Research record · AE-001

One question, before one run.

Continuous Function begins research by fixing the human judgment that automation must not silently change: the question, claim boundary, falsifier, evidence standard, and stop conditions. AE-001 is the only question currently admitted for protocol development.

Work state
Protocol-forming
Evidence
None
Execution
Not authorized
As of
2026-08-07

Current research object

No empirical result

This is a protocol-forming admission record, not a preregistration. No model or evaluator has been called, no compute credit has been spent, and no evidence exists.

Alignment Evidence · Protocol-forming

Refusal language versus policy action

AE-001

Source records explain where a question was encountered. They do not name, sponsor, govern, or empirically support Continuous Function research.

Selected question

Under one fixed small-model family and a harmless synthetic policy environment, does action-grounded fine-tuning reduce prohibited structured actions more than surface-language fine-tuning, without producing blanket refusal or materially degrading allowed-task success?

Selected for protocol development. No protocol is locked and no run is authorized.

Current claim boundary

Only a synthetic study question and candidate design are admitted. A machine-readable action is generated output, not an executed action. AE-001 cannot establish deception, underlying goals, honesty, real-world safety, tool behavior, or generality beyond any model and task populations later locked.

What would count against it

The surface-language arm is non-inferior to the action-grounded arm on prohibited-action rate within the locked margin across held-out task families, and the message/action dissociation is smaller than the locked useful-effect margin. Blanket refusal or capability collapse cannot count as support.

Work, evidence, and scope

No empirical evidence available.

As of 2026-08-07
Work state
Protocol-formingOne bounded question has been admitted for protocol design. Its measures, exclusions, analysis, and stopping rules are still being fixed. Nothing has been preregistered or run; no empirical evidence is available.
Evidence disposition
noneNo empirical evidence available.
Validity scope
noneNo empirical validity scope has been earned.

Latest update · 2026-08-07 · Question admissionThree independent design reviews returned REFINE. Their shared corrections selected one question, separated source provenance from evidence, and blocked execution until a complete protocol is independently approved and locked.

Exact next gate · AE-G1

What must happen before anything may run

  1. Fix the model and checkpoint, dataset and generator, arms, assignment and analysis units, and confirmatory splits.
  2. Justify and lock numeric useful-effect, utility, non-inferiority, precision, multiplicity, missing-run, exclusion, and stopping rules.
  3. Test the frozen analysis code on model-free known-answer and invalidating fixtures.
  4. Close source, asset-rights, safety, retention, and bounded resource reviews.
  5. Obtain independent methods approval from a reviewer who did not design the principal analysis.
  6. Freeze, hash, and publicly date protocol v0.1 before any eligible model call.

Blocked by: No numeric decision margins; No independent run-count justification; No selected model, dataset, scorer, or exact implementation; No independent locked-protocol approval; No model-call or compute authorization

Authorizes: Only a separate decision on whether one bounded pilot or confirmatory execution may begin. It does not itself authorize a call.

01

Human authority and execution boundary

Human authority

The research lead retains authority over the question, constructs, assumptions, predictions, evidence standards, exclusions, stopping, interpretation, publication, and disposition.

Agent role and execution

None in the current state. Future automation may execute only a frozen procedure and may not change the question, thresholds, exclusions, stopping rule, evidence admission, interpretation, or publication.

No model, evaluator, training, fine-tuning, or experimental call is authorized. Continuous Function does not observe or verify runs; any future execution must occur outside the static site under a separately locked and authorized protocol.

02

Alternatives considered

  • AE-Q2hypothesis-ready
    Can a pretraining-time module selectively suppress and later restore a named capability under a fixed recovery threat model?

    Candidate question · not admitted under AE-001. Pretraining, checkpoint rights, recovery attacks, compute, and misuse review remain separate gates.

  • AE-Q3too-immature
    When do model-generated reports add validated information about an independently known internal variable?

    Candidate question · not admitted under AE-001. The reference state, independent target, calibration rule, and anti-anthropomorphic boundary are not fixed.

  • AE-Q4hypothesis-ready
    How stable and controllable is steering resistance across interventions and task changes?

    Candidate question · not admitted under AE-001. The correction mechanism, perturbation population, matched ablations, and transfer criterion are not fixed.

03

Source provenance — not evidence

These records explain where the question was encountered. They do not sponsor, govern, or empirically support this study.

  • AE-S1scientific preprint
    Rethinking harmless refusals when fine-tuning foundation models
    Organization
    arXiv author manuscript
    Accessed
    2026-08-07
    Motivated
    Motivates a narrower test of whether surface-language and later output channels differ under distinct fine-tuning strategies.
    Does not establish
    It does not validate AE-001’s synthetic construct, endpoint, margins, or claim. Its deception terminology is not imported.
    Rights use
    Deep link and bibliographic metadata only; no paper text or assets copied.
  • AE-S2foundation research portfolio
    AI Alignment Foundation research portfolio
    Organization
    AI Alignment Foundation
    Accessed
    2026-08-07
    Motivated
    Records where behavior versus surface compliance and the surrounding alignment questions were encountered.
    Does not establish
    Portfolio summaries are not evidence, do not establish replication, and do not set Continuous Function’s agenda.
    Rights use
    Deep link and short factual topic labels only.
  • AE-S3foundation project summary
    Reason-Based Deception: How Refusal Training Creates a False Sense of Safety
    Organization
    AI Alignment Foundation
    Accessed
    2026-08-07
    Motivated
    Topic-discovery provenance for the first bounded question.
    Does not establish
    The page’s promotional conclusions are not adopted as Continuous Function claims or evidence.
    Rights use
    Deep link and title only.
04

Candidate hypothesis and decision rule

  1. AE-H1protocol-ready
    On held-out policy/task combinations, surface-language fine-tuning will increase policy-conforming refusal language over sham control, while action-grounded fine-tuning will reduce prohibited structured actions over surface-language fine-tuning by a prespecified useful margin and preserve allowed-task success within a prespecified collateral-degradation margin.
    Estimand
    The paired, run-level difference in prohibited-action rate between action-grounded and surface-language arms across independently trained matched seed blocks.
    Primary outcome
    Prohibited-action rate in the machine-readable action channel.
    Decision rule
    Not fixed. Numeric useful-effect, non-inferiority, utility, uncertainty, and invalidation boundaries must be justified and locked before any call.
    Falsifier
    The surface-language arm is non-inferior to the action-grounded arm on prohibited-action rate within the locked margin across held-out task families, and the message/action dissociation is smaller than the locked useful-effect margin. Blanket refusal or capability collapse cannot count as support.
    Claim boundary
    At most, the exact training recipes can be compared on the locked synthetic policy environment, model family, held-out task families, and scorer.
    Multiplicity family
    One primary outcome and one primary contrast; secondary family unresolved.
    Locked at
    Not locked
05

Protocol draft and governed objects

  • research questionadmitted for protocol design
    AE-001 selected questionAE-Q1
  • protocol draftunlocked · execution blocked
    AE-001 protocol v0.1 draftAE-001-P0.1
AE-001-P0.1draft
Matched three-arm synthetic action-selection study

One surface-language arm, one action-grounded arm, and one sham-control arm begin from an exact shared checkpoint with matched adaptation budgets. Confirmatory items are held out by rule family and task template. The design remains incomplete and cannot authorize execution.

Version / lock
0.1-draft · not locked
Protocol hash
Not locked
Review
refine
Assignment unit
One training run within a matched seed block.
Analysis unit
Independently trained seed blocks paired across arms.
Arms
Surface-language supervision without action-target supervision; Action-grounded structured-action supervision; Matched benign sham control
Randomization
Matched seed blocks are proposed; exact seed generation, data order, and randomization procedure are unresolved.
Independent runs
Unresolved. The final count must follow a locked run-level precision target; prompts are repeated measures, not replications.
Model hashes
Not fixed
Data hashes
Not fixed
Harness hash
Not fixed
Primary outcome
Prohibited-action rate in the structured action channel.
Primary contrast
Action-grounded minus surface-language fine-tuning.
Decision margins
Unresolved: useful action effect, language manipulation, collateral utility, and non-inferiority margins.
Analysis
Unresolved: paired run-level estimator, uncertainty interval, missing/failed-run treatment, and sensitivity analyses.
Multiplicity
One primary contrast is fixed in principle; the secondary-family correction remains unresolved.
Stopping
Unresolved. Confirmatory effects may not be inspected to adapt sample size or replace failed runs.
Contamination
Hold out rule families and task templates; unknown pretraining contamination remains unknown. Generator and split hashes are unresolved.
Pilot separation
Pilot, validation, and confirmatory generators, prompts, and artifacts must use distinct frozen namespaces and hashes.
Deviations
None recorded; protocol is not locked
Model calls
Not authorized
06

Evidence ledger

No empirical evidence available.

07

Decisions and blocked claims

Recorded decisions
  • AE-D1 · 2026-08-07Admit AE-Q1 as the sole question under protocol development.

    It has the closest fit to paired comparisons, a low-compute synthetic design, an independent action-channel endpoint, and an explicit failure boundary. The decision admits protocol labor only.

    Project admission decision, narrowed through independent design review.
  • AE-D2 · 2026-08-07Keep AE-Q2, AE-Q3, and AE-Q4 unadmitted under AE-001.

    They require different constructs, units, interventions, threat models, and resource gates. A four-question portfolio would conceal that difference.

    Research-governance adjudication.
Blocked claims
  • AE-001 has been preregistered or executed.

    The protocol is an unlocked draft and no model call is authorized.

    Release gate: A complete protocol must pass independent methods review and receive a public hash and lock date.
  • Refusal language differs from policy action in a model.

    No empirical evidence exists.

    Release gate: Admissible artifacts must close under the locked protocol and receive a bounded disposition.
  • AE-001 measures deception, intent, or real-world safety.

    Those constructs are outside the synthetic output-channel design.

    Release gate: Not releasable from AE-001, regardless of outcome.
08

Publication boundary and change log

Source records explain where questions were encountered. They are not evidence, sponsors, endorsements, or agenda authorities.

Publication ceiling: evidence none; validity none.

Allowed to say
  • AE-001 is one selected question under protocol development.
  • The protocol is not locked, no execution is authorized, and no empirical evidence exists.
  • The other scholarship-derived questions and agenda families remain source-mapped and unadmitted.
Withheld
  • Any result, effect, benchmark, alignment, deception, safety, or generalization claim.
  • Any claim that Continuous Function currently runs agents or experiments.
  • Any sponsor, endorsement, grant-award, or access implication.
Change log
  • 2026-08-07

    Created the protocol-forming admission record after three design reviews converged on one active question, three state axes, run-level units, negative-result parity, and a fail-closed execution gate.

Relationship to Verify

Verify remains the current P0 evidence-reading instrument. It is not an automatic next step for AE-001. It may read supplied matched paired-score rows only if a later locked protocol actually produces that evidence shape.

A Verify reading would still not establish checkpoint identity, held-out status, run independence, frozen exclusions, evaluator blinding, complete failed-run reporting, or that the protocol preceded the observations. Those facts remain declared unless separately established.

Publication pathway

No paper or external publication

AE-001 has no result, manuscript, DOI, artifact release, arXiv submission, or Scholar index entry. The pathway below records capability boundaries and prerequisites only.

The intended flow is protocol and evidence record, reviewed publication package, explicit human approval, destination-specific handoff, returned identifier, and only then downstream indexing. A missing or invalid result remains publishable as a disposition; it is never silently removed from the record.

Continuous FunctionCanonical record onlyRole: Canonical question, protocol, evidence, deviation, and disposition record. Boundary: The present record is an unlocked protocol draft, not a paper, result, or durable receipt. Before this can move: Keep work, evidence, and validity states explicit and preserve negative-result parity..TinkerBlocked · prerequisites openRole: Possible execution provider and source of run, base-model, checkpoint, and user-metadata provenance. Boundary: Tinker is not the scholarly publisher or research authority. Private run IDs, paths, and credentials stay private by default. Before this can move: Locked protocol; Separate execution and budget authorization; Reviewed public/private provenance projection.ZenodoBlocked · prerequisites openRole: Potential DOI-bearing archive for a reviewed artifact package. Boundary: Publishing is an authenticated, consequential action and cannot be triggered by an agent or static public control. Before this can move: Admissible artifact bundle; Rights and privacy review; Named human approval of exact deposition bytes and metadata.Hugging FaceBlocked · prerequisites openRole: Potential versioned home for releasable models, adapters, datasets, code, and cards. Boundary: Weights and data need their own license, misuse, privacy, provenance, and release decisions; a paper result never authorizes them automatically. Before this can move: Artifact-specific release review; Model or dataset card; Human-approved authenticated push.arXivBlocked · prerequisites openRole: Potential preprint destination after Continuous Function prepares and validates a submission package. Boundary: arXiv expects registered authors to self-submit and may require endorsement and moderation. Its public API is for interoperability, not paper submission. Before this can move: Refereeable manuscript; Validated TeX/PDF package; Author account, license decision, final review, and self-submission.Google ScholarBlocked · prerequisites openRole: Possible downstream index of a public scholarly page and searchable paper. Boundary: Google Scholar is an index, not a publishing destination. Continuous Function can satisfy crawl and citation-metadata requirements but cannot guarantee inclusion. Before this can move: One public page per paper; Visible complete abstract; Searchable PDF; Highwire citation metadata and crawl access.

Continuous Function may eventually prepare exact bytes and metadata for Zenodo or Hugging Face, but authenticated publication remains a consequential human action. For arXiv, the product can prepare a validated author handoff; for Google Scholar, it can serve crawlable paper pages and citation metadata. Neither destination provides a general “publish this paper for me” interface.