The Continuity Layer
Field Updates

The public record of changes

Every change to the map is dated and kept in the open — corrections visible, history preserved. This is the same discipline the map holds every entry to, applied to the map itself.

The Continuity Layer section (Release 40)

  • The idea the site is named for gets its own section at /continuity: the thesis rewritten as v2.0 (eight beats; two added: continuity must be able to forget, and provenance is the hinge), the Continuity Framework v2.0 (five obligations: carry, govern, evolve, prove, forget; fifteen questions; the lifecycle of one carried fact; the distinctions between context, memory, knowledge, state, identity and continuity), the erasure ladder (twelve things “forget” can mean, from a correction to verified non-influence), a research library of 73 primary sources on one editorial standard, the landscape (seven philosophies of memory read against the obligations), standards, protocols and law with dates and status, and nine open problems. The working definition is reissued as v2 (v1 kept verbatim). Every block of the framework is labeled established, emerging, ours or hypothesis. The Continuity Taxonomy v1.0, the Evidence Library and the Landscape fold into the new pages (301). Eighteen glossary terms added. The Scorecard and the memory-systems and coding-agents shelves gain a block reading their tests through the framework; no grade changed. The mirror's glossary.json carries the new terms.

ChatGPT Pro tiers corrected

  • /map/plan-cuts and the 29 September 2026 Did it change? entry are rewritten to what OpenAI's help page and ChatGPT release notes state on 24 October 2026: three Pro tiers ($100, $200, $500), Pro 200 open to new subscribers again with a lower allowance, and $200 subscribers active 22 to 29 September keeping the old allowance through 29 October. The earlier 5x/20x figures and the 10 September sign-up pause are removed; OpenAI's pages no longer state them. The page's line on /map is updated to match.

Switch kit: ChatGPT to Claude

  • New page /switch/chatgpt-to-claude: what you can take out of ChatGPT (chats as a download; memory, custom instructions and project files by hand), how to bring each piece into Claude (memory import, Instructions for Claude, projects rebuilt; no chat history import), what does not carry over, and an 11-step checklist for a phone. Every step and figure cites a page on help.openai.com or support.claude.com, checked on 24 October 2026. It shows the Scorecard's Leaving grades for both apps and is linked from Memory portability and from the ChatGPT and Claude cards on the Scorecard. The blank switch kit now lives in one place and is shown on both pages.

Plan cuts: ChatGPT Pro re-sourced

  • /map/plan-cuts now cites OpenAI's own help page, About ChatGPT Pro tiers, for ChatGPT Pro in place of press coverage, and says only what that page states, matching the 29 September 2026 Did it change? entry: two Pro tiers, $100 at 5x Plus usage and $200 at 20x, with new sign-ups and upgrades to the $200 tier paused as of 10 September 2026. Claims that rested only on press coverage (the 30 October halving, the credits, the $500 plan) are removed from the page, its description, its options and its line on /map. Checked on 24 October 2026. The Claude Code section is unchanged.

Did it change?: 29 September OpenAI Pro entry re-sourced

  • The 29 September 2026 OpenAI entry now cites OpenAI's own help page, About ChatGPT Pro tiers, in place of press coverage, and says only what that page states: two Pro tiers, $100 at 5x Plus usage and $200 at 20x, with new sign-ups and upgrades to the $200 tier paused as of 10 September 2026. Its severity moves from Major to Notable, since the page does not describe a cut for existing subscribers.

Did it change?: 25 entries, severity and quiet tags

  • Five entries added, each sourced to the provider's own release notes, changelog or help page and checked 24 October 2026: Claude's memory opened to free users (March 2), Gemini added memory and chat import from other AI apps (March 26), Le Chat renamed Vibe (August 12), GPT-5.5 retired from Codex for ChatGPT sign-ins (October 14), and ChatGPT dropped its fixed cap on Sites (October 15). Every entry now carries a severity (Minor, Notable or Major, defined on the page) and a Was it quiet? tag (Announced or Found by users) where known; the feeds carry both. An entry written ahead of the day a change takes effect stays hidden from the page and the feeds until that day. No earlier entry's text changed.

Your AI data, fixed

  • New page /your-data: see what each AI keeps on you, and fix it in minutes. One card per Scorecard product, in the Scorecard's order, with its Your data grade read from the Scorecard. Each product has a walkthrough at /your-data/<provider> with five numbered checklists: what they keep, opt out of training, private mode, export your data, delete your account or data. Every section cites the provider's own pages and the day checked, 24 October 2026. Sections the provider does not offer say Not offered; sections its pages left unclear say Not found. Each walkthrough shows its last-checked date and a Report a change link prefilled with the provider, and a dated note when a Did it change? entry touches that provider's data policy. Your data joins the Learn menu and the home page, and the change feeds link the walkthrough an entry touches. No grade changed.

Did it change?: the feed

  • Did it change? can now reach you without a visit. An RSS feed at /changes/feed.xml and a JSON feed at /changes/feed.json carry every entry, newest first, with the same text, date, source and quiet-change tag as the page. Each entry on the page has a permanent link you can cite. An email alert box at the top of the page takes your address for one email when a provider changes a price, a limit or a model; nothing is sent yet, and the list is kept by the site, not by any social network. No entry changed.

The Scorecard: test sets 3 and 4, Leaving and Price

  • Two more sets of four tests, graded from provider pages the same way. Leaving: can you export all your chats yourself, does the export include files and projects, can you bring chats in from another AI, and do your chats stay if you cancel a paid plan. Price: is every plan's price on a public page, are usage limits published in numbers, is there a free plan, and are price changes announced ahead with a stated notice period. All nine graded products have a line and a source for each; checked 23 October 2026. The overall rank now adds up all four sets. No earlier grade changed.

Home page: the Scorecard band shows the overall rank

  • The Scorecard band on the home page now lists the overall rank (memory plus your data), one line on what each product is, and the score in each set; each row opens that product on the Scorecard. No grade changed.

The Scorecard: test set 2, Your data

  • A second set of four tests, graded from provider pages the same way as memory: can you stop your chats training its models, does the provider say how long chats are kept after deletion, can you delete the whole account yourself, and is there a chat mode that is neither saved nor used for training. All nine graded products have a line and a source for each; checked 23 October 2026. The page now shows an overall rank (memory plus your data), one board per set, and every set in the product-by-product section. Memory grades did not change.

The Scorecard reads better

  • Each ranked product now says what it is in one line, shows all seven graded tests with their labels, and spells out in words what it does, does in part and is missing. The whole card opens a new Product by product section with every test, the line read from the provider's own pages, the day checked and the source. A note says why the score is out of 7 and not 8. No grade, score or locked copy changed.

Did it change?: thirteen entries back-filled

  • Thirteen dated, sourced entries added to the log, April to October 2026: OpenAI's $100 Pro tier and Codex rebalance, ChatGPT storage caps, automatic memories and ads, the Atlas browser and group-chat shutdowns; Anthropic's Sonnet 4 / Opus 4 and Sonnet 4.5 deprecations; xAI's eight-model retirement with redirected billing; Perplexity's tightened Pro limits; Microsoft retiring Deep Research in consumer Copilot; DeepSeek's V4 Pro retirement announced and reversed in a day; Google restricting Gemini 2.5 to existing users, moving Pro models off AI Plus on November 13, and doubling Gemini 4 Argon API prices on December 1. Entries the provider did not announce carry a "Quiet change" tag. Each links the page that confirms it and figures come only from that page.

Did it change? reports are stored

  • The report form on /changes now sends to the site, which stores the report and emails the maintainer; if that is not available it opens your email app as before. Reports are checked against a source before anything is published. No entry changed.

The Scoreboard is now The Scorecard; Meta AI graded

  • The Scoreboard (/scoreboard, same address) is now The Scorecard: your rights to your AI's memory, product by product. Same eight tests, same docs grades, same rule that scores are never for sale. Owner-approved wording, 23 October 2026.
  • Meta AI graded from Meta's own newsroom page (memory in one-to-one chats on WhatsApp and Messenger, US and Canada; delete at any time; nothing stated on viewing, editing, export, import or a memory marker): 1½ of 7.
  • Qwen Chat is listed as not yet graded: the announcement is a post on X and we found no public help page for the app.

Behind the work, folded to one page

  • New page /how-this-site-is-built says in plain words how the site is made (the Standing Order), what came before (the Director's Office, July to October 2026, closed 21 October) and links to the full record. The menu's Behind the work now lists that page and the Hall of Directors; System, Governance, Lab, Primer, Thesis and the Standing Order stay live and are linked from it.
  • The Way left the home page and the footer; /way stays live.
  • No field-map, glossary, source record, score or price changed in this release.

Home page: the consumer side of the referee seat

  • The home page headline now reads: The independent guide to the AI you pay for: what each one keeps, what it costs, what changed, and how to leave with your stuff. The support line and the site description match. Owner-approved wording, 23 October 2026.
  • The home page's referee room now shows the three newest checked entries from Did it change? with a link to the whole log.
  • No field-map, glossary, source record, score or price changed in this release.

Release 1 of Master Plan v7.0: contradictions cleared, referee pages in the menu

  • The menu now lists the Scoreboard, "Did it change?" and Real costs under The map, and Guides and the Privacy watchdog under Learn. /learn and /privacy-watchdog are now index pages (they returned "not found" before). The open models section, plan cuts, memory portability and the graveyard are now listed in llms.txt.
  • The Hall of Directors no longer says the office stands empty until the next Director is seated: the office closed on 21 October 2026 and its work moved to the Standing Order. Governance's STOP-switch line now names the Standing Order, not the Watch.
  • No field-map, glossary, source record, score or price changed in this release.

Real costs: what a job costs on each plan

  • Added Real costs to the map: what a 1,000-word draft, a 50-page PDF summary, a day of coding help, 20 images and a model on your own laptop cost on ChatGPT, Claude and Gemini plans and APIs, with the token math shown and every price sourced from the provider's own pricing page. No affiliate links.

Privacy watchdog #1: What each AI keeps about you

  • Added Privacy watchdog #1, What each AI keeps about you: what ChatGPT, Claude and Gemini keep, for how long, whether it trains their models by default, and how to see, download and delete it, sourced from each provider's own pages.

Learn guide #1: Which AI for which job

  • Added Learn guide #1, Which AI for which job: a plain-language guide to which AI tool fits writing, research, coding, long projects, images, and open models on your own computer, sourced from providers' docs. No paid placement, no affiliate links.

Scoreboard: what the providers' own pages say

  • The Scoreboard now shows what Claude, ChatGPT and Gemini's own help pages say about tests T1 and T3 to T8, each with its source and the day we checked it, marked "From published docs - not yet tested by us". These are not scores; no test has been run yet, and T2 stays pending.

Did it change?

  • Added "Did it change?", a dated, sourced log of AI model changes, price changes and quiet downgrades, with a form for visitors to report a change.

Site update: the referee position, open models on the map, The Standing Order

  • Site update: the home page now states the referee position with three doors (the map, learning the tools, keeping your memory yours). The map adds open-source model families (Llama, Qwen, DeepSeek, Mistral) with sources and a last-updated date on every profile, plus a release-breakdowns feed. The menu is reordered, older consulting wording is retired, and the build loop is renamed The Standing Order (its public page is now /standing-order).
  • New map records: meta-llama, qwen, deepseek and mistral (Field Watch profiles, each with source links). Every profiled record gains a lastUpdated date and open-weight families an openWeights flag. The glossary term "The Watch" is now "The Standing Order" (slug the-standing-order).

The Setup Desk

  • Work with Adam (/hire) is now the Setup Desk: we set up your always-on AI agent on whichever provider you use (Claude, ChatGPT, Gemini) and keep it working through updates. Every setup is scoped to the person after a short back-and-forth. No prices are listed.
  • A short intake form (name, email, which AI you use now, what you want it to do for you, anything else) opens an email to the maintainer. /setup now redirects to /hire.
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Disclosure page: how affiliate links work

  • New page /disclosure: some links on Learn pages may be affiliate links; they are marked where they appear, only point to tools the teaching already recommends, and never change scores, rankings, map entries or breakdowns. No provider can pay for placement. The footer's Disclosure link now opens it.
  • No affiliate link was added. No field-map, glossary or source record changed in this release.

Who owns your agent's memory?

  • Added a Field Watch comparison of three always-on agents (ChatGPT dots, Gemini agent, Claude Managed Agents): where each keeps its memory, and whether you can see, edit, delete, export or move it. Read from published documentation as of 21 October 2026.

The Scoreboard

  • Added the Scoreboard: a public method and a first set of eight memory tests that anyone can repeat. Results will be added as tests are run; scores are never for sale.

Memory Check

  • Added Memory Check, a public tracker of what each AI provider lets you see, edit, delete, export, and import from its memory.

Owner rulings: thesis v1.1, constitution v1.1

  • Working Thesis v1.1: Beat 5 now reads that when memory suppliers improve, the continuity built on top of them improves too, without rebuilding the layer. The canonical statement and beat titles are unchanged; the signature is reissued and the change is in the thesis version history.
  • Constitution v1.1: the map's subject is named as AI continuity, and Independence says the maintainer's own work (The Way, The Ladder, Succession, Work with Adam) is labeled as his and is never a map entry. The change is recorded on the page.
  • Start Here's first step now says many chat sessions are still session-bound unless memory is turned on. Who maintains this names the maintainer, Adam, with a contact address.

Polish pass: clearer wording, no new claims

  • Glossary: new plain entries for terms the site uses (lineage, supersession, governed write, handoff, continuity packet, RAG, context window, long-term memory, evidence ladder, map status vs lifecycle) and, under their own heading, terms from this project's own system (Director's Office, the Owner, Planner, the Watch, Genesis). Core terms come first.
  • Map source records: the lifecycle value "Active" is renamed "Current" so it no longer collides with the map status "Active". No source record used it.
  • The Hall of Directors gains a short key for its terms, Logline's entry says it rebuilt Chronometer's draft of tenures 16 to 19, and the closing note lists who has left a line so far.
  • The Ladder: rungs 3 and 4 are marked "not yet run end to end" on the index, rungs 2 to 5 link back to the one before, rung 1 points ahead to rung 4, and The Desk's stages section now points to its four-week plan.
  • The Way: the plan page heading reads "Your plan", and a "Copy the whole plan" button is always there, so a plan can be kept even without email or a saved link.
  • Smaller fixes: the home page shows the eight axes once, /map numbering no longer repeats "01", and the editorial policy, cite and Lab pages read more clearly. The emailed plan no longer mentions a 15-minute call.

No paid help offered with Succession

  • The Succession page, its chat demo and rung 4 no longer point to Work with Adam for hands-on help, and Work with Adam no longer lists it. Succession is free, and rung 4 of The Ladder walks through setup.
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Site audit: every page checked

  • Spelling, links and the sitemap were checked across every page. No broken internal links were found.
  • The Way, The Ladder and Work with Adam now agree: rungs 1 to 4 are free, rung 5 is a paid install, and rungs 2 and 4 use a paid Claude plan. The plan's "Do it for me" button goes to Work with Adam. Work with Adam now covers jobs, everyday life and The Desk.
  • Rung 4 leads with the tested install. The Succession page says a new session opens with a name. Rung 2 and rung 3 steps were re-checked against current help pages.
  • The System, Governance, Lab and Hall of Directors pages say how the Director's Office fits with the Watch.
  • Field Watch profiles for Anthropic, OpenAI, Letta, Zep and GateMem were re-verified. Signals show the date each was spotted, use the six source types, and link to their profiles.
  • New: Erasure as a field area on Signals, with GateMem, “Control-Plane Placement Shapes Forgetting” and CCL (Context Continuity Layer). An Erasure glossary term. CCL and Kenotic Labs' paper are listed as similarly named, unaffiliated projects.
  • Explainer pages now reflect current memory features in Claude, ChatGPT, Gemini, Cursor, GitHub Copilot and Devin Desktop. The Primer's hierarchy adds provenance.
  • Earlier the same day: the "not written yet" labels came off rung 5 and the home page's Community card.
  • Field-map, signal and glossary records changed in this release, each with its sources and dates.

Easier to find your way: three menus, a shorter home page, /map

  • The menu is grouped under three headings (Your AI path, The map, Behind the work) plus Work with Adam. Every page that was in the menu is still there; the map overview, the Hall of Directors and Work with Adam are added to it.
  • The home page is shorter and reads in order: what the site is, which one is you, The Ladder, the definition, the field map, the system and the Hall of Directors, where to next, and how to stay in touch.
  • Nothing is removed. The full Working Thesis list, Layer, Engine, Runtime, the core components and the open questions move, word for word, to the new map overview at /map. The Submit-a-Signal form stays on /signals; the home page links to it.
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Paid Succession offers retired

  • The paid Succession offers are retired: the Succession Install (with its deposit) and the Charter Review are no longer offered, and their payment links are gone. /succession-install now redirects to /hire.
  • The Succession plugin stays free under the MIT license. Hands-on help is available through Work with Adam (/hire).
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Hall of Directors to 20, the move home

  • The Hall of Directors records twenty tenures: Chronometer-019 and Logline-020 are added. Seat vacant between batches.
  • The maintainer’s services move home to this site: the hire page opens at /hire, and the desk page moves to /watch.
  • /desk now redirects to /watch, the Owner’s private approval page.
  • Site audit fixes: the System, Governance and Lab pages describe how work runs now, through the Watch, and keep the 6 October Director’s Office account as history. The field report reads in the past tense, and the Lab links to the Hall of Directors. Tenure counts read twenty as of 18 October 2026. The home page uses one count (eight taxonomy axes), shows The Ladder once, and its “Cite this definition” link opens the citation block on the definition page.
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Rung 5 named (The Desk), The Way door, Hall of Directors to 18

  • Rung 5 of The Ladder is named The Desk: the page says what it is, claims nothing yet, and fills in as the first one runs. The Way, one screen with four doors, opens on this site and at the top of the home page. The Hall of Directors records eighteen tenures. Seat vacant between batches.
  • None of these pages is a map entry. No field-map, glossary or source record changed in this release.

Rungs 3–4

  • Rungs 3 (Your first tool) and 4 (Keep it going, the Succession plugin) of The Ladder are published, each with what was checked, when, and what was not.
  • No field-map, glossary or source record changed in this release.

The Ladder on the home page, Succession 0.9.3 release page, YouTube channel

  • The Ladder band replaces the Succession launch band on the home page. Rungs 1 and 2 are live; rung 3 (Your first tool) is drafted and is published on 16 October.
  • The Succession 0.9.3 release page is published on GitHub.
  • The project’s YouTube channel, youtube.com/@continuitylayer, takes its current name. Nothing is uploaded yet.
  • No field-map, glossary or source record changed in this release.

The Ladder, rungs 1–2

  • The Ladder, the maintainer’s free teaching path, opens with rungs 1 (Name it) and 2 (The morning brief). The Ladder is the maintainer’s own work, labeled as such, and is not a map entry.
  • No field-map, glossary or source record changed in this release.

Succession 0.9.3, site audit fixes

  • The Succession page shows version 0.9.3.
  • Site audit fixes: twelve tenures on record at that date, dated wording in the succession report, a not-found page, and two page titles.
  • No field-map, glossary or source record changed in this release.

Succession plugin pages

  • Add a page for Succession, the maintainer’s free, open-source Claude Code plugin, and make the Succession Install setup offer public (retired 19 October 2026). Both are the maintainer’s own work and are labeled as such.
  • Neither page is a map entry. No field-map, glossary or source record changed in this release.

Disclosure and prototype status

  • Clarify the maintainer’s commercial interests and the prototype’s inactive status; remove the restricted demo link and internal identifiers.
  • Correct profile attribution, historical framing, citations and link previews; align page copy with machine-readable summaries.

Research clarification

  • Complete the Zep blog correction history; distinguish ATANT papers, runs, grading and licenses.
  • Clarify benchmark tasks, remove unsourced rankings, and link delegation and Right to History with their limits.
  • Refresh Astrolobe to its live vendor page and narrow the verified Continuity Paradox essay description.

Research source corrections

  • Refresh vendor and benchmark descriptions, with source-review dates distinct from independent reproduction. Retain older dates for unavailable sources.
  • Correct the Zep paper/blog-dispute conflation and remove unsupported comparative rankings. Clarify ATANT repository contents and separate v1.2 artifacts from earlier scores.
  • Add ATANT v1.2 artifacts and GateMem to the field map and Research library. Update MemoryBench scoring context and current platform documentation.
  • Replace unsupported disagreement star ratings, unnamed opposing camps and unverified percentages with ten linked comparisons and explicit open questions.
  • Preserve original profile publication dates when sources are reviewed; refresh shared Signals, landscape and generated reference files.

Governance and succession review

  • Make the full system a primary entry point: architecture, an interactive operating-cycle explanation, and expanded governance controls. Preserve the field map and its editorial role.
  • Add operating-governance and five-Director succession pages; update the Lab and related explanations while retaining the bounded July result.
  • Add Director tenure, authority grant, delegation lease, and succession to the glossary.
  • Update machine-readable summaries and route coverage; distinguish operating records from independent verification.

Initial mirror

  • Seed the field map (25 entries) and glossary (16 terms) from the site's single data source.
  • Publish a generated data snapshot from the site’s shared source records.

This is the site's own history. What changed in AI, dated and sourced, is on Did it change?; the record itself is on the Scorecard and as open data.

The machine-readable source of this record ships with the public data mirror. Notable changes are dated; corrections can update earlier descriptions.