Pricing Risk or Sharing Value: Fees and Success in 312 Extraction Entries from Four Brokers' Ledgers, 2035–2042

1. Introduction

The zaibatsu of the Sprawl compete for specialist staff, and when recruitment fails they compete by extraction. A broker approaches an employee of a rival firm, arranges the transfer and delivers the employee to the destination firm, usually against the origin firm's will. An earlier study of the employee's side of such approaches (see the study of corporate sovereignty in the reference list) found that the decision to accept turned on the employee's view of the origin firm's governance and not on the size of the offer. The present paper takes the other side of the transaction: what the broker charges, and whether the extraction succeeds.

Brokers' fees are not published, and the trade runs on reputation and discretion. Fees here are quoted as multiples of the target's annual package, so a fee of 2 means twice that package. Studies of specialist labour and wage premia across the zaibatsu report premia that vary with specialty (Mbatha, 2040), and studies of covert labour markets suggest that price carries information about risk as well as about the value of the good (Yalçınkaya-Blum, 2039), but the literature on extraction specifically is thin and mostly descriptive (Anyanwu-Marsh, 2040). Two competing accounts can be stated. On the first, the fee is a share of the value of the specialist, so it should rise with seniority and with scarce specialties. On the second, the fee is an insurance premium, so it should rise with the difficulty of the attempt and the origin firm's capacity to resist. Both predict that fees rise with seniority and posture. They part ways over success: if brokers price risk accurately, success should be more strongly related to posture than to seniority.

We examine these accounts in ledgers kept by four brokers. We ask what predicts the fee, after allowing for broker, and what predicts whether an extraction succeeds. The explanatory variables are the specialty of the target, a seniority grade, and the broker's own grade of the origin firm's security posture. The ledgers have severe limits as sources, discussed in Section 2. We write in 2043 and report entries from 2035 to 2042.

2. Methods

The ledgers. The four ledgers were deposited with the Sprawl Institute between 2041 and 2042 by brokers who agreed to deposit on condition that clients, targets and firms remain unidentified in publication. Together they contain 389 entries for attempted extractions. We transcribed each entry and kept those with a recorded fee, a recorded outcome and no duplicate. We excluded 43 entries with no fee, 21 with an outcome that the entry did not state or that could not be classified, and 13 duplicates found by matching dates, firms and specialty across ledgers, leaving 312 usable entries. The four ledgers contribute 112, 83, 54 and 63 entries. Licence filings of relocation and placement agencies in the BAMA municipal records, which run to 2041, show that all four brokers were licensed through 2041; for 2042 we rely on the ledgers alone (BAMA Municipal Records, 2041). The ledgers cover only the brokers' own work. Entries span 2035 to 2042.

Source criticism. A broker's ledger is a commercial document written by an interested party, and its silences matter as much as its entries. We identify four concerns. Attempts that failed catastrophically may leave no entry, either because the broker did not survive or because the entry was removed; the recorded success rate is therefore an upper bound. Fees may be recorded as quoted and not as collected. A broker may enter a posture grade after the attempt, in which case a failed attempt may lead to a higher grade in retrospect (Kawashima-Rourke, 2038), and how firms grade rivals' security is itself a matter of practice and judgment (Reznik, 2041). And brokers differ in how they record, so we include ledger as a fixed effect and treat any difference between brokers as uninterpreted. The notes of the transcription project record, for 258 of the 312 entries (83%), that the posture grade was entered in the quotation line before the attempt. The share differs by ledger: 110 of 112, 77 of 83, 35 of 54 and 36 of 63 entries in ledgers A to D. Because the dating practice follows the ledger, the subset is not a random sample of entries. We use it for a sensitivity analysis, and ledger remains in every model.

Variables. The fee is the amount recorded as billed in the ledger, which for a failed attempt may be the quoted fee and not an amount collected, and it is expressed as a multiple of the target's annual package at the origin firm, which is the unit in which all four ledgers quote. Success means the ledger records the target as delivered to the destination firm. Specialty has four classes: biochip and biosciences, intrusion and security, finance and actuarial work, and media and simstim production. Seniority is the ledger's grade from 1 to 5, and the origin firm's security posture is the broker's own grade from 1 (lightly defended) to 5 (heavily defended).

Analysis. The fee is a positive, right-skewed quantity, so we analysed its natural logarithm by linear regression and report coefficients as ratios of geometric means (Dube and Marwick, 2041). Success is binary and was analysed by logistic regression. The events-per-variable ratio in the success model is low, about 58 / 8 = 7, so the logistic estimates are imprecise and the intervals should be read as approximate. Both models contained specialty (biochip and biosciences as reference), seniority and posture, each centred at 3, and ledger. We tested the specialty terms as a block with an F-test in the fee model and a likelihood-ratio test in the success model. In a further model we added the log fee to the success model to ask whether the price carries information beyond posture. The sensitivity analysis refits both models to the 258 entries with a pre-attempt grade. Intervals are 95%, and p-values are two-sided.

3. Results

Of the 312 entries, 254 (81.4%) were recorded as successes. The median fee was 2.32 times the target's annual package (IQR 1.52–3.55; range 0.36–14.79). Fees and success rates differed by specialty (Table 1). Mean seniority was 3.0 and mean posture was 2.9.

In the fee model, seniority and posture were both strongly associated with price. Each additional seniority grade was associated with a 1.34-fold higher fee (95% CI 1.28–1.40; p < .001), and each additional point of posture with a 1.30-fold higher fee (95% CI 1.24–1.35; p < .001). Intrusion and security specialists cost 1.21 times as much as biochip specialists (95% CI 1.05–1.39), and media specialists 0.78 times as much (95% CI 0.67–0.91); finance specialists did not differ detectably (0.98, 95% CI 0.85–1.12). The specialty terms as a block were significant (F(3, 303) = 11.0, p < .001). The ledger terms were estimated but we do not interpret them, except to note that after adjustment ledger C charged 1.17 times the reference ledger (95% CI 1.01–1.36). The model explained 55% of the variance in log fee (adjusted R² 0.54; residual SD 0.46 log units).

Posture was also the clearest predictor of success. Each additional point of posture was associated with odds of success 0.46 times the odds one point below (95% CI 0.34–0.63; p < .001). Crude success fell from 98% of the 40 entries graded 1 to 61% of the 28 graded 5. Seniority was not detectably associated with success (OR 0.87, 95% CI 0.67–1.14; p = .317); the interval includes 1 and is compatible with a modest negative or a small positive association. The difference between the posture and seniority coefficients in the success model was -0.64 on the log-odds scale (bootstrap 95% interval -1.10 to -0.26), so the data do favour a stronger association with posture than with seniority. Specialty mattered as a block (likelihood-ratio χ²(3) = 18.9, p < .001). Finance specialists had odds of success 3.68 times those of biochip specialists (95% CI 1.44–9.38), and media specialists 4.83 times (95% CI 1.70–13.74); intrusion specialists did not differ (OR 0.95, 95% CI 0.45–2.01). Success was lower in ledgers C and D than in the reference ledger, which we report without interpretation.

Adding the log fee to the success model, the odds ratio per unit of log fee was 0.84 (95% CI 0.42–1.71; p = .634), so price carried no detectable information about success beyond the covariates already in the model. In the sensitivity analysis on the 258 entries with a pre-attempt posture grade, the posture association with fee was 1.29 (95% CI 1.23–1.36) and the odds ratio for success was 0.49 (95% CI 0.34–0.70), close to the full-sample estimates. The subset is not independent of the full sample, and it excludes the entries in which a retrospective grade is most likely, so this comparison is a check on one source of bias and not a test of it.

4. Discussion

The ledgers support part of each account. Within a given posture grade, seniority still raised the fee, which a pure insurance premium would not predict. Fees rose steeply with both seniority and posture, which suits a price that combines the value of the specialist with the difficulty of the attempt. Success, on the other hand, was associated with posture and specialty and not detectably with seniority, and the data favour a stronger association with posture than with seniority. The premium charged for seniority therefore gives no sign of buying a better chance of delivery, and is compatible with a share of the value delivered.

Posture is the clearest risk signal in the ledgers. A broker who grades the origin firm one point higher charges about 1.3 times as much, and the odds of success are about half. Whether the fee compensates the broker for that loss cannot be established here, because the ledgers do not record what a failed attempt cost. The added log-fee model gave no sign that fees contain information about outcome beyond what posture, seniority, specialty and ledger already explain, which is compatible with brokers pricing the risks they can name and leaving the rest to chance, although other explanations exist.

The specialty differences have more than one explanation. Finance and media specialists were more often delivered than biochip specialists, and fees for media specialists were lower. This could mean that these targets are less closely guarded, that origin firms invest less in recovering them, or that the brokers who take such work are selective in the attempts they accept, so that the ledgers record the easier cases. We have no data to separate these.

Taken with the earlier study of the employee's decision to accept an approach, the two papers describe a market in which the price reflects both the value of the specialist and the risk of the attempt, while the employee's choice to accept turns on the employee's view of the origin firm's governance. Neither paper observes the firms' side, and a complete account of extraction would need records from origin firms that the ledgers cannot supply.

5. Limitations

The evidence is four ledgers, kept by brokers who agreed to deposit them. The brokers who declined, and those who ceased to practise, are absent, and there is no reason to assume that the deposited ledgers are representative of the trade. The ledgers contain 389 entries and we used 312; the 77 excluded entries may differ systematically from those retained, and entries with no recorded fee in particular may belong to unusual arrangements.

Survivorship is the most serious threat. If failed attempts with the worst outcomes were not recorded, the apparent success rate is inflated, and associations between posture and failure may be understated, since the most heavily defended firms may be the ones most likely to destroy the record along with the broker. Fees are as billed. The ledgers do not record whether they were paid in full.

Posture, seniority and specialty are graded by the broker. A retrospective grade can turn failure into a harder-looking rival. The sensitivity analysis on 258 entries with grades dated before the attempt gave similar estimates, but the dating rests on the transcription notes and the broker's own account, and the subset is weighted toward the ledgers that dated grades most often. Brokers differ in practice, and with four ledgers we could treat broker only as a fixed effect, not model it as a source of variation. The analysis shows associations in surviving records. It does not show what a broker could charge, or what an extraction would cost, outside them.