John Collins Alpha / Singapore
JOHN COLLINS ALPHA / CUMULATIVE INTELLIGENCE
Most AI work resets at the next prompt. John Collins Alpha carries the work forward: every serious interaction, action, source, market response, failure and correction can change one living model of AI, markets and commerce. Deep Vertical AI makes the defensible part of that state public.
01 / THE CUMULATIVE STATE
The account's accumulated work selects the questions. External evidence challenges it. Markets and outcomes grade it. The resulting model must explain all four realities without pretending they are the same kind of evidence.
Repeated work across AI infrastructure, markets, filings, catalysts, specialist research, portfolio evaluation and system design has converged on one need: intelligence must preserve why a belief exists and what later happened to it.
Current lab evidence increasingly evaluates multi-turn behavior, tool use, accumulated context, prompt-injection resistance and long-horizon control rather than isolated response quality.
The account's research separates present AI-linked economics, incremental growth exposure, system criticality, competitive control, legacy dilution and execution. The latest session reinforced dispersion rather than a blanket AI bid.
Adoption is deepening where models are connected to context, tools, permissions and repeatable workflows, while low-quality synthetic volume is meeting visible reader and marketplace resistance.
02 / HOW EXPERIENCE COMPOUNDS
This is the architectural correction. Account experience is not an extra feed attached at the end; it determines what each stage notices, tests, retains and publishes.
prior cumulative account state
new interactions, actions and inspected evidence
verified AI and specialist evidence
market and outcome adjudication
surviving, changed and falsified beliefs
Reconcile questions, actions, artifacts, failures, corrections and project states with the model already held. The weekly interval is a delta, never a reset.
Use the accumulated state to select relevant lab work, research, news and specialist interpretation, then verify public claims independently.
Overlay market behavior, catalysts and later outcomes on active beliefs. Preserve misses and conflicts instead of narrating around them.
Explain what the combined experience now says about AI, markets, commerce and the direction of the work.
Publish the useful, independently supportable part; keep private memory private; freeze the record and feed later results back into memory.
03 / SINCE 17 AUG 2026 / 13:34 SGT
The interval is useful only where it changes or tests the accumulated model. No new specialist packet is a legitimate result; a one-session price move is an observation, not a story invented after the fact.
A forensic account-led audit showed that Project 09 had converted a cumulative eleven-project system into one static weekly article. The rebuild now defines the missing memory interface and declares incomplete capture instead of implying it.
Anthropic's new transparency evidence evaluates behavior assembled across turns and reports long-horizon limits. Earlier OpenAI evidence similarly emphasizes connected context, tools, trajectory monitoring and rollback.
No tracked SemiAnalysis or paid-Substack item arrived after the prior cutoff. Existing rights-safe themes still concentrate on capital, power, memory, optics and rack-scale systems; they shaped the market test but do not count as interval evidence.
On 17 August, semiconductors rose while software fell; memory and optical-hardware proxies led the selected cross-section. One session updates the prior toward selective physical constraints but cannot prove why prices moved.
SOXX outperformed QQQ by 1.74 percentage points while IGV lagged it by 1.85 points. COHR and MU were the clear leaders. This is consistent with selective attention to physical bottlenecks, but it is not causal attribution or an investment recommendation.
Regular-session close on 17 August 2026 versus 14 August 2026, using Interactive Brokers public historical OHLCV (`Last`). No account data was accessed.
04 / THE ELEVEN-PROJECT NERVOUS SYSTEM
Projects 00–08, this surface and Project 10 perform different functions in one loop. The surface is responsible for preserving their relationship—not flattening them into a weekly list of evidence nodes.
Freeze context and reconcile the research sequence
Discover and rank live market change
Reconstruct the completed session
Maintain the event and dependency clock
Ground claims in primary disclosures
Detect public fact and narrative change
Extract rights-safe specialist themes
Interpret the physical AI stack
Attack downside and rival explanations
Render the public-safe cumulative state
Observe what later confirmed or broke
05 / THE PRESENT PROOF
State of the art is comparative. This rebuild can show an architectural advance over the old surface; it cannot yet claim complete account memory, autonomous ingestion or proven human usefulness.
The existing surface could freeze an edition perfectly but could not show which account memories or project states produced it.
One hand-authored edition imported directly into the homepage; Recents and browsing compressed into a short theme digest; no memory schema, coverage receipt or eleven-project interface.
A forensic audit mapped the full system, defined a typed public-memory state, made the five operations cumulative and rebuilt the homepage and LinkedIn candidates around the state transition.
The candidate can render prior state, account-derived change, external evidence, specialist continuity, market adjudication, project roles, sources and uncertainty from one public-safe object.
Continuous account-wide capture, per-project cursors and a tested private event graph do not yet exist. Coverage remains partial; human usefulness remains unknown and later publication outcomes belong in append-only records.
06 / COVERAGE RECEIPT
Partial coverage can guide a candidate synthesis. It cannot support a claim that the entire account has been mechanically captured or that the resulting system is state of the art.
07 / A USEFUL TEST
Evaluate the whole operating system around a model—memory, tools, provenance, evaluation and outcome feedback—not the model or output volume alone.
08 / PUBLIC EVIDENCE
Anthropic / 17 August 2026
Cumulative multi-turn safety behavior, prompt-injection evaluation and explicit long-horizon capability limits.
Lab-reported evaluation; not an independent capability benchmark.OpenAI / 12 August 2026
Cumulative context for connected tools, workflows, permissions, governance and measured enterprise use.
Provider-reported adoption analysis; not causal proof of business value.OpenAI / 20 July 2026
Cumulative context for trajectory-level evaluation, incident-derived tests, monitoring and rollback.
Provider account of its own systems and mitigations.CGTrader / July 2026
Marketplace evidence that AI-generated supply volume and realized marketplace revenue can diverge sharply.
One marketplace and asset category; not a general measure of AI output quality.European Central Bank / July 2026
Cumulative macro context on AI optimism, concentrated exposure, valuations and circular relationships.
Macro risk context, not a security-level causal explanation.Interactive Brokers / 17 August 2026 close
Regular-session closing prices used for the one-session cross-sectional AI market observation.
A public price snapshot, not account data, an investment recommendation or evidence of causation.Deep Vertical AI / 17 August 2026
Prior public state and a concrete example of moving from nominal capability to verified economic gates.
A published research artifact, not proof of the cumulative memory architecture.
John Collins, PhD / Singapore09 / WHO IS BUILDING IT
John Collins builds and operates John Collins Alpha. Deep Vertical AI is the Singapore company through which the accumulated capability becomes research, architecture and client-owned systems.
Earlier work—investment-bank quant research, HSBC's Artificial Intelligence and Data Analytics laboratory, FTI Consulting's Asia AI practice and doctoral research— explains some of the judgment behind the system. It is context, not the proof.