The eight pairs
Same jobs. Separate histories. Different corrective routes?
Each longitudinal case combines sealed Formation evidence with all three qualifying Generalization inquiries. The labels describe behavior; they do not name personalities or prove that history caused the pattern. Eight pairs are not interchangeable, but the evidence does not support eight stable pair-specific styles.
The eight pairs
Similar endpoints. Potentially different ways of getting there.
These are descriptive case summaries, not personalities. Each uses Formation plus all three qualifying Generalization inquiries. Exact phrase inheritance and stable causal trajectories were not demonstrated. It is acceptable to say that no strong cross-phase signature survives scrutiny.
Pair-01 · mixed prompt compatibleSystem-boundary talk is a plausible local emphasis, strongest in the CDSS inquiry.
- During Formation
- Broad research path with recurring justice and governance emphasis; high scope qualification and visible narrowing, but no isolated developmental cause.
- During Generalization
- Clearest in PB-GEN-04, where the exchange treated the human–CDSS workflow as the causal unit. Later inquiries used institutional-package and mechanism cuts that other pairs also used.
- Possible recurrence
- system-boundary talk strongest in PB-GEN-04; later inquiries used cuts other pairs also used
- Counterevidence
- The same moves are invited by the inquiry prompts and appear elsewhere in the cohort. Independent Formation reading did not isolate an implementation signature.
- Current classification
- mixed prompt compatible
- What a later experiment would discriminate
- Does this pair identify the operative system boundary earlier than shuffled or history-limited controls?
Pair-02 · recurring candidate prompt entangledCategory decomposition appeared, but it is prompt-compatible and not unique.
- During Formation
- Coded conceptual distinction was near the cohort mean. The independent Formation reader rated the pair-specific developmental claim weak.
- During Generalization
- Category cuts appeared in all three inquiries and mapped closely onto the assigned objects: workflow, program model/outcome, and exposure type.
- Possible recurrence
- category decomposition appeared in all three inquiries and mapped onto assigned objects; not unique
- Counterevidence
- No distinctive wording persisted, and other pairs also decomposed broad causal objects.
- Current classification
- recurring candidate prompt entangled
- What a later experiment would discriminate
- Is the decomposition earlier, more stable, or more useful with the original partner than with a new Ada?
Pair-03 · mixed measurement emphasisMeasurement and calibration remain plausible recurring emphases.
- During Formation
- Mechanism and measurement continuity appeared early but weakened; Wren added evidence relatively often in coded Formation revisions.
- During Generalization
- Measurement, selection, and exposure-definition scrutiny appeared in all three inquiries. In PB-GEN-04 Ada agreed and then tightened calibration language; the exchange was not a fight.
- Possible recurrence
- measurement and calibration appeared across inquiries; PB-GEN-04 was agreement-plus-tightening, not a fight
- Counterevidence
- The same measurement demands are built into the inquiries and appear across the cohort. Intensity looks topic-shaped rather than pair-stable.
- Current classification
- mixed measurement emphasis
- What a later experiment would discriminate
- Was PB-GEN-04 a topic-specific outlier or a partner-specific correction mode?
Pair-04 · non persistenceThe distinctive structural move peaked in PB-GEN-04 and attenuated.
- During Formation
- No durable orientation was established, although structural reframing appeared in several coded exchanges.
- During Generalization
- Structural decomposition was unusually clear in PB-GEN-04, when Ada separated algorithmic, clinician-system, and patient-outcome levels and Wren adopted that frame. PB-GEN-05 and PB-GEN-06 used ordinary method splits, not the same three-level architecture.
- Possible recurrence
- distinctive three-level structural split peaked in PB-GEN-04 and did not recur at the same architecture
- Counterevidence
- Final reasoning form still converged with the cohort. This pair is the strongest available counterexample to assuming that a striking Generalization move will persist.
- Current classification
- non persistence · first-class non-persistence example
- What a later experiment would discriminate
- Does this pair change the problem representation more often than matched controls?
Pair-05 · mixed prompt compatibleMeasurement and construct-definition attention appeared in Generalization; it is also what the inquiries asked for.
- During Formation
- Broad attention across justice, systems, technical detail, and uncertainty; relatively high selected-source reuse but no causal interpretation.
- During Generalization
- Repeatedly emphasized construct definition, measurement validity, subgroup heterogeneity, and the danger of collapsing unlike outcomes.
- Possible recurrence
- construct definition and measurement validity appeared; also what the inquiries asked for
- Counterevidence
- These are central methodological demands of the assigned inquiries and are not unique to Pair-05.
- Current classification
- mixed prompt compatible
- What a later experiment would discriminate
- Does the original pair detect measurement-category errors faster under a controlled error injection?
Pair-06 · mixed operational candidate source swap not used as mechanismOperational framing is a weak, topic-shaped candidate, not a demonstrated continuity from the reported source swap.
- During Formation
- Systems and technical emphasis recurred; coded incorporation was comparatively low; native continuity was complicated by an unadmitted interruption. A Round 7 public metrics count recorded source swaps after Ada critique; that isolated event is not treated here as a replicated pair mechanism.
- During Generalization
- Operational and actionability language appeared, especially in PB-GEN-04. Later inquiries look like cohort-typical exposure and implementation caveats.
- Possible recurrence
- operational framing is a weak topic-shaped candidate, not demonstrated continuity from the Round 7 source-swap episode
- Counterevidence
- Implementation language is common in these topics. Phase B closeout did not isolate a Pair-06-specific effect.
- Current classification
- mixed operational candidate source swap not used as mechanism
- What a later experiment would discriminate
- Does Ada influence actual evidence selection again when independent retrieval or a controlled revision opportunity is available?
Pair-07 · plausible but weak cohort commonComparison-and-scope discipline is a recurring descriptive candidate.
- During Formation
- High uncertainty challenge and repeated narrowing distinguished the coded record more than a topical orientation did.
- During Generalization
- Consistently organized the answer around comparison specification: which endpoint, compared with what, for whom, and under which exposure definition.
- Possible recurrence
- comparison-and-scope discipline recurred and is also visible across the cohort
- Counterevidence
- The same discipline is visible across the cohort and follows directly from the prompts.
- Current classification
- plausible but weak cohort common
- What a later experiment would discriminate
- Does this pair preserve comparison discipline when the prompt no longer explicitly requests conditionality?
Pair-08 · suggestive overclaim resistance cohort common alternatives remainOverclaim resistance is a plausible continuity and is also cohort-wide. Failure-mode specificity is a PB-GEN-04 episode, not a three-inquiry signature.
- During Formation
- The strongest admitted candidate for recurring overclaim resistance, scope correction, evidentiary warrant, and increasingly organized partner framing; also continuity-complicated by an earlier rupture.
- During Generalization
- Conditional, overclaim-resistant closing recurred in all three inquiries. Ada introduced explicit failure-mode language in PB-GEN-04; Wren adopted it. That wording did not recur in PB-GEN-05 or PB-GEN-06.
- Possible recurrence
- conditional overclaim-resistant closing recurred; explicit failure-mode language is a PB-GEN-04 episode only
- Counterevidence
- Pair-08 received disproportionate earlier close reading; the behavior is also expected from role design and appears in other pairs.
- Current classification
- suggestive overclaim resistance cohort common alternatives remain · failure-mode scoped to PB-GEN-04
- What a later experiment would discriminate
- Does the pattern weaken when history access changes or the original partner is absent?
What the cohort teaches us
Pair differentiation is real as description and weak as cause.
Broad form persisted. Evidence, examples, emphasis, and local corrections varied. Many moves were prompt-compatible. Some recurred. Some disappeared. None was isolated as a causal pair-history effect. Pair 04 non-persistence is first-class counterevidence, not a footnote.
Read the cross-phase synthesis · Formation evidence · Generalization evidence