FlowingFlowing
← Back to all development cases
This case was judged by an anonymized AI reviewer and has not been independently reviewed by people; inspect the draft, cards, and both outputs to assess it.
AutocompleteWithout cards better56 / 280 · beu-4f245456d6adadf1

Cosmic ray feedback in galaxies and galaxy clusters -- A pedagogical introduction and a topical review of the acceleration, transport, observables, and dynamical impact of cosmic rays

Astrophysics · 2306.03141v2

FLOWING EVIDENCE BENCHMARK

How do we compare writing with and without cards?

At the same writing position, the same model produces one first result with cards and one without. All 280 pairs enter the overall anonymized review, which assesses whether a decision not to continue fits the draft.

Overall: 138 with-card wins, 26 ties, 107 no-card wins, 5 pairs where neither was usable, and 4 inconclusive judgments. Both sides wrote text in 218 pairs; 48 had a deliberate stop and 14 included an invalid output.

SAME DRAFT · WITH OR WITHOUT CARDS

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED VERSIONS

SHARED BY BOTH

Draft, writing task, and generation model

A · With cards

Relevant literature cards supplied

B · Without cards

No literature cards supplied

Generation model

Same model and version

A → first output

B → first output

Anonymized AI judge

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Review again with the answer order swapped

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The AI judge sees the same draft and both first outputs without knowing which received cards.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are positions and cards selected?

We selected 40 writing positions in each of seven fields, for 280 comparisons. The position and relevant literature cards were set before each comparison, so both versions began from the same draft.

The main result retains every frozen position, including places where cards might not add a useful next move. This avoids selecting cases based on their outputs, but 49% describes the whole cohort, not a win rate where evidence clearly had an opportunity to help. This cohort did not record that opportunity label for each position before generation, so those two groups cannot yet be separated reliably.

CARD USE · 140/280

Reported card use in continuations with text

The model reports whether a card affected its continuation. In 140 of 280 tests it produced continuation text and reported card use. The previous count of 150 included 10 cases without continuation text and has been corrected.

COMPARISON · 70 / 7 / 30

Which written version was better?

Of the 107 reported-use cases where both versions produced text, the blinded AI judge preferred the with-card version 70 times, found 7 about the same, and preferred the without-card version 30 times.

How are the two table percentages calculated?

ADOPTION COVERAGE

140reported use with text
÷
280all positions
=
50%

WITH-CARD VERSION PREFERRED

70with cards better
÷
107comparable tests
=
65%

The percentages use different groups: 50% counts continuations with text reporting card use among all 280 tests; 65% covers the 107 such cases with text on both sides. The generating model reports card use, while an anonymized AI judge compares writing quality.

Card use is self-reported by the generator. The anonymized AI judge sees the draft and source snippets, so source-aware preference remains possible; there was no independent human review. The overall result describes these 280 frozen positions, not a win probability for any writing task.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Cosmic ray feedback in galaxies and galaxy clusters -- A pedagogical introduction and a topical review of the acceleration, transport, observables, and dynamical impact of cosmic rays

1 Introduction > 1.3 Role of cosmic rays in galaxy evolution

…sqrt{s} and the beam energy E lab E_{\mathrm{lab}} in a fixed target experiment (measured in the laboratory frame) are related via s = ( 2 ​ E lab ​ m p ​ c 2 ) 1 / 2 \sqrt{s}=(2E_{\mathrm{lab}}m_{\mathrm{p}}c^{2})^{1/2} so that the energy reachable by the Large Hadron Collider s ≈ 14 \sqrt{s}\approx 14 TeV corresponds to a CR particle energy E lab ≈ 10 17 E_{\mathrm{lab}}\approx 10^{17} eV, which falls short by a factor of 3 × 10 3 3\times 10^{3} in comparison to the highest CR energies ever measured. Here, m p m_{\mathrm{p}} and c c denote the proton rest mass and light speed, respectively.

…nd the star formation efficiency falls off rapidly toward both ends of the mass spectrum. These missing baryons either did not fall into dark matter potential wells (or remain undetected) or were expelled due to feedback processes, which can be ejective (when gas is expelled from galaxies before it can form stars) and/or preventative (when gas is heated and prevented from accreting onto galaxies). Star formation suppression is thought to be due to stellar and supermassive black hole feedback below and above this critical halo mass threshold, respectively ( Somerville and Davé, 2015 , e.g.,) .

Key challenges facing commonly employed thermal and radiation pressure–driven stellar feedback models are (i) the overcooling problem, where the thermal energy injected by supernovae (SNe) is quickly radiated

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionOverall AI review: Without cards better

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Read overall AI review rationale

A = without cards · B = with cards

Both outputs correctly complete the sentence by adding a second challenge (ii) to parallel the first (i). Both are grammatically and stylistically appropriate continuations. Output A is preferable because it maintains tighter thematic coherence with the draft's focus on 'thermal and radiation pressure–driven stellar feedback models.' Output A's 'difficulty of launching outflows from the dense, high-pressure interstellar medium of massive galaxies' directly addresses a well-known limitation of these specific feedback mechanisms in realistic galactic environments. Output B's 'accelerating gas to the escape velocities required to eject it from the disc' is more generic and somewhat misaligned: the overcooling problem already concerns the inability to drive winds, and 'escape velocity' framing is more relevant to mechanical/radiative feedback scenarios rather than the thermal/radiation pressure mechanisms specified in the draft. Output A's 'high-pressure interstellar medium of massive galaxies' precisely captures why thermal feedback fails in massive systems (the high ambient pressure suppresses wind launching), making it the more accurate and scope-appropriate completion for this specific argument.

Evidence

Same model and position · retrieved passages supplied · first output

Not preferred in this comparison

away before it can drive galactic winds, and (ii) the difficulty of accelerating gas to the escape velocities required to eject it from the disc.

Passages supplied to the Evidence version

Card 1 · Stellar feedback-regulated black hole growth: driving factors from nuclear to halo scales

ABSTRACT Several recent simulations of galaxy formation predict two main phases of supermassive black hole (BH) accretion: an early, highly intermittent phase (during which BHs are undermassive relative to local scaling relations), followed by a phase of accelerated growth. We investigate physical f…

Read full passage excerpt

ABSTRACT Several recent simulations of galaxy formation predict two main phases of supermassive black hole (BH) accretion: an early, highly intermittent phase (during which BHs are undermassive relative to local scaling relations), followed by a phase of accelerated growth. We investigate physical factors that drive the transition in BH accretion in cosmological zoom-in simulations from the FIRE project, ranging from dwarf galaxies to galaxies sufficiently massive to host luminous quasars. The simulations model multichannel stellar feedback, but neglect AGN feedback. We show that multiple physical properties, including halo mass, galaxy stellar mass, and depth of the central gravitational potential correlate with accelerated BH fuelling: constant thresholds in these properties are typically crossed within ∼0.1 Hubble time of accelerated BH fuelling. Black hole masses increase sharply when the stellar surface density in the inner 1 kpc crosses a threshold $\Sigma^\star _{1\,\rm kpc}\app …

Card 2 · Black Hole Growth and Feedback in Isolated ROMULUS25 Dwarf Galaxies

Abstract We investigate the effects of massive black hole growth on the structural evolution of dwarf galaxies within the R omulus 25 cosmological hydrodynamical simulation. We study a sample of 205 central, isolated dwarf galaxies with stellar masses and a central BH. We find that the local M BH –…

Read full passage excerpt

Abstract We investigate the effects of massive black hole growth on the structural evolution of dwarf galaxies within the R omulus 25 cosmological hydrodynamical simulation. We study a sample of 205 central, isolated dwarf galaxies with stellar masses and a central BH. We find that the local M BH – M star relation exhibits a high degree of scatter below M star < 10 10 M ⊙ , which we use to classify BHs as overmassive or undermassive relative to their host M star . Within isolated dwarf galaxies, only 8% of undermassive BHs ever undergo a BH merger, while 95% of overmassive BHs grow through a mixture of BH mergers and accretion. We find that isolated dwarf galaxies that host overmassive BHs also follow different evolutionary tracks relative to their undermassive BH counterparts, building up their stars and dark matter earlier and experiencing star formation suppression starting around z = 2. By z = 0.05, overmassive BH hosts above M star > 10 9 M ⊙ are more likely to exhibit lower centr …

Card 3 · The main sequence of star-forming galaxies across cosmic times

ABSTRACT By compiling a comprehensive census of literature studies, we investigate the evolution of the main sequence (MS) of star-forming galaxies (SFGs) in the widest range of redshift (0 < z < 6) and stellar mass (108.5–1011.5 M⊙) ever probed. We convert all observations to a common calibration a…

Read full passage excerpt

ABSTRACT By compiling a comprehensive census of literature studies, we investigate the evolution of the main sequence (MS) of star-forming galaxies (SFGs) in the widest range of redshift (0 < z < 6) and stellar mass (108.5–1011.5 M⊙) ever probed. We convert all observations to a common calibration and find a remarkable consensus on the variation of the MS shape and normalization across cosmic time. The relation exhibits a curvature towards the high stellar masses at all redshifts. The best functional form is governed by two parameters: the evolution of the normalization and the turnover mass (M0(t)), which both evolve as a power law of the Universe age. The turn-over mass determines the MS shape. It marginally evolves with time, making the MS slightly steeper towards z ∼ 4–6. At stellar masses below M0(t), SFGs have a constant specific SFR (sSFR), while above M0(t) the sSFR is suppressed. We find that the MS is dominated by central galaxies. This allows to turn M0(t) into the correspon …

Card 4 · FEEDBACK FROM CENTRAL BLACK HOLES IN ELLIPTICAL GALAXIES. III. MODELS WITH BOTH RADIATIVE AND MECHANICAL FEEDBACK

We find, from high-resolution hydro simulations, that winds from active galactic nuclei effectively heat the inner parts (≈100 pc) of elliptical galaxies, reducing infall to the central black hole; and radiative (photoionization and X-ray) heating reduces cooling flows at the kpc scale. Including bo…

Read full passage excerpt

We find, from high-resolution hydro simulations, that winds from active galactic nuclei effectively heat the inner parts (≈100 pc) of elliptical galaxies, reducing infall to the central black hole; and radiative (photoionization and X-ray) heating reduces cooling flows at the kpc scale. Including both types of feedback with (peak) efficiencies of 3 × 10 −4 ≲ w ≲ 10 −3 and of EM ≃ 10 −1.3 respectively, produces systems having duty cycles, central black hole masses, X-ray luminosities, optical light profiles, and E+A spectra in accord with the broad suite of modern observations of massive elliptical systems. Our main conclusion is that mechanical feedback (including energy, momentum, and mass) is necessary but the efficiency, based on several independent arguments, must be a factor of 10 lower than is commonly assumed. Bursts are frequent at z > 1 and decline in frequency toward the present epoch as energy and metal-rich gas are expelled from the galaxies into the surrounding medium. For …

Card 5 · Exploring supermassive black hole physics and galaxy quenching across halo mass in FIRE cosmological zoom simulations

Feedback from accreting supermassive black holes (SMBHs) is thought to be a primary driver of quenching in massive galaxies, but the best way to implement SMBH physics into galaxy formation simulations remains ambiguous. As part of the Feedback in Realistic Environments (FIRE) project, we explore th…

Read full passage excerpt

Feedback from accreting supermassive black holes (SMBHs) is thought to be a primary driver of quenching in massive galaxies, but the best way to implement SMBH physics into galaxy formation simulations remains ambiguous. As part of the Feedback in Realistic Environments (FIRE) project, we explore the effects of different modeling choices for SMBH accretion and feedback in a suite of $\sim500$ cosmological zoom-in simulations across a wide range of halo mass (10^10-10^13 Msun). Within the suite, we vary the numerical schemes for BH accretion and feedback, the accretion efficiency, and the strength of mechanical, radiative, and cosmic ray feedback independently. We then compare the outcomes to observed galaxy scaling relations. We find several models that satisfy the observational constraints, and for which the energetics in different feedback channels are physically plausible. Interestingly, cosmic rays accelerated by SMBHs play an important role in many successful models. However, it i …

Model-reported card use

The model did not report that any card materially affected this completion.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

away before it can drive a galactic wind, and (ii) the difficulty of launching outflows from the dense, high-pressure interstellar medium of massive galaxies.

Earlier card-use comparison

The earlier diagnostic compared only cases with reported card use and text on both sides; its scope differs from the overall AI review above.

This case does not have both with-card continuation text and a model-reported use record. It remains among all 280 tests but does not enter the reported-use comparison.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.