Integrate ERRATA.md Williams Bound into README, DESIGN, PLAN, and TODO; delete ERRATA.md

[Copilot]

• Updated README.md — added Sublinear Growth (Williams Bound) and Three-Zone Memory to Core Design Principles
• Updated DESIGN.md — integrated Williams Bound theorem, node salience, tier quotas, community quotas, resident-first retrieval, dynamic subgraph bounds, new types, updated storage stores, updated Daydreamer responsibilities, policy-derived constants
• Updated PLAN.md — added HotpathPolicy and SalienceEngine modules, restructured implementation phases, updated progress summary and blockers
• Updated TODO.md — added P0-F (HotpathPolicy foundation), P0-G (SalienceEngine), updated all downstream tasks, added P2-F (community detection), P3-D hotpath scaling benchmarks, P3-E hotpath guard
• Deleted ERRATA.md
• Fixed DESIGN.md:167 — policy-constants guard wording updated to reflect planned (not yet implemented) state (P3-E3)
• Fixed TODO.md notes — "P3-E3 will enforce this" instead of implying guard already exists
• Fixed DESIGN.md:129-130 — replaced max(1, ceil(...)) community quota formula with largest-remainder method that guarantees quotas sum exactly to tier_budget; excluded zero-share communities explicitly
• Fixed DESIGN.md:153-161 — added t_eff = max(t, 2) bootstrap floor and max(1, …) secondary guards; documented valid domain, cold-start safe values, and that the clamp is inactive at scale
• Fixed DESIGN.md:289-323 — clarified HotpathEntry dual role: live RAM resident index and periodic IndexedDB persistence checkpoint; Daydreamer owns write path; startup reconstructs RAM index from IndexedDB without full corpus replay; updated Hotpath glossary entry to match
• Fixed TODO.md:52-60 — rewrote HotpathPolicy test bullets to be concretely testable: monotonicity checked over finite range [0,1,2,10,100,1_000,10_000,100_000]; quota assertions check derived counts sum to capacity (not ratios sum to 1.0); added NaN/Infinity/edge-case guards for H(t), tier quotas, community quotas, and salience
• Fixed README.md:59 — COLD zone definition now explicit: metadata in IndexedDB + raw vectors in OPFS; semantically isolated (no strong nearest neighbours at insertion); only discoverable via deliberate random walk; zones control discoverability, not data lifetime

Original prompt

---

This section details on the original issue you should resolve

<issue_title>The ERRATA.md needs to be properly integrated</issue_title>
<issue_description>The ERRATA.md describes updates and changeswe need to make to our design that will make the system overall closer to something that is maximally efficient in information theoretic bounds. It allows us to grow indexes sublinearly while ensuring biology inspired dynamics have a mix of room to grow and constraints on that growth that will seek optimizations as a function of running in the environment. Literally trading time for space and vice versa.

In any event this document needs to be followed, suffused and diffused through the README, DESIGN, PLAN and TODO.md documents.

The README just needs to gloss it, how does it drive the DESIGN.

The DESIGN.md needs to explain it in a way that incorporates, revises etc the design itself. Mostly this is a process of dialing things in for the design, tightening it, trimming the loose ends, answering questions we didn't think to ask earlier and assessing the consequences, what needs to be revised to accomodate that. A lot of the design is probably ok as is verbatim except where the errata calls out something.

The PLAN.md is the actual delivery vehicle. This thing maps directly to the github project, it's how we know what system to work within and the high level of each deliverable. I expect there to be changes from the errata and also from trying to integrate it.

The TODO.md is a detailed, step by step list of the work remaining., This document is ground truth for our issues. We use issues to drive drive development. It will perhaps be the most impacted by the integration of the errata and that's ok. We want to deliver something that has never been built before. So we need careful consideration of what exactly we are building and how exactly we build it.

This is your task. When it is done, do one final review to make sure you got it all, then delete the ERRATA.md file itself because it will serve no further purposes.

Thank you!
</issue_description>

<agent_instructions>You should consider the code base as written and what will need to change and how as you review. Let that guide your thinking since it will help shape the TODO.md</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• Fixes The ERRATA.md needs to be properly integrated #12

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[Copilot]

Pull request overview

This PR integrates the former ERRATA.md “Williams Bound” architecture into the project’s living docs (README/DESIGN/PLAN/TODO) and removes ERRATA.md, making sublinear hotpath growth a first-class design/planning concept for CORTEX.

Changes:

• Added a Williams-bound-based sublinear growth framework (H(t), salience, tier/community quotas, resident-first retrieval, bounded maintenance) to DESIGN.md.
• Rescoped execution sequencing in PLAN.md and TODO.md to put HotpathPolicy/SalienceEngine foundations first, and updated downstream tasks to integrate them.
• Updated README.md design principles to reflect the new HOT/WARM/COLD model and Williams-bound constraints; deleted ERRATA.md.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
TODO.md	Introduces P0-F/P0-G foundations and threads hotpath/salience/community quota requirements through later phases and tests.
README.md	Updates core principles to include Williams-bound sublinear growth + three-zone memory framing.
PLAN.md	Updates module status and phases to prioritize HotpathPolicy/SalienceEngine and hotpath-aware storage/query work.
DESIGN.md	Adds a comprehensive “Williams Bound & Sublinear Growth” section plus hotpath entities/stores and resident-first query path.
ERRATA.md	Removes the now-integrated errata document.

[devlux76]

@copilot Please fix as recommended.

[Copilot]

Fixed in 4de4453. DESIGN.md now reads "is planned (see TODO.md P3-E3) to prevent these constants from being hardcoded elsewhere; until that guard is in place, discipline is enforced by convention." The matching note in TODO.md was updated to "P3-E3 will enforce this automatically; until then, enforce by convention."

[devlux76]

@copilot Good catch! Please fix as described using your best reasoning for the domain involved while respecting the constraints imposed by the design.

[Copilot]

Fixed in df00a1c. Replaced max(1, ceil(...)) with a largest-remainder method that guarantees quotas sum to exactly tier_budget regardless of community count or sizes. The four-step allocation is now explicit in the design: compute fractional shares, floor to base allocations, distribute remaining slots to largest-remainder communities, and exclude zero-share communities from that tier. This makes the quota system both implementable and directly unit-testable.

[devlux76]

@copilot This was an excellent catch! For small corpora we do need to implement something that won't introduce more problems at scale. Clamping ought to do that but let's make sure whatever this solution looks like, that it doesn't interfere with the dynamics at scale. Perhaps a corpora sized based constraint that falls away above the threshold. Use your best judgment here.

[Copilot]

Fixed in 06702d5. Added an effective-mass floor t_eff = max(t, 2) applied before all three formulas, plus max(1, …) secondary guards on maxHops and perHopBranching. At cold-start (t < 2) the formulas evaluate to conservative safe values (1/1/1 — a single-node subgraph, the only sensible result for fewer than two nodes). Once the corpus grows past the floor the clamp becomes completely inactive (t_eff = t for all t ≥ 2), so large-corpus dynamics are unaffected. The valid domain and this fall-away behaviour are now documented inline.

[devlux76]

@copilot I'd periodically cache it somewhere so that we handle the situation where the user rebooted the machine. IndexedDB feels like a good place for it. It also means that the daydream module can touch it directly and restart/ reload the process which may be simpler since daydream is the default state.

[Copilot]

Fixed in 2ce4dc6. HotpathEntry now documents its dual role explicitly: (1) the live RAM resident set that every query scans first, and (2) the hotpath_index IndexedDB store which holds a periodic snapshot used to reconstruct the RAM index on startup/reboot — no full corpus replay required. Daydreamer owns the write path to hotpath_index and checkpoints it at the end of each maintenance cycle (LTP/LTD pass), so the persisted snapshot is never more than one cycle stale. The object store description and the Hotpath glossary entry are updated to match.

[devlux76]

@copilot Good catch, we should make sure to follow TDD throughout. Nothing scales to infinity but we should always expressly guard against NaN, MAX and MIN and be reasonably thorough in our testing while being reasonable in general about the expectations.

Use your best judgement here and fix it please.

[Copilot]

Fixed in 2905eb9. The test bullets in P0-F2 are now concretely testable and TDD-ready:

• Monotonicity is checked over the finite range [0, 1, 2, 10, 100, 1_000, 10_000, 100_000] instead of "for all t"
• NaN/Infinity/bounds guard added for H(t): edge inputs t=0, t=1, t=Number.MAX_SAFE_INTEGER must return a finite integer ≥ 1
• Quota assertion now checks that derived tier-quota counts sum to capacity, not that ratios sum to 1.0
• Community quota edge cases added: budget=0, empty sizes array, budget < sizes.length; result must never contain NaN, Infinity, or negative values
• Salience finitude guard added: extreme weight/hit-count inputs must never produce NaN or Infinity

[devlux76]

@copilot to me a cold item is one that exists in the graph and perhaps has prev/next nodes connecting it to other nodes to form a book but is otherwise an island unto itself. It can only be discovered during a random walk because at insertion it had no nearest neighbors in vector space.

This might need to be made explicit, cold means far away from the hot path of search and unlikely to be discovered except by a random walk.

[Copilot]

Fixed in 5d1abcb. The COLD definition in README.md now reads:

COLD (metadata in IndexedDB + raw vectors in OPFS, but semantically isolated from the search path — no strong nearest neighbours in vector space at insertion time; only discoverable by a deliberate random walk)

Also updated "zones control lookup cost, not data lifetime" → "zones control lookup cost and discoverability, not data lifetime" to make the random-walk-only nature explicit at the principle level.

[devlux76]

@copilot agreed, please use your best judgement and fix it.

[Copilot]

This was already fixed in 4de4453. The note in TODO.md now reads: "P3-E3 will add a guard to enforce this automatically; until then, enforce by convention." — future tense throughout, no implication the guard currently exists.