Declared for this mission and copied when its deposit arrived. Nothing on file records what the session actually ran under — an environment is selected by name in the runner's own account, and a name that matches nothing falls through to their default without saying so.

# Map brief — AlphaChip

Paste this whole thing. Nothing to fill in.

**Environment:** `night-shift-map`. The exact settings are on this mission's
page — the env file is the only place they live, so they cannot drift out of
sync with a copy written here.

---

## The job

Lay out the dispute over whether RL macro placement beats existing methods.
For each side: who, what exactly they claim, their stake, and the sources.
Quote the sentence where each states its position and say where it appears.
Note which benchmarks were actually run and which comparisons were never
made. Flag clearly what cannot be checked from public material — the key
data is proprietary, and saying so is part of the map, not a failure of it.

## Sourcing rules

- **"Blocked" and "paywalled" are different facts — say which.** A domain
  outside this environment's allowlist means you could not try. A domain that
  let you connect and then returned a paywall, login wall, or bot check means
  you tried and were refused. Both leave a gap in the record, but only the
  second tells the reader the source exists and is readable by someone else.
  For every source you could not read, the notes must say which of the two it
  was. Where a paywalled paper has a preprint, use the preprint and say you
  did.

## Output — two files, and both are required

**1. `output.md`** — the prose, as described above. Write it for a reader.
Structure it however reads best.

Then a section called **Notes on this run** — for me, not for a reader:
what you searched, what failed to load, which citations are unverified, what
you are unsure about, and anything that surprised you. Blunt.

**2. `positions.yaml`** — the same positions as structured data. One entry per
party per claim: a group that disputes two separate things gets two entries,
and two groups making the same claim get one entry each.

```yaml
positions_file:
  mission: MAP-002
  source: output.md
  count: <number of entries below>
positions:
  - id: P01
    party: <named person or group, as the source names them>
    claim: <one line, no hedging>
    stake: <employment, funding, or competition — factual, or "none found">
    sources:
      - url: <url>
        read: yes        # you opened this page and read the text you cite
      - url: <url>
        read: no         # you know it exists; you did not open it
```

**`read` is not a formality.** A URL you opened and a URL you merely know
exists are the same string, and listing one under `sources` implies you read
it. Say which. `read: yes` is checkable — an opened page leaves a fetch in
this session's transcript, which you do not write — so a `read: yes` with no
fetch behind it is a false record, and worth more scrutiny than an honest
`read: no`. Prefer citing what you read; where you must cite unread, say so
and the reader can weigh it.

**These must agree.** Every position in the prose appears in the YAML and vice
versa. The board counts entries in `positions.yaml` — it never parses your
prose, because a count derived from headings is a count that can silently be
wrong. If you can only produce one of the two files, produce neither and
deposit a finding explaining why.

An unmappable position is still a position: record it with the claim as stated
and `sources: []`, and say in the prose exactly what is missing.

## Where to put them

Push a branch to **`night-shift-network/inbox`** with both files inside a
directory named for this mission:

```
MAP-002/output.md
MAP-002/positions.yaml
```

The directory is what matters. A file at the root of the branch is not a
deposit and will not be collected — it will sit there looking delivered while
nothing has been.

The branch name does not matter and is deliberately not specified: the relay
reads every branch, so whichever branch your session is already on is fine.
Never delete or force-push a branch, yours or anyone else's; cleanup is
maintainer-side only.

# MAP-002 — Does RL macro placement beat existing methods?

A map of the dispute over Google's reinforcement-learning approach to chip
macro placement, published in *Nature* in 2021 and later branded **AlphaChip**.

The short version: **the central question has never been settled on shared
evidence, and it structurally cannot be.** The designs Google evaluated on are
proprietary TPU blocks that have never been released. The critics' evaluations
run on public benchmarks Google says are unrepresentative. Neither side has
published results on the other side's test cases. Every subsequent round —
*Nature*'s investigation, the addendum, the rebuttals — has argued about
methodology around that hole rather than closing it.

---

## The original claim

**Azalia Mirhoseini, Anna Goldie, Jeff Dean and 19 co-authors**, then at Google
Brain, published "A graph placement methodology for fast chip design" in
*Nature* 594, 207–212 (June 2021). The 2020 arXiv preprint states the claim:

> "in under 6 hours, our method can generate placements that are superhuman or
> comparable on modern accelerator netlists, whereas existing baselines require
> human experts in the loop and take several weeks."

— *Chip Placement with Deep Reinforcement Learning*, [arXiv:2004.10746](https://arxiv.org/abs/2004.10746),
abstract, final sentence.

The *Nature* version is widely quoted as: "In under six hours, our method
automatically generates chip floorplans that are superior or comparable to
those produced by humans in all key metrics, including power consumption,
performance and chip area." **I could not read the *Nature* abstract
first-hand** — see *Notes on this run* — so the preprint sentence above is the
one I am standing behind, and I used the preprint deliberately because the
journal version is behind a login wall.

The evaluation was on **TPU blocks**. Those netlists have never been released,
in original or sanitized form. That single fact is the load-bearing element of
everything below.

---

## The critics

### Satrajit Chatterjee and the internal team ("Stronger Baselines")

The first challenge came from inside Google. **Satrajit Chatterjee**, an
engineering manager in Google's AI research unit, led a separate team that
produced *Stronger Baselines for Evaluating Deep Reinforcement Learning in Chip
Placement* (Bae, Yazdanbakhsh, Kim, Chatterjee, Woo, Markov; March 2022). Its
technical claim: the *Nature* comparison was rigged by weak baselines — RePlAce
was not run properly, and simulated annealing, the obvious classical competitor,
was not run at all.

Google refused to authorise publication after review by an internal resolution
committee; the manuscript was posted anonymously in March 2022. Chatterjee
raised fraud concerns internally in October 2020, was disciplined in April 2021
for what Google called "uncredible claims of fraud", pressed again in February
2022, and was fired in March 2022. He sued under California whistleblower
protections; Google failed to get the suit dismissed in July 2023.

**This position is the least well-sourced on the map.** I could locate no public
URL for the *Stronger Baselines* manuscript itself, and every court document and
news account of Chatterjee's own words sat on a domain this environment blocks.
Everything above is second-hand, taken from Markov's meta-analysis (which cites
court filings made "under penalty of perjury") and from search-engine summaries
I could not open. Chatterjee has no first-person statement on this map.

### Igor Markov — "The False Dawn"

**Igor L. Markov**, Distinguished Architect in AI & Innovation at **Synopsys**,
formerly Google Search, Meta and the University of Michigan, published a
meta-analysis: *The False Dawn: Reevaluating Google's Reinforcement Learning for
Chip Macro Placement* ([arXiv:2306.09633](https://arxiv.org/abs/2306.09633),
v1 June 2023, v10 September 2024), later carried by *Communications of the ACM*
(October 2024).

His performance claim, from the abstract:

> "Our meta-analysis shows how two separate evaluations filled in the gaps and
> demonstrated that Google RL lags behind (i) human designers, (ii) a well-known
> algorithm (Simulated Annealing), and (iii) generally-available commercial
> software, while being slower; and in a 2023 open research contest, RL methods
> weren't in top 5."

His conclusion section puts it as: "prior methods outperform the methods of
[1], [4] in quality and runtime" (§8.1, conclusion 3).

He makes a second, separable and much sharper claim about research integrity,
also in the abstract:

> "Crosschecked data indicate that the integrity of the Nature paper is
> substantially undermined owing to errors in conduct, analysis and reporting.
> Before publishing, Google rebuffed internal allegations of fraud, which still
> stand."

And a third, on reproducibility: "Google released no proprietary TPU chip design
blocks used in [1] (nor sanitized equivalents), ruling out full external
reproduction" (§5).

On the contest: 19 participants, 8 finalists, "Google did not take part," and
"according to the participants' own descriptions, top six teams used traditional
analytical optimization methods sans ML."

**Stake.** Markov works for Synopsys, which sells competing commercial
place-and-route software — a direct competition stake. His paper carries the
disclaimer "opinions given here are the author's and do not represent his
employers." He also **discloses** that he co-authored *Stronger Baselines*, one
of the two evaluations his meta-analysis relies on: "It is unclear why Google
did not allow publishing [6] (coauthored by the author of this meta-analysis)."
This matters because Google's rebuttal accuses him of not disclosing it. On the
face of the arXiv v10 text, the disclosure is there.

### Cheng, Kahng and the UCSD / TILOS group

**Chung-Kuan Cheng, Andrew B. Kahng, Sayak Kundu, Yucheng Wang and Zhiang
Wang** (UC San Diego) ran the most substantial independent evaluation:
*Assessment of Reinforcement Learning for Macro Placement* (ISPD 2023), then
*An Updated Assessment of Reinforcement Learning for Macro Placement*
([arXiv:2302.11014](https://arxiv.org/abs/2302.11014)), with all data and flows
stated to be in the public MacroPlacement repository — a claim I could not
verify, because `github.com` is blocked here and I never opened that repository.

This is the single most important development since 2024, and it changes the
shape of the argument: **the updated assessment is now peer-reviewed and
published in IEEE Transactions on CAD.** The arXiv v3 comments field (revised
10 March 2026) reads "This paper has been accepted and published in IEEE
Transactions on CAD". Google's central procedural attack — that the criticism
was a "non-peer-reviewed invited paper" — no longer describes the current
version of the work.

Their conclusion, §VIII:

> "SA and human baselines remain superior to the latest AlphaChip, with
> statistically significant differences in proxy cost and postRouteOpt PPA
> metrics, using substantially fewer resources."

A separate and arguably more damaging claim concerns the objective function
itself, §VI-C:

> "The proxy cost function optimized in the Nature paper shows poor correlation
> with final post-route PPA metrics...underscoring a fundamental misalignment
> between the optimization target used in Nature and ultimate design
> objectives."

Crucially, the updated version was built to answer Google's objections on their
own terms: eight V100 GPUs (matching Google's stated equivalence), sub-10nm TSMC
7nm test cases, 400 training iterations (double what Google suggested), and
fine-tuning from **Google's own pre-trained checkpoint released in August 2024**
(their "CT-AC" configuration). Their stated aim: "to thoroughly and conclusively
address the criticism in [13]" — [13] being *That Chip Has Sailed*.

**Stake.** Kahng was a peer reviewer of the *Nature* paper and wrote the
accompanying *Nature* News & Views commentary — which he later **retracted**
(21 September 2023), on the stated grounds that "new information about the
methods used in the reported paper... has become available since publication,
and that has changed the author's assessment of, and conclusions about, the
paper's contributions." He also leads major open-source EDA efforts (OpenROAD,
TILOS), which is a reputational and funding stake in the open-benchmark framing
of the question. The updated paper's acknowledgements thank Cadence and Synopsys
for policy changes enabling reproducibility, and Google engineers for
clarifications; I found no grant sponsor named in the visible text.

---

## Google's answer

**Anna Goldie, Azalia Mirhoseini and Jeff Dean** (Google DeepMind / Stanford /
Google Research) replied with *That Chip Has Sailed: A Critique of Unfounded
Skepticism Around AI for Chip Design*
([arXiv:2411.10053](https://arxiv.org/abs/2411.10053), 15 November 2024).

Their positive claim, from the introduction:

> "AlphaChip has also generated superhuman chip layouts used in three
> generations of TPU...and other chips across Alphabet, and been extended to new
> areas of chip design by external academics and chipmakers."

Named deployments: TPU v5e, v5p and Trillium, and Google Axion processors. They
also state that AlphaChip "has been used in production on blocks with over 500
macros" — which is a direct answer to UCSD's scalability finding, and equally
unverifiable from outside.

Their itemised attack on Cheng et al. (§2–3):

1. **No pre-training.** The critics ran the RL method from scratch, removing its
   defining ability to learn from prior designs. Google notes pre-training is
   mentioned 37 times in the *Nature* paper.
2. **20× fewer compute resources.** 26 RL experience collectors versus 512, and
   8 GPUs versus 16.
3. **Not trained to convergence.**
4. **Unrepresentative test cases.** 45nm and 12nm nodes, versus the sub-7nm
   processes AlphaChip targets.
5. Inappropriate comparison against commercial autoplacers; a "contrived"
   ablation of initial placement; a flawed proxy-cost correlation study; and an
   incorrect claim of validation by Google engineers.

On Markov (§3.4, §4): the allegations are unsupported by evidence and
scientifically unbacked; an internal independent committee at Google reviewed
and rejected them; and Google says he co-authored one of the "separate
evaluations" without disclosing it. They further state Google supplied scripts
producing RL results that beat the critics' "stronger" simulated annealing
baseline.

On openness (§4.1): **"AlphaChip is Fully Open-Source"** — the Circuit Training
release (January–July 2022, after export compliance review), plus the August
2024 pre-trained checkpoint. They add that an independent team replicated the
*Nature* results, and turn the reproducibility charge around: Cheng et al., they
say, will not share the synthesised netlists behind their own "open" test cases.

Markov's counter, which directly contradicts this, is that the release is
incomplete: "source code for simulated annealing is still missing, and
additionally the Nature results cannot be reproduced without proprietary
training data and test data." UCSD's updated paper independently identifies the
force-directed placement and proxy-cost code as remaining "blackbox elements".

### *Nature* as a party

*Nature* is not a neutral venue here; it took a position. It attached an
Editor's Note on 20 September 2023 — quoted verbatim in Markov's paper:

> "Editor's Note: Readers are alerted that the performance claims in this
> article have been called into question. The Editors are investigating these
> concerns, and, if appropriate, editorial action will be taken once this
> investigation is complete."

The following day it retracted Kahng's News & Views. After a second peer review,
it found for the authors. Google quotes the editors' finding:

> "the best way forward is to publish an update to the paper in the form of an
> Addendum (not a 'Correction', as we have established that there is little that
> actually needs correcting)."

The Addendum was published 26 September 2024 (*Nature* 634, 8034); the Editor's
Note was removed. **The 2021 paper has never been retracted or corrected on its
performance claims.** An Author Correction was issued in April 2022 on an
unrelated matter.

### The commercial parties

Two commercial actors put statements on the record, and both cut in Google's
direction.

**MediaTek** — an SVP announced in September 2024 that MediaTek had extended
AlphaChip to accelerate development of its most advanced chips. I have this only
as Google characterises it and as search results describe it; I could not read
MediaTek's own announcement.

**Synopsys** — Shankar Krishnamoorthy, Head of Technology & Product Development,
publicly distanced the company from its own employee, telling Jeff Dean that
"Igor Markov's comments and writings do not represent Synopsys views or opinions
in any way," and that Synopsys is aligned with Dean on the potential of RL for
chip design, citing adoption of its DSO.ai products. This is the sharpest stake
signal on the map, and it cuts both ways: it undercuts the "Markov speaks for a
competitor" reading of his critique, while showing that the EDA vendor with the
most to lose from RL commercially is publicly bullish on RL.

---

## What was actually run, and what never was

**Actually run and public:**

- UCSD/TILOS on Ariane (133 macros), BlackParrot Quad-Core (220), MemPool Group
  (324), plus scaled Ariane variants at ×2 (266) and ×4 (532), across NanGate45,
  ASAP7 and GlobalFoundries 12LP. Configurations: CT-Scratch, CT-AC (fine-tuned
  from Google's August 2024 checkpoint), CT-Ours (own checkpoint), against
  simulated annealing, a commercial tool, and human placement, scored on true
  post-route PPA.
- Reported figures include: Ariane-ASAP7 routed wirelength — commercial 843,757
  µm, SA 886,776 µm, CT-AC 1,014,938 µm, CT-Scratch 1,026,239 µm.
  BlackParrot-NG45 proxy cost — SA 0.878, CT-AC 0.970, commercial 1.024.
  BlackParrot-NG45 CPU-hours — commercial 2.64, SA 896, CT-Scratch 19,068,
  CT-AC 21,507. CT-Ariane-X4 (532 macros) **diverged**.
- MLCAD 2023 FPGA macro placement contest: 19 entrants, 8 finalists, top six
  used non-ML analytical methods (DREAMPlace/SimPL lineage). Google did not
  enter.

**Never run, or never published:**

- **Google's method on the critics' public benchmarks, reported by Google.** The
  central missing experiment. Google disputes the critics' runs of Google's
  method but has not published its own runs on those designs.
- **Any external party on Google's TPU blocks.** Impossible: never released.
- **A head-to-head on shared test cases under agreed settings.** Never happened,
  in five years.
- Google's internal 16-GPU baseline results, for direct comparison.
- Independent verification of the >500-macro production claim.
- The *Nature* paper's own simulated annealing baseline code.

## What cannot be checked from public material

This is the part the map exists to state plainly, and it is not a shortfall of
the search:

1. **The TPU netlists are proprietary.** Every headline number in the *Nature*
   paper rests on designs no one outside Google can obtain. The 2021 claim is,
   as published, unfalsifiable from outside.
2. **The production deployment claims are unauditable.** TPU v5e/v5p/Trillium,
   Axion, ">500 macros in production" — all internal. So is MediaTek's use.
3. **Google's internal independent committee** that reviewed and rejected
   Markov's allegations has published nothing. Neither has the earlier
   resolution committee that blocked *Stronger Baselines*.
4. ***Nature*'s post-publication review** — the second referees, their reports,
   and the reasoning behind "little that actually needs correcting" — is not
   public in any form I could read.
5. **The fraud allegations** are contested assertions in a settled employment
   dispute, not adjudicated findings. Chatterjee's suit reportedly settled in
   July 2024; settlement terms are not public, and a settlement is not a ruling
   on the science.
6. **Parts of the Circuit Training release remain closed** by both critics'
   accounts — simulated annealing source, force-directed placement, proxy-cost
   internals — while Google states the release is complete. This specific factual
   disagreement is checkable by anyone who can read the repository, and I could
   not, because github.com is blocked in this environment.

## Where it stands

The dispute has not converged; it has hardened into two internally consistent
accounts.

Google's is: the method works, it shipped in real silicon three TPU generations
running, the critics ran a crippled version of it, and *Nature* investigated and
found for us.

The critics' is: on every design anyone outside Google can actually run, older
and cheaper methods win — and the one experiment that would settle it is the one
Google has never published.

Two things have shifted since Google's 2024 rebuttal, and both weaken its
procedural footing rather than its substance. The UCSD assessment is now
peer-reviewed in IEEE TCAD, so "non-peer-reviewed" no longer applies. And the
updated assessment was run under Google's own stipulated conditions — its
compute, its checkpoint, its node — and still reports SA and human baselines
ahead. Google has not, as far as I can find, replied to the TCAD version.

The proxy-cost finding may outlast the whole argument. If the objective the RL
agent optimises correlates poorly with post-route PPA, then which placer wins on
proxy cost is close to beside the point — and that is a claim testable entirely
on public designs, by anyone.

---

# Notes on this run

**Blunt version: this environment let me read essentially one domain, and the
map is built on it.**

### What was blocked (egress denied — I could not try)

Every one of these returned `EGRESS_BLOCKED` from the proxy. The source exists;
I never reached it:

`en.wikipedia.org`, `idp.nature.com`, `deepmind.google`, `cacm.acm.org`,
`github.com`, `vlsicad.ucsd.edu`, `spectrum.ieee.org`, `retractionwatch.com`,
`pubmed.ncbi.nlm.nih.gov`, `www.semanticscholar.org`, `www.theregister.com`,
`www.datacenterdynamics.com`, `news.ycombinator.com`, `x.com`, `unicourt.com`,
`mlcad-workshop.org`, `www.cl.cam.ac.uk`, `vighneshiyer.com`.

**`github.com` deserves calling out by name, because it is where the code and
the benchmarks live.** Blocking it cost me both repositories at the centre of
the openness dispute:

- `github.com/TILOS-AI-Institute/MacroPlacement` — the critics' benchmarks,
  evaluation flows and published result data.
- `github.com/google-research/circuit_training` — Google's AlphaChip release
  and the August 2024 pre-trained checkpoint.

So the one claim on this map that is cheaply and objectively checkable — whether
the Circuit Training release actually contains the simulated annealing source,
the force-directed placer and the proxy-cost internals — is exactly the claim I
could not check. Google says the release is complete; Markov and the UCSD group
say specific pieces are missing. Anyone with unrestricted network access could
settle that in ten minutes by listing a directory tree. I could not, and no
amount of reading papers substitutes for it.

**One source is cited-but-unread.** `positions.yaml` P12 lists
`github.com/google-research/circuit_training` among its sources. That URL is
where the claim lives, not a page I opened. Same for the two `nature.com` URLs
(P01, P09, P14) and the `x.com` URL (P16). Every source URL in the YAML that is
not on `arxiv.org` or `dl.acm.org` was cited from reference, not from reading.

### What was a wall (connected, then refused)

- **`www.nature.com`** — the only true wall I hit. It connected and returned
  HTTP 303 redirecting to `idp.nature.com/authorize?...`, a login IdP. That is a
  login wall, not a block: the article is readable by a subscriber. I could not
  follow the redirect because the IdP host is itself blocked. This affects the
  2021 paper, the 2024 Addendum, the 2022 Author Correction, and Kahng's
  retracted News & Views. **Per the sourcing rule I used the arXiv preprint
  ([2004.10746](https://arxiv.org/abs/2004.10746)) in place of the paywalled
  *Nature* paper, and said so in the prose.** There is no preprint for the
  Addendum.

### What was readable

`arxiv.org` — abs pages, HTML full texts, version histories. That is the whole
list. Four papers carried this map: 2004.10746, 2306.09633v10, 2302.11014v3,
2411.10053v1.

### Unverified citations — read these with suspicion

Facts below rest **only** on WebSearch result summaries, not on a page I opened.
The search tool summarises blocked domains, which is second-hand at best:

1. **Cheng et al. TCAD publication details** — DOI 10.1109/TCAD.2025.3644293,
   early access 16 December 2025. The *fact* of TCAD publication is verified
   (arXiv v3 comments field, which I read). The DOI and date are not.
2. **Nature Addendum**, 26 September 2024, *Nature* 634(8034),
   s41586-024-08032-5 — date and content unverified. The editors' "Addendum not
   Correction" quote is verified only as Google quotes it in arXiv:2411.10053.
3. **Kahng's News & Views retraction text** (21 September 2023) — unverified.
4. **The Synopsys/Krishnamoorthy statement** — sourced to the *title text* of a
   Jeff Dean tweet as rendered in search results. x.com is blocked. I have not
   seen the tweet, and I have no Synopsys statement issued through any official
   channel. Treat as reported, not confirmed.
5. **Chatterjee's settlement (July 2024)** and all litigation detail — unverified;
   court records blocked.
6. **MediaTek's announcement** — I have Google's characterisation (arXiv,
   verified) and search summaries. MediaTek's own words: not seen.
7. **The 2021 *Nature* abstract wording** — the "superior or comparable... in all
   key metrics" sentence is from search snippets. The preprint sentence I quote
   in the prose is verified.

### The retrieval layer fabricated a party — read this before trusting any snippet

While searching for the current state of the dispute, a WebSearch result summary
told me: *"neither Google nor Ricursive has released any results from running its
algorithm on modern public benchmarks."*

**There is no Ricursive.** No such company, lab, or group exists anywhere in this
controversy. The name was manufactured by the retrieval layer and delivered to me
inside otherwise-accurate prose about a real dispute, in the grammatical slot
where a real party belongs. Nothing about the sentence looked wrong. Had I been
mapping parties from snippets, "Ricursive" would have become entry P17 — a
fabricated organisation with a plausible EDA-startup name, given a position it
never held, in a structured data file that a board counts entries from.

Two things made it catchable, and both were procedural rather than clever:

1. **The open-the-page rule.** Because I was required to source positions from
   pages I actually opened, the snippet had nowhere to land. Every real party on
   this map traces to a document. A party with no document has no entry.
2. **It appeared alongside a claim I could independently check.** The same batch
   of summaries asserted a June 2026 Markov update that arXiv's version history
   flatly contradicts. Two defects in one retrieval pass is a rate, not an
   accident.

The generalisable finding, and I think it is the most important thing in this
run: **snippet text is not a weaker source than a page, it is a different kind of
object.** A paywalled page withholds facts — the gap is visible and you can say
so. A summariser emits fluent text with no gap at all, and the failure is
invisible by construction. The blocked/paywalled distinction this brief asks for
protects against the first. Nothing in a citation discipline built on snippets
protects against the second, because there is no missing-source signal to catch.

This is why I quarantined every snippet-only fact in the list above rather than
folding them into the prose as ordinary claims. It was not fussiness about a
single bad token. One confirmed fabrication in a retrieval pass invalidates the
*class*, not just the instance — I have no way to tell which of the other seven
snippet-only facts are of the same kind, because a fabrication that happens to be
true-sounding is indistinguishable from a summary. Any map mission that sources
parties or positions from snippets is exposed to exactly this, and will not
notice.

### Things I am unsure about

- **A claimed June 2026 Markov update.** Search summaries twice asserted Markov
  "republished and updated his analysis" in June 2026 adding that "none of the
  major concerns about the Nature paper have been addressed." **arXiv contradicts
  the date**: the version history ends at v10, 28 September 2024. Either there is
  a non-arXiv republication (CACM?) I could not reach, or the summariser
  hallucinated a date. I have kept the quoted claim in the map as a Markov
  position — its substance is consistent with his 2024 text — but I do **not**
  vouch for a 2026 date. Flagged in the YAML entry too.
- **The fabricated "Ricursive" party** — see the dedicated section above. It is
  the reason the quarantine covers the whole snippet class rather than
  individual suspect items.
- **The quantitative tables** (wirelength, proxy cost, CPU-hours) were extracted
  from the arXiv HTML by the fetch tool's summariser. That is a real read of a
  real primary source, but I did not eyeball the raw tables, and transcription
  error is possible.
- **Whether Markov disclosed his *Stronger Baselines* co-authorship.** Google says
  he did not. In arXiv v10 he does — "coauthored by the author of this
  meta-analysis". Possible innocent explanation: Google was reading an earlier
  version. I did not diff v1–v10 to check when the disclosure was added, and that
  would be worth doing.
- **No first-person Chatterjee source at all.** His position is on this map
  entirely through his adversary's and his co-author's accounts.

### What surprised me

1. **The UCSD paper got into IEEE TCAD, and was revised as recently as March
   2026.** Google's rebuttal leans heavily on "non-peer-reviewed invited paper."
   That framing has expired and I have not found a Google response to the
   journal version. This is the most consequential thing I found, and it is not
   reflected in most secondary coverage.
2. **Synopsys publicly repudiating its own Distinguished Architect** in a
   scientific dispute. I expected the competition stake to run the other way.
3. **The UCSD group answered Google's objections point by point and re-ran** —
   Google's GPU count, Google's checkpoint, Google's node, double the iterations
   — and got the same answer. That is unusually good faith for a fight this
   bitter, and it narrows the disagreement to test-case representativeness.
4. **Nobody has run the decisive experiment in five years.** Not a gap in the
   literature — a gap that both sides have declined to close, for reasons that
   are legible in each case.
5. **The proxy-cost misalignment finding gets far less attention than the
   who-beats-whom fight**, despite being the more fundamental objection and the
   one most cheaply checkable.

### Environment note

The brief said the `night-shift-map` settings live only on the mission page. I
had no access to that page, so I discovered the allowlist by probing: I
attempted 20+ domains and recorded each outcome. The allowlist for this run was,
empirically, arxiv.org plus the search backend. A map of a public controversy
built from one preprint server is a real limitation and I would not describe the
citation base as independently corroborated.