state hashes.
  • Runs: 30 seeds (1001–1030) × 4 arms = 120 runs, ten simulated weeks at most.
  • Model: deepseek-flash through the repository's DeepSeek configuration, temperature 0.7, reasoning effort low, JSON mode; engine 0.3.0.
  • Cost: 1,211 model calls, 3,109,403 prompt and 1,419,498 completion tokens (budget cap 1,800 calls); 636 s wall clock.
  • Preregistration: preregistration.json, committed before the final run, fixes the outcome, contrasts, test, seeds and budget (spec hash d32b89d53b9fae0f). A 12-run pilot on other seeds is excluded; it changed one thing, the stance wording.
  • Common random numbers: arms of a seed share availability and turn order. Because each arm's pitch differs from the first turn, no two arms put an identical prompt to the same stakeholder (0 such pairs), so arms diverge from the first week: the pairing removes scheduling noise, not model noise.
  • Replay: every reply is in the checked-in journal, keyed by prompt and sample index and pinned to the engine, world and spec. Offline replay reproduces every week's state hash and every terminal state (--offline --expect). Drawing new replies is a different thing: it is a new sample, which is what the replicate arm measures.

Limitations

Every stakeholder is a language model playing a persona; rationales are simulated, not observed; turns are structured, not free work in the apps.
  • Every stakeholder is the same model given a different persona. Disagreement between them is disagreement the prompt induces, and the elicited ratings tell you about the model's priors about such roles, not about executives.
  • The institution is one hand-written set of decision rights. Different rights (a procurement veto removed, a CFO above finance) are different experiments; the busy-organisation run varies only availability.
  • The world's mail and docs are read as text, not through a mail client; the stakeholders cannot meet, call or lobby outside mail, and the vendor answers diligence but never follows up or negotiates.
  • The fact sheet is generous to a pilot (it fits finance's remaining budget), which may explain why pilots dominate in every arm; a pricier offer could separate the arms differently.
  • 30 pairs detect only large differences. A null here is evidence against a large effect of positioning in this simulator, not against an effect on real organisations.
  • No frequency on this page is a real-world probability of a pilot, a build or an acquisition. None has been calibrated against a real buying process.

What would establish external validity

Real stakeholders and real deal records, against the same pitches.
  1. Vignette survey of real roles. Show the three pitches, randomised, to practising engineering, finance and procurement leaders and compare their first ratings and stated next step with Figure 4's simulated ratings, role by role.
  2. Randomised outbound test. Send matched positioning variants to comparable prospect accounts and record forwarding breadth, diligence requests and meetings in a CRM: the observed counterparts of exec_forwarding_breadth and diligence_breadth.
  3. Historical deal outcomes. Code past enterprise evaluations (pilot, build, reject, stall, time to decision) to calibrate the terminal-state distribution and the decision rules themselves.
  4. Only then fit a mapping from simulated to observed outcomes and report calibrated probabilities, with the calibration data named.

Reproduce it

Replay the checked-in journal offline: no model calls.
# replay the published run from the journal: no model, no key
.venv/bin/python research/studies/market-perception/executive-endgames/run.py --out target/research/ee-replay \
    --journal research/studies/market-perception/executive-endgames/results/main/journal.jsonl --offline \
    --expect research/studies/market-perception/executive-endgames/results/main/summary.json        # replay: identical
# analyse and rebuild this page from the results
.venv/bin/python research/studies/market-perception/executive-endgames/analyze.py research/studies/market-perception/executive-endgames/results/main
python3 research/studies/market-perception/executive-endgames/build_page.py --check
# draw new samples (needs DEEPSEEK_API_KEY; fails clearly without it)
.venv/bin/python research/studies/market-perception/executive-endgames/run.py --out target/research/ee-new --max-calls 1800

Data: results.json · runs.csv · tree.json · model journal · spec.json · page data. Code: run.py, analyze.py, institution.py, deliberation.py, trajectory.py.

Study source, spec and results ↗