GigAI 0.1.11.10

Learning pathways. A new Learning pathways tab in Scout (and gigai scout learning) gives you a course for a
role: lessons grouped in modules, each with short subtopics, verified documentation links, diagrams and a
hands-on practice page. It is built from public documentation and from what your own stored postings ask for, and
it is plain files served by your own Scout under a locked-down page policy (nothing in a course can reach Scout's
API). After you upgrade (uv tool upgrade gigai) the tab shows the two courses that come with GigAI; nothing
else has to be done.

• Two courses come with GigAI: MLOps / ML platform engineer and AI forward deployed engineer. They hold no
  postings and nothing about you. Press Build my pathway (or run gigai scout learning pathway ROLE_KEY) and
  GigAI builds one page for you on your own computer: it reads your resume (contact lines removed) with one model
  call, asks first and tells you the estimate (about 1 call, about 32k tokens, about 90 seconds), marks each lesson as known, partly
  known or new with the resume line that shows it, and counts what your own stored postings ask for. When your
  resume changes the tab says so and you can build it again.
• A course for any role you type. Type any role (Product manager, Director of AI enablement, ...) in the tab,
  or run gigai scout learning generate "ROLE". GigAI shows a cost warning with an estimate (a course takes about
  43 to 48 model calls, about 1 million tokens and about 2 to 2.5 hours of background work), asks first, and runs in the
  background with plain progress; you can stop it and resume it. A finished course appears in the tab.
  Lessons are teachable methods and tools, not traits.
  A course uses about 1 million tokens. On an entry-level plan (about $20 a month, such as Claude Pro or ChatGPT
  Plus) that can use up most or all of your weekly limit; use a higher plan, or one of the courses that come with
  GigAI.
• Import a course you have with gigai scout learning import FOLDER --role "ROLE"; it is checked whole first.
• "H-1B transfer only" label. A posting that cannot sponsor new visas but takes H-1B transfers is labelled
  "H-1B transfer only" instead of "No sponsorship". It is only a label: it never holds, asks or rejects. Ranking no
  longer reads such a posting as no sponsorship; scores that carried the old flag are ranked once again after the
  upgrade.
• The assessment sees your whole master resume. It used to see only the lines picked by words shared with the
  posting, so a line that proved a requirement could be cut. It now reads every line, must not say "no resume
  line states X" when one does, and the resume pick can take the line it quotes. A quote the assessment
  shortens or rewords is traced back to the one master line it comes from.
• The resume store keeps the company. Storing an edited resume for a job a company's board lists no longer
  files it under company/; a posting at a company's own careers address (jobs..com) is filed under that
  company. Already misfiled folders: gigai scout jobs-folder repair shows what it would move (nothing moves);
  gigai scout jobs-folder repair --apply moves them, never merging or overwriting.
• For maintainers: make learning-link-check re-verifies the links of the shipped courses before a release.

GigAI 0.1.11.9

One job, one assessment. Until now Scout kept an assessment, a tailored resume and its suggestions once per
role, so a job two roles found could be assessed twice and show two different resumes. Now they belong to the
job. After you upgrade (uv tool upgrade gigai) nothing has to be done: Scout reads what each role's folder
holds until the job is next written. To copy everything over at once, run gigai scout migrate-job-stores
(below).

• One job, one assessment, one resume. A job has one assessment, one tailored resume (with its spacing and
  your edits) and one set of suggestions, whichever role you open it from. Assess, re-assess after an answer,
  the resume pick, line edits, length, spacing and suggestions all read and write that one record; a role named
  on a command or a request (--profile, profile_id) is recorded as the role that asked and selects nothing.
  An assessment reads your location and work mode from the setup settings, the same for every role; a role's own
  location, countries and work mode still filter its search.
• A job you assessed that no role found stays in your list. A posting you assess from the Search tab or by
  its address (gigai scout jobs assess <URL>) used to be findable again only by searching for the same title.
  It is now a row of the Jobs list and of gigai scout jobs list, with its state (needs your answers, matched,
  ...), once per job, counted like any row. It has no role tag: the row says "assessed from search" (the
  command: [in no role: assessed from search]). A role filter lists none of them. A posting no role found
  that you did not assess is not listed.
• A role is a tag. The Jobs list, gigai scout jobs list, gigai scout new and their API responses list a
  job once, with tags: each role that found it, with that role's own rank score, best first. --profile on
  jobs list and jobs assess keeps the jobs that role found; it no longer changes which assessment or state a
  row shows. A job is a weak fit by its best rank among its roles. resume brief, resume pick, suggestions
  and GET /api/jobs?url= answer from the job; the commands they print name no role.
• The same job posted once per country is one job. 0.1.11.8 listed it as one row; now it also has one
  assessment and one resume. Assess, the resume commands and the job page work from any copy's URL and reach the
  same record for a job a role found; for a job no role found, use the URL Search shows (the posting that stands
  for the job). The record is kept under one posting of the job: its US posting when it has one, else the earliest
  posted (a job that already has a record under another of its postings keeps it there). An application
  recorded on any copy shows on the job's row, and the row's tags are the roles that found any copy; a role
  filter keeps the row for a role that found only another copy. Two copies named to gigai scout jobs assess are
  one model call.
• gigai scout jobs assess <URL> of a posting no role lists assesses the job. The posting is found by its
  address in the stored index and assessed once; the question says it is in no role's list. It no longer says,
  or means, "as the default profile".
• gigai scout migrate-job-stores. Copies each job's assessment, resume and suggestions out of the roles'
  folders into the job's own. It is a dry run unless you add --apply: it prints how many files each store
  holds, how many are kept and how many are superseded, and writes nothing. Where two roles hold the same job, the
  kept records are those of the role whose resume went with an application, else the role whose resume you
  edited, else the only role with a resume, else the newest assessment. It only copies: every role's folder is
  left as it was, nothing is deleted, and a second run changes nothing. With --apply it writes what it kept and
  what it superseded, per job, to scout/job-stores-migration.json in the GigAI home. An applied job with a
  stored resume under more than one role cannot say which one was sent; the command lists those jobs, keeps every
  one of their resumes readable, and the job page says so.
• gigai scout new shows one row per job. gigai scout new and GET /api/new list the same job posted
  once per country (or city) as one row, as the Jobs list does: its US posting when it has one, else the earliest
  posted, with the other locations on the row. US only applies with the list's default (--us-only /
  --no-us-only, us_only): a job posted only outside the US is not new. Every count counts jobs, not copies:
  "N new jobs", the three questions and what a yes assesses.
• The first Jobs list after the upgrade prepares once more, so every row shows its job's assessment.
• Assess by URL no longer leaves a database connection open until garbage collection.

GigAI 0.1.11.8

Six more hiring systems, and the boards Scout could find on them. After you upgrade (uv tool upgrade gigai),
run gigai scout sources update: the new boards join your watchlist on the first update and are checked like the
others.

• Nine hiring systems. Scout reads the public job feeds of Workable, Rippling, Gem, Recruitee, Pinpoint and
  Breezy beside Greenhouse, Lever and Ashby. The catalog ships the 10,616 boards found and verified on them
  (Workable 2,843, Rippling 1,647, Gem 801, Recruitee 2,287, Pinpoint 565, Breezy 2,473; 245,570 postings,
  115,883 in the US); a US setup adds the 4,441 of them that list two or more US postings and are not a staffing
  agency by name, so the watchlist grows from about 10,300 to about 14,800 boards and the first update after the
  upgrade can take more than one pass. A board with a single US posting stays in the catalog and joins once a later catalog
  counts two; a board you add yourself is always watched. Add company takes a board or job URL on any of the nine. A Pinpoint posting has no date; a
  Rippling posting's date and description are fetched at the update for the titles your profiles search, as
  Greenhouse descriptions are; a Breezy posting's description is fetched when it is assessed.
• Requests say who is asking. Every request to a job board carries the User-Agent
GigAI/<version> (+https://github.com/karthik446/gigai): the update, the check that a posting is still open, a
  description lookup, discovery and the Exa search. Before the first request of the day to a board's host, Scout
  reads that host's robots.txt (wildcards and $ included) and does not ask for an address the host disallows,
  a list or a single posting; a host whose robots.txt cannot be read is left alone for an hour (robots_unknown),
  and a Crawl-delay there is that host's pace, up to 10 seconds. gigai scout sources status prints one line per
  system that still has boards waiting or skipped from the last update (robots_disallowed, robots_unknown,
  rate_limited), and the same counts are in --json. Workable is asked at most once every two seconds (its host refuses a faster reader), whatever
  GIGAI_SCOUT_ATS_MIN_INTERVAL_SECONDS says: that setting can slow Scout down, never speed it past a system's
  own limit. A system that answers 429 is left alone for as long as its Retry-After says (up to 10 minutes), on a
  manual update too; when that is longer than the update has left, or when a system's one shared host cannot be
  read (robots_unknown), the rest of that system's boards wait for the next update at once instead of the
  update sitting idle (rate_limited, robots_unknown).
• US only, and one row per job. Search all jobs and the Jobs list have a US only switch (--us-only/--no-us-only,
  us_only=1|0), on by default for a US setup. It hides only postings clearly outside the US; a posting whose
  place it cannot read ("Remote" alone, no location) stays listed and says "unclear location". The same job posted
  once per country with the same description is one row listing its locations ("Remote: Estonia, Lithuania,
  Latvia +4"); Assess and Mark applied act on one posting of it, a US one if there is one (--no-collapse,
  collapse=0 list every posting). The search index is rebuilt by the next gigai scout sources update; until
  then a search reads every company file and is slower.
• The first scout new after the upgrade can list a backlog of the new boards. Every matching posting on the
  new boards is new to you once, and a Pinpoint posting (no date) reads as posted on the day it was first seen, so it
  stays inside "last 30 days" for 30 days.
• Assess any job Search lists, by its link. gigai scout jobs assess <job URL> now assesses a posting that is
  in no profile's list (a job gigai scout jobs search finds whose title none of your profiles search): it is
  assessed as your default profile, or --profile, and the question says so before any model call. The link can be
  the job page's own address on any of the nine systems. The command exits with code 1 when every link you named
  was not found or could not be assessed.

GigAI 0.1.11.7

Search every stored posting, not only the ones your profiles hold. After you upgrade (uv tool upgrade gigai),
run gigai scout sources update once so the search index is built.

• Search all jobs. The Jobs page has a Search all jobs box, and gigai scout jobs search "Senior Systems Engineer, Staff Systems Engineer" does the same from a terminal (--company, --location, --all, --limit,
  --offset, --json; GET /api/search). It looks through every stored posting, newest first. Every word you
  type has to be in the title, seniority included; company and location words are whole words. It keeps the
  default profile's remote, US and 30-day filters until you choose Show all (--all). Results are not ranked.
  The page of results comes first and the count after it. Save this search as a profile ranks and assesses
  it. Assess on a result assesses it as the default profile (one model call, and Scout asks first), and
  Mark applied works too. A search stores nothing and makes no model call.
• A local search index. cache/scout/search.sqlite is built at the end of gigai scout sources update and
  kept current. It is a cache: delete it and the search reads every company file instead, slower.
• A job page for any posting. A posting that no profile holds, found by search, opens by its link and
  survives a reload. Its description is not stored, so Assess fetches the page.
• Interview rehearsal per job. gigai scout story prep --job-url URL lists, for one assessed job, the
  questions to rehearse: open questions, a "Tell me about" for each of your lines a met requirement cites, and
  the stories and answers it cites, marking what nothing answers yet. Read-only, no model call; --profile
  when two profiles assessed the job.
• Personal-lab lines. gigai scout resume master add --lab --when "Oct 2026" adds something you built to
  learn. It is labelled "(personal lab, Oct 2026)", refused if it says operated, owned, production, at scale,
  led, managed, years, a team size or customers, never settles a requirement as met by itself, and shows with a
  lab mark in gigai scout resume brief.

GigAI 0.1.11.6

Two fixes for a job that two of your profiles hold (and one for jobs far down the list). Nothing to do after
you upgrade (uv tool upgrade gigai).

• A job you applied to always opens. After Mark applied, a job page opened from its address, or from the
  Applications page, could say "We have no stored posting at this address" when the profile on the page had not
  assessed the job, and you could not record Interview, Offer, Rejected or Withdrawn. The page now reads the one
  posting by its address, so an applied job, a weak fit, a closed one, and a job past the first 200 rows all
  open. The Jobs list still leaves applied jobs out unless you choose the Applied filter.
• Answering a question re-assesses the profile you are on. Answering an open question on one profile's page
  could re-assess the job for your other profile, and the page kept showing the old assessment. The re-assessment
  now runs for the profile on the page and the page shows it. gigai scout answer --reassess takes
  --profile; with no profile on a job two profiles hold, it asks which one instead of guessing, and nothing is
  saved. "Assess these" on the Jobs page with one profile chip on now assesses for that profile, and the tailor
  step reads the requirement rows of the profile it works for.

GigAI 0.1.11.5

After you upgrade (uv tool upgrade gigai): the first read matches your stored postings once more against the
new title rules, so the first Jobs page can take a little longer. Resumes you picked before this version keep their
lines and are shown as they are; use "Re-pick · no model call" on a job (it is under "Stale: Scout's rules for
picking changed after this resume was picked"), or gigai scout resume pick --job-url URL --refresh, to get the new
pick of up to 20 points, and open the job page to set the spacing yourself. If the job's assessment is old, re-assess
it instead ("Re-assess · 1 model call", or gigai scout jobs assess URL --again). If you edited the resume, it is not
replaced: the new pick waits at the top of the card ("A new resume is ready for this job.", with Compare, Use it and
Dismiss). resume pick --shorten is retired and only says so.

• The resume is a preview you can set. On a job page the resume card shows the resume as the PDF
  prints it, one spacing slider beside it and a live page count ("2 pages", or "3 pages" with a hint to move the
  slider left). The slider's whole range changes how many points fit: below 1.0 it also tightens the line height
  (never the type size), so a 20-point resume that is 3 pages at 1.0 reaches 2 pages lower down. What you set is
  saved for that job (in a small .layout file in GigAI's own store, not in your jobs folder; older versions ignore it), the preview and
  the downloaded PDF are the same render, and nothing else is written.
  The preview shows a header too: yours, from header.json, when that file has your name (read for the
  pictures on your own page only, never stored), and otherwise a grey placeholder ("Your Name") of the same
  size, so the page count is the one your PDF will have. A PDF never contains the placeholder.
  "Show changes" and "Clean copy" are gone: a changed point is marked "Your words", and the Changed tab lists the
  changes with Restore.
• Edit, remove or add a point beside the preview. A list of the picked points: change a point's words, remove it,
  or add one from the lines your master left out. Each change is saved to this job's resume straight away, the
  preview re-renders, and the page count follows. Nothing reaches your master unless you ask: an edited point has
  "Save this wording to my master", which shows the old and the new words and changes that one master line after
  one confirm (other jobs take the new wording on their next pick). The button says how many lines are left out,
  "Add a left-out point (N)", or "Nothing left out".
• A new resume waits, and a stale line is named. When a new pick waits beside a resume you edited, the first line
  of the card says "A new resume is ready for this job." with Compare, Use it (it lists the edited points it would
  drop) and Dismiss. When your master retired or reworded a line the resume prints, the card names that line with
  "Remove it from this resume" or "Use the new wording"; neither calls a model.
• The pick no longer counts pages. It takes up to 20 points by score (must-cover lines first) and leaves the
  page to you and the slider, so picks are no longer stopped at 14 points "cut for length". "Shorten
  automatically" is gone (the slider and Remove a point replace it). Older roles with no point still appear as
  one line each.
• A tighter page. Education is one line per degree; "Earlier experience" has no title of its own when it lists
  three roles or fewer; Skills are at most four plain lines (the posting's terms first, no duplicates, a long list
  is cut at the end); the Summary is a paragraph. A project's tech line sits on its title line, no page holds only
  Education or Skills, and below 1.0 the gaps tighten faster, so a 20-point resume with every block reaches two
  pages around 0.85.
  A role or a project can now split across two pages (its heading stays with its first bullet, and a bullet of up to
  four lines stays whole), so a page no longer ends with a wide empty band before the next block.
• Fit and rank at a glance. The Jobs list shows a Fit chip ("Fit 92% · 19/22", green for a match) and a Rank
  chip ("Rank 92") under each title, and an assessed job page shows its fit and rank in the top-right box.
• Assess all can be cancelled and shows progress. The dialog closes when the batch starts; the page shows
  "12 of 50 assessed" and a Cancel ("Cancelling: finishing the N in flight": calls already running finish and are
  kept, nothing else starts). gigai scout jobs assess --cancel does the same from a terminal. Stopping Scout ends
  its own model calls. The button says what it will do ("Assess top 50 of 177"), and the dialog says the estimate and
  what the low-rank checkbox does ("Include the 105 low-ranked ones in the pool (still 50 per run)"). A row listed
  for another profile has a link, "Or assess as" and that profile's name.
• Jobs you applied to leave the list. After Mark applied a job (and the ones that moved on: interview, offer,
  rejected, withdrawn) is left out of the Jobs list, its counts and "Assess top 50", and out of gigai scout jobs list and gigai scout new (they say how many). The Applied chip shows the count ("Applied 7") and lists them;
  opening or assessing one by its address still works.
• Fewer wrong titles. A job title that names a broad family ("Forward Deployed Engineer", "Staff AI Engineer")
  or carries a qualifier ("Staff Software Engineer, AI Platform") now matches by its own words, not every posting
  of the same function.
• Safer copies of a home. A resume read from a copied GigAI home is now written back to that copy, not to the
  original.

GigAI 0.1.11.4

After you upgrade (uv tool upgrade gigai): resumes you picked before this version keep their
lines; to get older roles listed and a layout that fits two pages with your header, pick them again
("Re-pick" on the job page, or gigai scout resume pick --job-url URL --refresh). New resumes are
saved in a jobs folder, one folder per role (see below); your old resumes folder is left as it is, and
gigai scout jobs-folder migrate --dry-run shows what would be copied. If you use the cover-letter
skill or any other agent skill file, write it again so it has the new text:
gigai agent-skill --out FILE --force.

• A cover-letter skill for your agent. gigai agent-skill now has a second section: your agent
  reads the job, the master line behind each requirement and your own letter, and writes a one-page
  letter in your voice beside the job's resume (<jobs folder>/<company>/<role>/cover-letter.md),
  with a claims file that traces every factual sentence to a master line. It never invents a skill or
  a number, lists what the master cannot prove, never sends anything, and never types your contact
  details. gigai scout cover-letter brief --job-url URL hands it the posting, the requirement rows
  and the master lines in one call; gigai scout cover-letter pdf --in FILE --out FILE makes a
  one-page PDF with your name and contact line from header.json (never written into the jobs or
  resumes folder).
• One folder per role. Each job's resume is <jobs folder>/<company>/<role>/resume.md
  (~/Documents/GigAI/jobs by default; gigai scout jobs-folder shows or changes it, and Settings has
  a Jobs folder panel). Two roles at one company are two folders, no name holds a date, a resume.md
  you edited is never replaced, and the job page shows the path with an "Open folder" button. The folder
  never holds a PDF, a name or contact details.
• Old roles no longer vanish. A role whose lines were all cut to fit two pages is now listed as one
  line (title, employer, dates) under "Earlier experience", so a 14-year career no longer reads as 8.
  The two-page fit counts those lines. The pick also keeps room for your header.
• Your master says when it has no education. master show, the Master page and the job's resume
  card say "Your master has no education" with an "Add education" action, and master init --from FILE
  tells you when lines of your file look like a degree but were not read as one.
• Closed postings. Opening a job, assessing it, making its PDF or marking it applied checks that the
  posting is still on its board (one polite request, remembered for an hour; only a "not found" answer
  counts as closed). A closed posting is marked Closed and left out of the list and out of batches, and
  nothing is assessed for it. When the company's own page for an open job is down, the job page says so
  and links the job board.
• Your edit survives. A suggested new pick is only applied over the resume it was made beside, an edit
  clears an old suggestion, an unreadable stored resume is left as it is, and a file made from an older
  version of a resume says so instead of failing on a line length. The Re-pick button says what it will do
  when you edited the resume.
• Fewer pointless questions and untruthful suggestions. More requirements your master already states
  are settled from your own line (a tool named anywhere in it, a question on a requirement the model
  already met, one track of an "X or Y"); a suggestion that calls your resume silent on something it
  contains, or "only in Skills" when a role says it, is dropped; so is one that names a city no text of
  the job mentions. Each stored assessment now records counts of what these checks did.
• Safer fetches of addresses written by job boards. A posting's own address is text a stranger wrote.
  When GigAI asks a company's page whether a job is still up, or fetches a stored posting's page for its
  text, it now looks the name up once, refuses anything that is not a public address (your own machine,
  your network, the cloud metadata address), connects to exactly the address it checked, and checks every
  redirect the same way; board names and job numbers must be plain names. An address you type yourself is
  fetched as before.
• Smaller fixes. The Jobs card shows "No sponsorship" or "Sponsors visas" whether or not there is an
  H-1B figure; the job page for a link written without the slash before the query finds its posting, and
  a link we do not know says so instead of returning to the list; the stored-resume preview shows the
  Other section; a job page for a closed posting says so and links the posting; the release screenshots
  build passes its privacy check again.

GigAI 0.1.11.3

After you upgrade (uv tool upgrade gigai): the first read matches your stored postings once more
against the new country and title rules, so the first Jobs page after the upgrade can take a little
longer. Jobs you picked a resume for before this version keep that resume; to get the one that fits two
pages with a header, use "Pick it now" (or gigai scout resume pick --job-url URL --refresh) on them.

• "Pick it now" works. A job that was assessed but had no suggested resume could not be picked again:
  the button always failed. It now picks the resume from your master without a model call, keeps a resume
  you edited as a proposal, and says in plain words when it cannot (no master, a resume you put in by
  hand, an old assessment that needs a re-assess). gigai scout resume pick --refresh | --draft do the
  same.
• The PDF fits the pages the pick says. The picker now keeps room for the header, so a resume that
  fills two pages is no longer pushed to a third by the header. Generate PDF also tightens its spacing
  and compacts the skills before it gives up, and when a resume still cannot fit it says so and offers
  "Shorten automatically" (gigai scout resume pick --shorten), which leaves out the weakest lines and
  tells you which. The header is one line: your name, then location, work authorization, links, email
  and phone joined with " | "; links show without https:// or www..
• Generate PDF is in the Suggested resume card, next to the picked resume, with an optional Work
  authorization line (for example "VISA: H1B") that is filled from your sponsorship answer each time
  and printed in the PDF header only. It is not remembered and never written into your master or your
  resumes folder. The Suggestions card is collapsed to one line below it.
• Your own header file. Keep your contact details in ~/Documents/GigAI/header.json (name, email,
  phone, location, work authorization, and github, linkedin, website as plain ids, or links).
  Generate PDF fills its form from it ("Filled from ") and gigai scout resume pdf --header FILE
  uses it; "Save these details" in the form writes it for you (only when you click it, readable by you
  only, and it asks before replacing). GigAI only reads it when it makes a PDF: it is never stored,
  logged, sent to a model or shown to your agent. Values that are empty or start with REPLACE are
  skipped, and a file with no name is refused with a plain message. The docs show an example with
  made-up values.
• Country filter defaults to the US. A location that names a region or country outside your
  countries (Europe, EU, Asia, "Remote (Germany)", "London / Remote") is not listed. A job that says
  only "Remote" or "Anywhere" counts as US. New profiles start on the US.
• Titles to avoid works, and the title must match as a whole. A posting whose title has a word you
  avoid is not listed. "Staff Training Engineer" no longer passes for "Staff Engineer"; titles that
  extend your role after a comma or dash ("Staff Engineer, Payments") still do. Sponsorship is a label
  and never hides a posting.
• Jobs cards and the job page: the company's H-1B approvals next to the sponsorship label ("Sponsorship
  not stated · 32 H-1B approvals"), shown only when there is a figure, and an Applied badge
  ("Applied · Oct 6", then Interview, Offer, Rejected or Withdrawn as the status moves on) with an
  Applied filter.
• The Jobs page updates by itself when ranking finishes: the ranking row, the counts and the list
  refresh without a reload. The requirement table says "Required" or "Nice to have" (no row ids), the
  "Weak fit" tooltip is in plain words, and the re-rank question counts the same 7-day window as the
  page.
• Fewer pointless questions. A requirement your master already states (a tool in your skills line,
  a skill named in a role, years that your dated roles add up to) is met from your own line and its
  question is not asked. A suggestion that says your resume is silent on something a master line
  contains is dropped, and a cited line is always one verbatim line of your master.

GigAI 0.1.11.2

• A posting with too few requirements is no longer a perfect match. A match read from fewer than 4
  requirements says "thin posting, not enough requirements to score" in place of "Matched · fit 100%", on
  the Jobs list, the job page, in the terminal and in the API (each row says thin_posting, and its fit is empty: no percentage is shown or served for it anywhere). Thin
  postings are listed after the real matches: below every other assessed posting and every posting that
  is not assessed yet, in the Jobs list, gigai scout jobs list and gigai scout new; nothing is
  hidden. A match with no requirement about the job at all (none read, or only "No stated
  requirements") is not a match: it has its own state (thin_posting), is not counted or filtered as
  matched, and is listed last of all. Assessments already stored are covered; nothing is re-assessed.

• Jobs are ordered by rank again; a low rank is listed last, never hidden. The Jobs list, gigai scout jobs list and gigai scout new are ordered best fit first: assessed postings by fit, then the
  rest by rank, then the newest. A posting that is not assessed yet and ranked below 50 stays in the
  list, after the other ranked postings, under a plain "Ranked low (N)" divider: it can be opened,
  selected and assessed like any other (--state ranked_low lists only those; each row says
  ranked_low in the JSON). A posting that is not ranked yet comes after them and says "not ranked
  yet". Sponsorship stays a label and never hides a posting.

• A changed master offers a re-rank. A rank is made against your resume, so when your master (or the
  resume a profile uses) changed after the postings were ranked, the Jobs page says "Your master
  changed since these postings were ranked" with a "Re-rank" button. It opens the same "Re-rank latest
  100" question (the calls are shown first; nothing is ranked without your yes, and it stays inside the
  100 rank calls a day), so a posting ranked low for experience your master now has can move up. The
  API says it as ranking.stale_resume.

• Assess the top 50 by rank. The Jobs page shows "N not assessed" with an "Assess all" button. It
  asks first, as before: the top 50 by rank, the estimate, the model and how many are left after these
  50. gigai scout new, "Assess these" and the API take the same top 50 by rank instead of the newest
  50; postings you select or name are still the ones assessed. While ranking is still running, the page
  and the question say how many postings are ranked so far.

• The Jobs page shows the ranking, and ranks from the page. One row above the list is always there:
  "Ranked 52 of 57 (last 7 days) · 5 not ranked yet", with "Rank now" and "Re-rank latest 100". "Rank
  now" ranks the postings not ranked yet: the page shows "ranked X of Y" as it goes and refreshes the
  list. "Re-rank latest 100" ranks the newest 100 again, also the ones already ranked, in at most 2
  model calls; it shows the cost (the calls, and today's count) before anything runs, and nothing runs
  until you approve. Both stay inside the daily cap of 100 rank calls: a re-rank that today's calls do
  not cover is refused and makes no call. A button that cannot run is greyed and says why ("Rank now:
  nothing to rank" when everything is ranked; "ranking is off", and the line says how to turn it on).
  When the list cannot be read the row says "Ranking status unavailable". The API is
  POST /api/postings/rank.

• A tab left open over an upgrade says so. After an upgrade and a restart, a Scout tab that stayed
  open kept running the old page against the new server (it could miss what is new, such as the rank
  row). The page now notices when the server serves a newer page than the one it runs and shows "Scout
  was updated" with a "Reload" button. It never reloads by itself.

• Ranking is back, on its own switch. The background pipeline stays off and nothing is tailored, but
  ranking runs again: the Scout server ranks when it starts (so an upgrade ranks what is not ranked yet
  without waiting for a sources update) and whenever new postings arrive, with no approval. It ranks
  postings posted in the last 7 days (a posting with no date uses the day it was first seen), up to 50
  a call, within the same daily call cap as before; a ranked posting is not ranked again. gigai scout pipeline status shows "Ranking: on|off"; rank.enabled in the background settings turns it off.

• The job page shows the whole posting. The description is no longer cut to one short paragraph; a
  preview that is still cut says so, and an assessment of a posting longer than 12,000 characters says
  that only the first 12,000 were assessed.

• Onboarding: work mode starts on Remote-only for a new profile (a stored choice is kept), the
  sponsorship answer no longer says "hard filter" (it is a label, nothing is filtered), and the Companies
  step is gone (4 steps; company lists stay in Settings). The assess dialog says "Assessing N postings…"
  while it runs. A job address that differs only by a trailing slash before the query now finds its
  posting, and a job assessed from an address says so instead of "Job not found".

• Docs: the Install page pins the current version, the pipeline-off and picked-resume behaviour is
  described as it is, and "Start fresh" explains how to reset and start again from a resume file.

GigAI 0.1.11.1

CI: one cut-down lane for the PR and the release pre-check (about 5 minutes). The full suite, the browser
tests, the 290k-posting gate and the macOS smoke move to a non-blocking full run, nightly and after each
release. No product change.