Synthetic hands: a protocol for Alter and Adversary

Version 0.1 · October 2026 · built from round two of the literature review. Untested: the first study is its calibration.

Scope

A synthetic hand is any model output that enters an entry. Only two roles are allowed. Alter drafts the Fiction. Adversary argues the Test. Trace and Reading are always human: no synthetic hand supplies a Trace, writes a Reading, codes, themes or interprets fieldnotes. The method text should avoid saying that models “read” or “understand”; they produce objections and drafts, and a person decides.

The objection and what the method concedes

An open letter signed by 419 people from 32 countries rejects generative AI for reflexive qualitative research (Jowsey et al. 2025; Jowsey et al. 2025). Only its abstract was read. Its two grounds, as far as can be told: reflexive analysis needs a positioned human researcher, and the technology has social and environmental costs.

The first ground the method can meet in full. Analysis and interpretation are the Reading register, which is human by rule. The second ground no protocol answers. The protocol limits and counts use; it does not claim to settle the point.

The replies mostly refuse a for-or-against split and propose the model as an interlocutor that points to passages, offers contrasting views and looks for negative cases, while the researcher interprets and documents everything (Friese et al. 2026; Nguyen-Trung et al. 2026; Lester et al. 2025; Silver 2026). That is close to the Adversary as defined here. No statement from the letter’s authors on a critic or devil’s-advocate role was found, so whether they would accept it is inference.

What the evidence says, and what follows

FindingConsequence for the protocol
A prompted model critic drifts towards agreement, more so when the user states confidence or writes in the first person; an instruction to be critical is the weakest remedy (Cheng et al. 2025; Dubois et al. 2026; Hong et al. 2025; Sharma et al. 2024)Give the Adversary a neutral third-person question. Run it on the Reading and its negation. Treat agreement as a fault.
Models favour text from their own family and are swayed by position and authority cues (Mahbub et al. 2025; results 2025; Ye et al. 2025; Panickssery et al. 2024)Adversary from a different model family than Alter; at least two families; independent runs.
Debate gains are shown where an answer key exists; voting accounts for much of the gain; models shift after reading peers (Choi et al. 2025; Michael et al. 2023)Several independent runs, not an exchange. No claim that the Adversary improves interpretations.
A role-played devil’s advocate is weaker than authentic dissent (Nemeth et al. 2001)The method says plainly that the Adversary is prompted advocacy. Failure to break a Reading is weak evidence for it. The outside reader is not replaced.
AI dialogue changes beliefs through information and conversation, not warmth; the same persuasiveness works for false claims (Boissin et al. 2025; Costello et al. 2026; Hackenburg et al. 2025)Adversary output is a set of leads. Factual-looking claims stay unverified until a person opens the source. Dry register. Alter is the promoting condition, so its output stays apart from the Adversary’s.
Sycophantic chatbots can drive a user into confirmed false belief even when the user reasons well (Cheng et al. 2026; Chandra et al. 2026)Structural limits, not disclosure: one structured pass, no open chat, no memory between entries, turn caps, a stop right for the outside reader.
Model-written fiction converges: same plots, tidier, less ambiguous; prompting and temperature do not fix it (Rettberg and Wigers 2025; Xu et al. 2025; DK et al. 2026; Russell et al. 2026; Padmakumar and He 2024)Fiction is evidence about the model, not the place. The human edit is the unit of record. Control run without the Trace.
One preprint reports that a model continues inside an elaborate fiction and lets real findings pass as confirmation (single model, not replicated) (Oliveira et al. 2026)Alter gets only its own entry’s Trace and Reading, never continues across entries, and every Fiction ends with a fiction-leak check.

The Science paper. The widely cited result that a model arguing with a conspiracy believer cut belief by about a fifth (Costello, Pennycook and Rand, 2024) (Costello et al. 2024) has carried an Editorial Expression of Concern since 11 June 2026 (Thorp 2026). The notice reports screening criteria that differed between manuscript and published pipeline, and spliced rows in the public dataset. It is not a retraction. The authors posted correction materials on 27 September 2026 (Costello et al. 2026) and say results hold; Science’s outcome was not found. The method text will not cite its effect sizes. The notice itself could not be opened; the details come from the AAAS release and Retraction Watch as reported by the auditor.

The protocol

A. Order of work in each entry

  1. Human first. Write Trace, then Reading, then your own Test (what would show it wrong, and the check) before any synthetic output is seen. (Lester et al. 2025)
  2. Adversary run. Input: the Trace and the Reading restated as a neutral third-person question. No stated confidence, hopes, biography or preferred outcome, and never the Fiction. Run it on the Reading and on its negation. (Dubois et al. 2026; Hong et al. 2025; Cheng et al. 2025)
  3. Alter run. Input: this entry’s Trace and Reading only. The prompt names the tension in the Reading and asks for a version in which it is not resolved. At least two runs under different human-written constraints. (Rettberg and Wigers 2025; Russell et al. 2026)
  4. Keep them apart. The Fiction is not shown in the same block as the Adversary’s output for the same Reading.
  5. Human adjudicates and logs. Each objection is accepted, rejected with a reason, or unchecked. The Test result is written by a person. (Nyaaba et al. 2026; Eschrich et al. 2024)

B. Adversary rules

  1. A different model family from Alter. At least two families, each run independent and blind to the others; no debate exchange. (Mahbub et al. 2025; results 2025; Choi et al. 2025)
  2. Every objection names a checkable observation (a passage, place, count or document) and the cheapest check. Objections that cannot be checked are counted as generic. The outside reader does the rating, not a model. (McAleese et al. 2024; Han et al. 2026)
  3. Anything factual-looking in an Adversary output (a count, date, clause, quotation) is a lead until a person opens the source and records hit or miss. (Hackenburg et al. 2025; Hicks et al. 2024)
  4. Approval of both a Reading and its negation is a failed run. Agreement with no checkable contradiction flags the entry; it does not close it as held. (Cheng et al. 2025)
  5. Dry, non-personal register: questions and pointers to sources, not assertions of counter-fact. (Boissin et al. 2025)
  6. One structured pass: a brief in, a list of checks out. No open chat, no memory between entries, a cap on turns. (Chandra et al. 2026)

C. Alter rules

  1. The human edit is the unit of record. Keep the model output and the edited Fiction as a diff and log the share of model text that survives. A mostly unedited Fiction is marked a model default, not a finding. (Padmakumar and He 2024)
  2. Once per study, run the same Fiction prompt with the Trace removed. What differs from the real run is what the Trace added; what is the same is what the model supplies. (Rettberg and Wigers 2025)
  3. Fingerprint checklist for the editor and outside reader: theme stated outright, single thread, plot resolved, emotion given as bodily sensation. A Fiction with every tick may not prompt a new Reading. (Russell et al. 2026)
  4. Close every entry with the fiction-leak check: list each factual-looking claim in the Fiction and mark it invented. A later Reading resting on one is invalid until a Trace exists. (Oliveira et al. 2026; Hicks et al. 2024)
  5. The label carries three things: the word Fiction, the model name and date, and the human role. Each Fiction is signed by a named person who answers for it.
  6. Fiction is judged by what it opens, not how it reads: record whether it raised a question still in use two or more entries later. Expect most will not, and report that.

D. Records: the hands card

For each synthetic hand in an entry: model and exact version, date, prompt, raw output, rejected runs, and the human response (kept, edited as a diff, rejected, overruled, with a one-line reason). Keep the Reading’s wording before and after any Adversary run. Per register, one line each on method, technology, humans and knowledge: why a synthetic hand was used and what was not delegated (Lester et al. 2025; Silver 2026). Checksum the export so someone else can audit it. The aim is auditability, not replication, because outputs vary between runs.

E. Limits, costs and the opt-out

The legitimate-use test

A list an objector can check against the study:

  1. The researcher holds every Reading.
  2. Synthetic output is labelled and is cited as evidence by no one.
  3. Every use is logged with its human response.
  4. Costs are counted.
  5. An opt-out version exists.
  6. The study reports where the synthetic hand changed nothing.

Calibration

No source tests a model critic on a researcher’s interpretation of a place or document. The first study, in whichever proposal it is run, therefore doubles as a test of the Adversary, and its figures are reported as a pilot.

Before the study

During the study

What the protocol does not claim

Open points and weak spots

Existing disclosure formats in government

Formats for recording the use of AI in government exist at several grains. None records, for one output, who or what produced each part and what the human did with it. That is the gap a hands card is meant to fill, and the formats below are what it has to be compared with.

GrainFormatWhat it records
Tool registerUK Algorithmic Transparency Recording Standard (GDS 2021)An owner by role title, how humans are involved, how far the process is automated, and the review route. Published records: 153 on 14 October 2026.
AssessmentCanada’s Directive on Automated Decision-Making and Algorithmic Impact Assessment (Government of Canada 2019); the EU AI Act’s fundamental-rights assessment, Art. 27 (EU AI Act 2024)A risk level, with duties that scale to it.
Decision noticeFrench law since 2017 (French and Italian law)That a decision rests on an algorithm, and on request the degree and mode of its contribution.
Document ruleEU AI Act, Art. 50 (EU AI Act 2024)Disclosure of AI-generated text published to inform the public on matters of public interest, unless a human has reviewed it and someone holds editorial responsibility.
Data modelW3C PROV-DM (Moreau and Missier 2013)Agents, activities and entities, and delegation between agents.

What none of them records is what the reviewer changed, the time available, and the reviewer’s authority to override or stop. The EU AI Act names the duty (Arts 14 and 26) but no format records it (EU AI Act 2024). A hands card for a government setting adds those three fields to the fields the formats already hold.

Testing the human response

Replace “human in the loop: yes” with observed behaviour: the override rate, the edit rate, the time per item, and whether the reviewer wrote a view before seeing the machine’s. The evidence on over-trust is mixed. Dutch experiments found selective use of advice that fits a prior view and not general deference (Alon-Barkat and Busuioc 2023; Selten et al. 2023). Policies that require human oversight rest on a capacity people may lack (Green 2022), and the reviewer can become the one who absorbs the blame, a moral crumple zone (Elish 2019). A human response that never differs from the machine’s is a finding to test, not a pass.

Cautions about the card itself

Sources cited above