Adversarial review, pass 1: summary of the four documents

This project is about learning two languages: Assembly (the A side, the native language of the machine) and Fortran (the B side, the language of mathematics). The four Gemini documents and this adversarial review are background research. Their job is to show areas that the owner did not think about.

Date: 2026-09-27 Method: I read all four documents in full, with the embedded salary images and the page scripts. I ran a rough automated count to find which listed sources the documents actually cite. The findings come from internal consistency checks and domain knowledge. I did not fetch any cited sources in this pass. Items with the tag [VERIFY] wait in the queue for pass 2.

How the four documents relate

Report (Markdown)	Interactive page (HTML)	Question it answers
01 Assembly-Fortran-Comparison.md	03 assembly_vs_fortran_learning_advisor.html	Which language to learn in 90 days for career and skills?
02 Assembly-Versus-Fortran-Comparison.md	04 retro_game_dev_language_advisor.html	Which language to learn in 90 days to build an 80s/90s-constrained game?

Scorecard

Document	Goal met?	Confidence	Critical	Major	Headline problem
01 Career report	Partly	Medium-low	3	9	Broken salary data. A LANL claim cites a Freelancer page. The milestones do not compare like with like.
02 Retro game report	No	Low	5	10	It claims Fortran advantages: array syntax, SIMD, fast floating point. These do not exist on 80s/90s DOS. The history contradicts the genre mapping.
03 Career advisor page	No (quiz) / Partly (summary)	Low	4	4	The quiz is tautological. It gives a verdict after one click. The salary chart contradicts its own source. It invents radar scores.
04 Retro advisor page	No	Low	5	5	It confuses fixed-form and free-form FORTRAN 77. It makes impossible “locked 60 fps” claims. It invents genre and CPU charts.

Problems that run across documents

1. Weak sourcing in all documents.

   • Report 01 cites only about 28 of its 56 listed sources.
   • Report 02 cites only about 10 of 27. Also, 42 of its ~81 citations point to one Reddit thread.
   • Report 02 appears to copy its bibliography from report 01. It includes malware-salary and Freelancer pages that have no relation to retro games.
   • The most-cited sources are listicles, Quora, Reddit, Medium and job-board pages.

2. Citations do not match their sentences. The documents attach numbers to claims that the source cannot support. For example, the citation for the LANL study is a hiring page. The citation for assembler error behavior is an LFortran blog post. Pass 2 should compare every high-stakes sentence with its citation.

3. Invented “quantitative” visuals. The two HTML pages show made-up radar and bar scores as analytics. Where a number has a source (salary), the chart changes it.

4. Built-in bias in the quizzes. The two quizzes put Assembly as option A every time. They show a 100% verdict after one answer. The code computes the answeredCount variable but never uses it. Neither quiz asks about the baseline of the user (prior C experience, hours per week, hardware, goal).

5. False dichotomy. Neither report seriously considers C. C gives most of the literacy benefits of Assembly. It was the dominant language for early-90s PC games. It is also the realistic bridge to the two targets.

6. The two reports reach opposite conclusions.

   • Report 01 leans toward Fortran (“highest probability of shipping”).
   • Report 02 ends on Assembly (“the essential foundation”). This overrides its own decision tree.

   The real goal of the user (career? retro game? curiosity?) decides which report is relevant. Neither report asks.

7. Overclaimed 90-day outcomes. The reports claim a “Production-ready PDE solver”, “HPC cluster deployment” and a “locked 60/70 fps bare-metal raycaster”. None of these has support for a learner who starts from scratch.

What holds up

• The framing of report 01, Assembly = reading literacy versus Fortran = numerical delivery, is broadly reasonable.
• The ISA section of report 01 (x86-64 for RE, AArch64 for Apple/cloud, RISC-V for teaching) is broadly accurate.
• The input-latency section of report 02 (hook INT 09h, do not poll INT 16h) is correct.
• The toolchain lists (NASM, GDB, Ghidra, Compiler Explorer, gfortran, fpm, fortls, OpenWatcom, DOSBox-X, 86Box) name real, appropriate tools.

Pass-2 plan

1. Fetch and check the high-stakes citations. These are the LANL Fortran report, coarrays/OpenCoarrays, LFortran status and the Microsoft FORTRAN graphics library. Also check the implementation language of Elite, Frontier and M.U.L.E.
2. Use these reviews to complete the claims tables in the per-segment worksheets in ../segments/.
3. Decide with the user which goal is primary (career or retro game). Then revise the surviving document against that goal.

Files

• 01-Assembly-Fortran-Comparison.review.md
• 02-Assembly-Versus-Fortran-Comparison.review.md
• 03-assembly_vs_fortran_learning_advisor.review.md
• 04-retro_game_dev_language_advisor.review.md