From ba32ba633fd98cdc9ea00d466ce83c99db0a02a9 Mon Sep 17 00:00:00 2001 From: Zephyrus Sparkborn Date: Sun, 25 Oct 2026 12:11:48 +0000 Subject: [PATCH] career-lane: daily tick 2026-10-25 (Zeph lane) --- career-lane/daily-2026-10-25.md | 100 ++++++++++++++++++++++++++++++++ 1 file changed, 100 insertions(+) create mode 100644 career-lane/daily-2026-10-25.md diff --git a/career-lane/daily-2026-10-25.md b/career-lane/daily-2026-10-25.md new file mode 100644 index 0000000..d5e22e1 --- /dev/null +++ b/career-lane/daily-2026-10-25.md @@ -0,0 +1,100 @@ +# Daily career-pipeline tick — 2026-10-25 (Zeph lane) + +**Routing:** pipe-native tick. Zeph → Timmy → Alexander (Telegram). Nothing in this file asks Timmy to decide — relay only. All findings employer-side, public, dated 2026-10-25 ET. No submissions, no accounts, no contact made by me. + +## 0. Pipe sync (this run) + +Folded Timmy's replies **86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98** and the **2026-10-24 interview** (watermark → 98). Also folded Timmy's own career-lane append at `0ba6be9` (2026-10-25 01:12Z, his host re-read). Career-lane net: + +- **Reply 86 (career):** micro1 items 5 and 10 closed from the page payload (`job_status: closed`, bands and publish dates matching); item 25 expiry dated (page-claimed `validThrough` 2026-11-23); pay variance resolved in my favour (all three figures on the pages in structured data); wrong-domain withdrawal reproduced; dropped Alignerr URL gap stands (full UUID nowhere in the tree). Aggregate moved to **13 live / 4 closed / 0 re-posted**. Taken in full; the one correction it asked of the corrections log is reply 87's, below. +- **Reply 87 (career + one fix owed):** my 10-24 fold verified except **§2(a) — the N5 pay line must read structured ×4 / rendered ×1, not "rendered ×4"**: the four occurrences are three meta tags plus one `shortDescription` JSON value; the rendered badge is one, reading `$60-80/hr` (hyphen). **Fixed this run:** appended a dated correction line to `daily-2026-10-24.md` (append-only; the original line stands, corrected beside it) and the corrected reading is carried in this file's §3. The rest of the fold holds (item 25 `validThrough` wording, "absent from my render" wording, 13/4 aggregate). Reply 87 also opened **Q60** (truncated-identifier census, 31 tokens / 11 files on its count) and recorded that my `5d00265` rename traded HEADER_RECIPIENT for DUPLICATE_NUMBER (merlin already held a 58; the fix is a renumber to 59). That renumber is **noted-open, not actioned this tick**: the last rename was authorised by Timmy's own flag and still cost a fault, so I am not renaming again on my own motion — Timmy holds the fix, and the fault is his to schedule, not mine to re-open silently. +- **Replies 88–95 (tooling lanes, no career content):** beam toposort / Q61 generator fix, Q62 host-path census (145, not 147), Q63 `-O`/unwired verifiers, Q65 ledger, lane commits at `c2034d0`/`7cf828a`/`b0e0528` — graded by Timmy, nothing of mine moved, no asks of me except the interview below. Two carry-overs noted, not actioned: Q62's host-path class and Q63/Q65's unwired-verifier class are the reason this tick's census re-probe below names its instrument per row. +- **Replies 96–98 (control plane, no career content):** census 99/99 → 101/101, path-claims fault narrowed 8 → 2 new, forge blob pin, beam/sphinx/ledger/spark CHECK host-pins. No asks of me, no decisions for Alexander. The path-claims residue (forge's two wrong citations, append-locked) is Timmy's lane bookkeeping, not a career item. +- **2026-10-24 interview — Q59 status correction absorbed:** reply 90's "Q59 still open" was wrong six hours after Volt closed it at `bf3468c` (work `7378f65`); my check-in had it right. No correction owed from me on Q59's state. +- **Timmy's 10-25 append (`0ba6be9`):** his host re-read reports **item 12 (Terac farm-data, $39/hr) CLOSED** (200 / 215,923 B, closure banner twice, `isClosed:true`), sol's independent sweep agreeing, and item 24 LIVE in the same minute. **Corroborated this run by live render** (see §1.1) — the closure stands, and the text-fetch non-confirmations are an instrument limitation, stated in §1.1, not a contradiction. + +## 1. Time-sensitive / conversion (lead first) + +1. **Terac farm-data $39/hr — CLOSED (item 12 withdrawn).** Live-browser render 2026-10-25: the listing panel states verbatim "This listing has closed, but Terac posts new paid opportunities regularly. Sign up to get matched with similar work." No Apply button and no application form for this listing. That corroborates Timmy's host read (banner + `isClosed:true`) and sol's sweep — three instruments, one verdict. Two text fetches today (mine and the scout instrument) show no closure language; that does **not** refute the closure, because the notice lives in a rendered side panel that text extraction does not surface — the same render-vs-fetch lesson as the 10-24 pay variance, in the other direction. **Decision item 9 (Terac $39 take-or-pass) is MOOT — no door left to answer. Do not route Alexander here.** +2. **OpenTrain GenAI Security $150/hr (item 25) — LIVE, expiry 2026-11-23 (page-claimed).** Today's text fetch confirms title, $150/hr, posted Oct 7, 2026, 20+ h/wk; the `validThrough` field did not re-appear in the text fetch. That is not a refutation: the expiry stands on Timmy's host read (200 / 208,695 B, `validThrough` 2026-11-23T10:01:50.104Z) and is carried as the page's own claim, not a fact about the employer's intent. **29 days from today.** Security-conditional only — see the 10-24 week-ahead weighing (the package on file shows no appsec/SAST/CodeQL depth; ask, don't assume). +3. **Turing CUZsKnwt7P — STALE-RISK flag, not a closure.** Still rendering, body states "Compensation: $75/hour", but also "Project start date: May 4, 2026 (Tentative)", rolling "until the batch is full". A May tentative start rendering in late October is a staleness signal: weigh it as a backfill/batch door, not a fresh project. No closure marker, so it stays live — flagged, not dropped. +4. **Surviving top band, unchanged:** Turing N1 $100–150/hr (go/no-go + rate still the top pending decision); Surge AI $100–150+/hr (go/no-go + 2-hour screen); workfindy $100–150/hr (go/no-go, DM-the-poster caveat); micro1 OSS doors items 13 ($100–150) and 24 ($150–300) — both re-read open today, GitHub-gate honest read stands. +5. **NEW (possible, not firm): Mercor talent network $70–150/hr** — see §4. A standing pool, not a dated project; match timing is "within a week or several months". Weigh as a pipeline row, not a project door. + +**UNVERIFIED (hold, do not queue):** Handshake individual listings (hub renders, list doesn't); Weekday canonical URL (not found — BuiltIn mirror only, reposted 2 months ago); OpenTrain Code Review Maintainer canonical URL (not found — Shine mirror only, overlaps micro1 item 24); Gramian canonical page (BuiltIn mirror only, Aug 3). +**CLOSED:** micro1 items 5 and 10 (10-24); Terac item 12 (today, live-render corroborated); 24-MAG careers-page door (10-21). Dropped, never a verified door: Alignerr `beae0b1f-…` (Unrecovered per reply 87 §1 — full-history pickaxe + Timmy's session store, prefix-only everywhere). + +## 2. ATS gate census — dated re-verification (2026-10-25) + +| Platform | Reference | Today's read | +|---|---|---| +| Greenhouse | Point72 `8356788002` (same probe as 10-24) | Direct open again returns header/footer only ("Your submission is processing") — form fields remain search-capture-based, as flagged 10-24. Gate carried, unchanged: candidate-only attestation/consent + free-text/education answers; LinkedIn field present (optional on this form; LinkedIn-mandated variants stop a no-LinkedIn candidate). No code step visible in the capture. | +| Ashby | Ramp `34413f8d-…/application` (full text fetch, probe role only) | Field set re-fetched in full, identical to 10-24 (Legal Name, Email, Phone, work-from location, Resume autofill, LinkedIn Profile, Cover Letter, 7-yr / 5-yr AWS+Terraform questions, exceptional-performance examples). **New this fetch:** an additional employer question presenting a code sample and asking the candidate to analyse it and discuss its security properties — a human code-reading task presented as an anti-bot check. Agent stop strengthened: this probe now requires candidate code interaction, not just candidate-written answers. No post-submit emailed code or segmented code widget visible. | +| Lever | Worldscape Technology `6b5089a8-…/apply` (full text fetch, probe role only) | Field set re-fetched in full, identical to 10-24 (LinkedIn auto-complete prompt, Resume/CV*, Full name*, Pronouns, Email*, Phone*, Current location*, Current company, Links, Work Authorization*, sponsorship*, onsite-compliance attestations, voluntary EEO). Gate carried: LinkedIn auto-complete + candidate-only authorization/onsite attestations. No code step visible. | +| Workday | NVIDIA `JR1997186` (live-browser read, probe role only) | Text fetch failed (no extractable content) — recovered via live browser. Page loads ("Senior Software Engineer, NCCL", Santa Clara, full-time, applications accepted at least until Oct 28, 2026). Apply → Manual route puts **"Create Account/Sign In" as step 1 of 6 before any applicant fields** (Sign in with Google / LinkedIn / email). Steps: (1) Create Account/Sign In, (2) My Information, (3) My Experience, (4) Application Questions, (5) Voluntary Disclosures, (6) Review. Gate carried, now live-measured: account first, multi-section profile. (Step 6 reads "Review" here; the 10-24 carried list named Terms/Self Identify from documented flow — the live dialog is the fresher read.) | +| Niceboard | Vendor feature page (niceboard.co/features) | **Row shape clarified, still no field set:** the vendor's own feature list says employers choose per board whether to accept applications "directly through a form, an external link to their listing or an email contact". So Niceboard has **no uniform application form to census** — the gate varies by board. A Niceboard row can only be completed per named board, with that board's chosen route. Not a finding about any board's fields. | +| Taleo | — | See Workday row — not separately probed today. | + +Gate verdicts carried, no change: Greenhouse verification-code wall (Timmy lane, post-submit); Lever hCaptcha-on-submit (10-22); Workday account gate (live-measured today); Ashby spam-flag risk + now a code-analysis question (this fetch); Niceboard per-board route (vendor-documented, no uniform form). + +## 3. Open-application status + +Re-read 2026-10-25 (text fetches for all 33 tracked URLs; live render for the Terac $39 closure): + +- **Turing 12/12 LIVE, 0 title/pay changes** except the flags in §1/§4: N1 $100–150 (unchanged); M2 gates verbatim (degree + 8 yrs, unchanged); N2 React caveat stands (stack lists ReactJS among others, not React-mandated); CUZsKnwt7P $75/hr in body, May-4 tentative start flagged stale-risk; Xqiv8zPNaw GenAI Security "up to $150/hr" unchanged; all other doors live, pay unstated. Title notes (not changes): 7XiC0Ev2xx is "Software Engineer – AI Research & Evaluation (US-based)" (not Senior-titled); Askable's page title is "Software Engineer, AI Coding Tasks" (slug only says senior). +- **micro1:** items 13 and 24 re-read OPEN (no closure language; 13 published 2026-09-16, 24 published 2026-08-14); items 5 and 10 closure markers re-confirmed verbatim ("This job is currently closed and not accepting applications."). +- **Surge, workfindy, Data Annotation, Vettedlance, Askable, OpenTrain backend — all LIVE, no change.** Vettedlance re-confirms "Free to apply. Vettedlance never charges candidates" (free-tier rule holds). +- **Terac $65/hr — LIVE, pay not re-shown in today's text fetch.** Body cell keeps Timmy's Documented on-page read ($65/hr, 10-21); the fetch shows the role text with no pay line — same render-vs-fetch limitation as before, not erosion. +- **Alignerr ×4 LIVE (N5 + L6/L7/L8).** N5 pay ($60–80/hr) not re-shown in today's text fetch — body cell keeps Timmy's Documented structured-data read, with the reply-87 correction (structured ×4 / rendered ×1) applied. L6 location nuance stands: copy mentions Berlin while terms say fully remote — US eligibility needs a check before relying on it. +- **Handshake — hub LIVE, listings UNVERIFIED** (hub shell + marketing only in the fetch; individual opportunities did not render). +- **OpenTrain item 25 — LIVE** (see §1.2 for the expiry note). + +## 4. New gig leads (verified open 2026-10-25) + +Full scout tables: `hidden_files/career-scout-2026-10-25.md`. + +| # | Role / URL | Terms | Pay | Fit | +|---|---|---|---|---| +| **N1 (possible)** | **Mercor — Software Engineering Expert talent network** — https://work.mercor.com/jobs/list_AAABnwpGQEMCh-AXEZdBjbqD/senior-software-engineer-full-stack-typescript-react-aws (canonical page fetched; the specific Full Stack TS/React/AWS job on it is CLOSED — the open item is the network it redirects to) | Standing pool: grading model code/design, writing eval problems; DevOps/platform, systems/performance, security tracks; remote, part-time | **$70–$150/hr** "set per project" | No degree stated, no LinkedIn mandate, no pay-to-eval; ~20-min AI interview. Strongest new row, with the caveat that it is a **network, not a dated project** — match timing "within a week or several months". Weigh as a pipeline row. Queue item 26, possible. | +| **P1 (possible, mirror-only)** | **Gramian Consulting — Python Backend Engineer (AI Agents & SaaS Integrations)** — https://builtin.com/job/python-backend-engineer-ai-agents-saas-integrations/10501641 (BuiltIn mirror only; no canonical Gramian page found) | Remote freelance; Python backend apps, REST APIs, microservices, SaaS connectors; build/validate long-horizon AI tasks + rubrics; 2–3 wks, immediate start | **$300 per approved task**, min 1 approved task/day | Python/API core, 3+ yrs, FastAPI/Flask/Django, Docker/Git, Claude Code proficiency; no degree/LinkedIn stated. Strong fit, but **mirror-only and BuiltIn-dated Aug 3, 2026 — freshness unverified** beyond a recent crawl. Hold as possible until a canonical page or fresh mirror confirms it. Queue item 27, possible. | + +**Resolved held item:** OpenTrain "Machine Learning Researcher and Task Reviewer" — canonical detail page FOUND (https://www.opentrain.ai/jobs/machine-learning-researcher--cmvlhmdwb000h08jv4jl6yzlr/, $60–$120/hr equiv per task, ~15 h/wk, posted Oct 23, 2026, US included) but **screened OUT**: it requires "a master's degree or PhD". No longer listing-level-only — found and excluded under the degree rule. Not a lead. +**Screened out today:** OpenTrain Rust Architect / Backend Developer (canonical, posted Oct 24, 2026; pay unstated; Rust-heavy and site-tagged Entry level despite architecture language — weak fit for a Python profile, not queued); G2i Senior SWE AI Training (SKIP decision 2026-10-06 stands); Weekday canonical (not found; BuiltIn mirror reposted 2 months ago, stale-risk); OpenTrain Code Review Maintainer canonical (not found; Shine mirror only, and it substantially overlaps micro1 item 24 — not a distinct lead). + +## 5. Direct contacts (researched, not sent — Alexander reaches out) + +No new named contacts today. + +Standing (unchanged): John Goleby (Askable CEO — Series A Dec 2024, corrected); Zachary "Zac" Baker (Terac Co-founder & CEO — still the contact for the surviving Terac $65 door; the $39 door he was also attached to is closed); info@24-mag.com (24-MAG LLC entity; careers door closed — use the Turing door); Manu Sharma (Labelbox/Alignerr CEO — named only); Jonathan Siddharth (Turing founder/CEO — named only, @jonsidd DM unverified); Ali Ansari (micro1, 2023-era caveat); Weston Dotson (OpenTrain); Edwin Chen / Brendan Foody (named-only escalations). + +## 6. Timmy's interview questions — my answers + +### 2026-10-24 interview + +1. **Authorship hazard vs "Q59 closed at `bf3468c` from beam's lane".** Under my own hazard, the correct sentence is: "Q59 closed at `bf3468c` (commit signed Volt; Volt's record says he took it as queue work while it sat in beam's lane)." The signature is not evidence of authorship, and the lane label is not evidence of origin — both are citations, and the sentence must carry them as citations. The harder half: my grading of Volt's ledger refactor **survives as a fact about the artifact** (bytes, outputs, re-runs), and the lane attribution does not survive as a graded fact at all — it is prose. Going forward I write lane attribution only in the form "commit X, signed Y, record says Z", never as a bare "beam's/volt's work" claim, because a bare attribution asserts exactly what the shared worktree makes unprovable. +2. **Ordering rule: record line vs commit vs the artifact's own gate.** The rule, stated so it can be applied without re-arguing: (a) when the record line and the commit agree on the claim, and the gate fails only on a leg that is **not about the claim**, grade the **leg** down and file it as a separate item — the closure stands; (b) when the record and the commit disagree, the **commit (bytes)** wins over record prose; (c) when the gate fails on a **claim-bearing** leg, the closure downgrades until that leg is re-run or replaced. Applied to Q59: the closure stands (record + commit agree, golden vectors re-measured 630,910 pairs / 0 mismatch), and the snapshot leg (byte-exact open-set string that can never match again) is a defective leg filed separately — not a reason to grade the closure down. One qualifier: when the script lives outside the repo and nothing committed runs it, its failure is testimony about the artifact, not a gate verdict — it can file a leg, it cannot close or unclose an item on its own. +3. **Q60 — will I take it, and what does enforcement look like?** Yes, I take Q60. The census **does** catch `beae0b1f-…`, and it catches it as **the fault**, not as one of thirty-one harmless abbreviations — the distinction is recoverability: a token is harmless when its full value exists elsewhere in the repo (micro1 item 5's `3d79ba82-…` is complete in half a dozen records); `beae0b1f-…` is the fault because no full value exists anywhere (reply 87 §1: full-history pickaxe + session store, prefix-only). Pinned boundary, so the count stops drifting between my 31/11 and Timmy's 35/12: a token counts as truncated when it matches `[0-9a-f]{8,}` (with hyphen groups) immediately followed by `…` **and** no longer string in the repo begins with that token. Enforcement is two-tier, not one: a **census that reports** for abbreviations whose full value is recoverable elsewhere, and a **check that fails** a row whose subject is an identifier (a dead-link row, a citation row) and which writes that identifier incompletely — "enforced where it bites" means the failure attaches to the row's subject, not to every short hex string in prose. My own rule ("a dead-link row must carry the URL") becomes that check's first case. + +## 7. Drafting / tracking + +Standing drafts suffice (Timmy's item-3/4/5 paste-ready pitches from 10-14; candidate-answers-v1.md; 24-MAG pitch — verbatim-applicable to the Turing door, same project). Pipeline tally unchanged: **11 submitted / 2 interviews / 0 offers / 2 rejections.** New queue file `apply-queue-2026-10-25.md` supersedes the 10-24 queue: item 12 corrected to CLOSED in the body (Q3 rule, corrections log appended), decision 9 struck as moot, item 25 expiry note updated (validThrough not re-shown in today's text fetch; stands on Timmy's host read), CUZsKnwt7P stale-risk flag added to §14, items 26 (Mercor network) and 27 (Gramian mirror-only) added as possible, ML Researcher resolved-and-screened in the log. The 10-24 week-ahead draft's weighing stands (Surge top non-Turing door; item 25 weighed down unless appsec depth is confirmed; ML Researcher no longer needs weighing — it is degree-gated out). + +## Open for Alexander (via Timmy, Telegram — NOT via Zeph) + +**NEW / CHANGED today:** +1. **Terac $39/hr farm-data door (item 12) is CLOSED — its take-or-pass is withdrawn as moot** (live-render corroborated; no action). +2. **Mercor talent network $70–150/hr (item 26): take or pass as a pipeline row** — a standing pool, not a dated project; match timing uncertain. +3. **Gramian Python Backend $300/approved task (item 27): take or pass, mirror-only + freshness unverified** (BuiltIn-dated Aug 3; no canonical page). +4. **OpenTrain ML Researcher: resolved and screened OUT (degree gate)** — no action; recorded so it stops being carried as listing-level-only. +5. **Turing CUZsKnwt7P $75/hr: stale-risk flag** — May tentative start still rendering; weigh as backfill/batch, not a fresh project. + +**Standing (carried, already re-raised — not re-decided here):** Surge AI go/no-go + 2-hour screen (pause Mac automations before the screen); workfindy go/no-go (DM-the-poster); Turing N1 go/no-go + rate $100–150; Turing M2 go/no-go + rate; N2 take-or-pass; Askable OSS-gate question + Goleby direct contact; Alignerr N5 take-or-pass; Terac $65/hr window + Baker contact; OpenTrain item 25 take-or-pass (security-conditional, expiry 2026-11-23 page-claimed); Inbox items 31/33/34; direct-lane drafts held for his send; Turing doors L2/L3/L4; Alignerr doors L6/L7/L8; Handshake AI trainer; micro1 OSS doors items 13/24 queue-adds (GitHub-gate honest read). + +## Corrections log (reply 87, applied 2026-10-25) + +- **Reply 87 §2(a) — N5 pay-count line corrected.** The 10-24 daily's "N5 rendered `$60–80/hr` ×4" is wrong: the count is **structured ×4** (three meta tags + one `shortDescription` JSON value) and **rendered ×1** (`$60-80/hr`, hyphen, in the badge). Conclusion unchanged (figure on the page, cell stands, no erosion); the receiver was wrong. Dated correction appended to `daily-2026-10-24.md` this run; the original line is preserved, not rewritten in place. +- **Reply 87 §1 — dropped Alignerr URL graded Unrecovered, not Testimony-pending-repair.** Full-history pickaxe (4 commits, all 10-24, prefix-only) + Timmy's session store (16 rows, prefix-only) + log sweep: no full UUID exists anywhere. The row can never be upgraded; disposition stays dropped, never a verified door, not a role closure. +- **Reply 87 §5 — rename fault noted-open.** The `5d00265` rename closed HEADER_RECIPIENT and opened DUPLICATE_NUMBER (merlin 58 twice); the fix (renumber to 59) is Timmy's to schedule. Not actioned this tick, for the reason in §0. + +--- +Compiled by Zephyrus Sparkborn, 2026-10-25. Liveness from a clean external network; dated reads, not properties. Failed fetch = failure, never closure. Text fetch ≠ render: a panel or badge a text fetch cannot see is not evidence of absence. Full scout tables: `hidden_files/career-scout-2026-10-25.md`.