HackerNews Digest

The DeepSeek Search & Fetch APIs provide read‑only access to live web content for agents and RAG applications. Both use the base URL https://api.deepseek.com with bearer authentication (Authorization: Bearer <API Key>) and Content-Type: application/json.

Search API (/v1/tools/search) accepts JSON with query (supports operators like site:, -site:, lang:, after:, before:), count (1‑50), and need_content (boolean). It returns HTTP 200 with a results array; each entry includes rerank_score, title, url, snippet, summary, optional content, site_name, site_icon, date, and authority. Empty results are not billed.

Fetch API (/v1/tools/fetch) takes a single url field and returns the page’s title and Markdown content. Availability depends on the target site’s policies.

Billing is per successful request (HTTP 200 with data), not per token or result count. Errors follow OpenAI‑compatible format; retry on 408, 429, or 5xx, and resolve 402 by topping up. Rate limits are enforced per‑account (QPS) for each API, with an additional per‑domain QPS limit for Fetch. Server‑side tool calls via DeepSeek Harness, Claude Code, or Codex are currently free; per‑request billing begins 00:00 Beijing time on 1 Nov 2026.

Sphinx is a Git hook that adds an “understanding check” to the push workflow for agent‑assisted development. When a developer or AI coding agent pushes changes, Sphinx pauses the push and presents a scenario‑based question derived from the diff (e.g., handling Stripe webhook retries). The user must answer; an incorrect response triggers a learning loop, while a correct answer allows the push to continue. The tool operates with minimal token usage and can interface with any OpenAI‑compatible endpoint, including Codex, Claude, Cursor, OpenCode, and Pi. Answers never reach the coding agent, and environment files or secrets remain hidden. A “yolo” mode permits pushes to proceed automatically, logging any skips. Installation is via npm (npm install -g sphinx-understand) on Node 22.19+ across macOS, Linux, or WSL, after which normal git push commands invoke the Sphinx checks. The system aims to ensure developers understand the functional impact of code before it ships.

The documentary “You Can See Everything” (directed by Lance Oppenheim and Nathan Fielder) examines Elizabeth Holmes and the Theranos scandal while simultaneously illustrating how documentary form can shape audience perception. The film interleaves close‑ups, wide shots, and narrative reordering to shift Holmes’s portrayal from villain to victim, prompting viewers to experience real‑time emotional reassessment. The author notes that Holmes was acquitted of patient‑fraud charges but convicted of defrauding investors who contributed over $700 million, emphasizing the nuance often lost in popular retellings. An referenced 2015 Canon experiment—photographers receiving divergent descriptions of the same subject produced markedly different portraits—serves as an analogy for how framing influences interpretation. The piece argues that viewers first form an emotional impression, then construct beliefs, and finally seek supporting facts, a process the documentary exploits to demonstrate the ease of psychological manipulation. The author suggests the film warrants study for its demonstration of narrative control and its impact on audience cognition.

Scientists demonstrated rapid reversal of coral bleaching by physically joining bleached fragments with healthy ones from the same genet. In laboratory trials using Pocillopora damicornis, ten paired fragments (one bleached, one healthy) were set up; five pairs were allowed to touch, forming a tissue bridge, while five were separated by millimetres. In touching pairs, the bleached fragments regained color within eight weeks, with symbiont populations doubling every four days and full recovery occurring after a week of contact even when later isolated. Separated pairs showed no recovery and died after ~20 weeks. Microscopy confirmed that fusion and symbiont transfer required genetic identity; fragments from different individuals or species did not fuse or transmit algae. The findings suggest a potential, scalable method for restoring bleached corals by leveraging intra‑genet tissue connectivity to re‑establish Symbiodiniaceae symbiosis.

Lean 4.35.0 introduces the lake check command, which exports a development and rechecks it with the official leanchecker kernel inside a sandbox (optional with --inadvisably-no-sandbox). The --paranoid flag runs the same check against four additional kernels: nanoda, lean4lean, con‑leche, and con‑ron, plus a hardened‑allocator variant of the official kernel.

• con‑leche – first Lean kernel written in Lean and proven consistent (modelled in set theory per Mario Carneiro). Its consistency proof is verified by the official kernel; performance is 0.69× the official kernel, checking Mathlib in ~100 s on eight threads.
• con‑ron – Rust translation of con‑leche, refined and verified using the Aeneas tool, providing protection against bugs in the Lean compiler/runtime.
• lean4lean – verification in progress; implementation mirrors the C++ kernel, aiming for a third verified kernel.

The renovation plan, led by Arthur Adjedjis and reviewed by Joachim Breitner, aligns kernel changes with proof updates. Future work includes a more user‑friendly lake check, plug‑in support for external kernels, binary export formats, and incremental checking to scale to projects of tens of millions of lines of code. Discrepancies between kernels should be reported to [email protected].

Since 1912 more than 2,300 U.S. dams have been removed, with a surge in the past 25 years; 2024 saw a record ≈ 2,500 river miles reopened, highlighted by the Klamath River’s four‑dam removal (≈ $500 M) that restored > 400 mi of chinook salmon habitat. While western salmon runs dominate headlines, most removals occur in the Northeast, where ≈ 35,000 barriers still impede fish. Recent projects include Vermont’s Blake Higgins Dam, Maine’s Edes Falls Dam, and New Hampshire’s Bellamy and Oyster River dams. Removal of the Talbot Mills Dam (Billerica, MA) is pending. Studies by UNH biologist Nathan Furey use PIT‑tag monitoring to assess alewife passage in restored streams such as the Bellamy River. Community‑driven efforts in Plymouth, MA have eliminated multiple Town Brook dams, increasing herring returns from ≈ 25 k to > 200 k annually. Ongoing challenges involve funding, permitting, and opposition from “NOOMBY” groups, while aging hydro‑dams face safety and economic liability pressures.

DeepSeek‑V4.1‑Flash‑1017 is a minor revision of DeepSeek‑V4.1‑Flash that delivers measurable gains in reasoning, coding, agentic tasks, and visual benchmarks while reducing execution time and output length, improving the performance‑cost trade‑off for various agent harnesses. The release does not include a Jinja chat template; instead, it provides a self‑contained Python encoder (encoding.py) with test cases covering multi‑turn dialogue, tool calling, thinking mode, numeric reasoning, system messages, and interleaved images. For production, the deepseek‑recipe Rust libraries (with Python bindings) offer a protocol‑aware toolkit that converts API messages into the DeepSeek V4/V4.1 prompt format, encodes token IDs, and parses model outputs, handling thinking, tool calls, images, and generation settings while leaving inference and transport to the caller. An inference folder supplies weight‑conversion steps and local execution guidance, with recommended sampling parameters (temperature, top‑p, context window, max‑tokens). The code and model weights are released under the MIT License. Citation information and contact ([email protected]) are provided for academic referencing and support.

Pele is a JavaScript library that converts Mermaid diagram syntax into SVG graphics. It supports theme customization through CSS variables—e.g., --pele-bg, --pele-fg, --pele-muted, --pele-line, --pele-surface, --pele-surface‑alt, and --pele-border—allowing color and font changes without re‑rendering the diagram. Users import the render function from the pele module, pass a Mermaid string (such as a flowchart definition), and receive an object containing the generated svg markup. The SVG can then be inserted into the DOM, for example via document.querySelector('#diagram').innerHTML = svg;. Example usage demonstrates rendering a simple flowchart where “Mermaid text” connects to an SVG node. The library’s API is minimal: call render(mermaidCode) and embed the returned SVG.

The snippet demonstrates a C‑preprocessor technique for applying a macro to each argument in a variadic list. It defines:

• CAT/CAT_ for token concatenation.
• NARG and NARG_ to count the number of arguments, using __VA_OPT__ to handle empty variadic lists and a trailing sequence 8,7,…,0.
• A series of MAP_n macros (MAP_0 … MAP_8) that recursively invoke a user‑supplied macro m on each argument, chaining calls until the list is exhausted.
• __map selects the appropriate MAP_n by concatenating MAP_ with the count from NARG.

An example defines PRINTINT(x) as printf(#x " = %d\n", x); and calls __map(PRINTINT, 1, 2, 3, 4, x, y, z). After preprocessing, the code expands to separate printf statements for each argument, printing their names and values. The approach works for up to eight arguments and illustrates how variadic macros and token‑pasting can emulate a map operation in the C preprocessor.