Dante Blog ▮

About

I’m Dante Lautaro Perea (b. March 23, 1994). I build AI systems professionally. This blog is written primarily for me: a long-horizon archive of models, decisions, and reasoning.

My objective function is functional lifespan extension with agency. The plan is staged:

  • Phase 1: value creation that compounds.
  • Phase 2: capital allocation + influence to (a) extend my personal time horizon using current best practice, and (b) increase global throughput toward new longevity technologies by shifting incentives, adoption constraints, and gatekeeper behavior.

Epistemic protocol (how entries are produced)

My epistemic protocol is mechanism-first:

  • Operational definitions (what, precisely, is being asserted or measured).
  • Causal structure (what implies what, through which mechanism, under which assumptions).
  • Update conditions (what observation would force revision, not rationalization).
  • Uncertainty as a first-class object (weights, not allegiance).

Corollaries:

  • I prefer models that generate decisions, not prose that generates mood.
  • I separate map from territory: what I observed vs. what I inferred vs. what I want.
  • I treat “insight” as a liability unless it can be translated into procedure (a test, a rule, a constraint, a design).

Purpose (why the vault exists)

The purpose of this vault is continuity and compounding of judgment. I’m not trying to produce a narrative that feels coherent; I’m trying to preserve structures that remain useful: the assumptions I was operating under, the tradeoffs I accepted, the signals I trusted, the ones I ignored, and what happened afterward.

An entry is successful if future-me can reconstruct the decision logic from the text alone and resume execution without re-deriving fundamentals—or re-inventing the same self-deception in a new costume.

What I optimize for (clarifying terms)

  • Functional lifespan: not “being alive,” but retaining usable cognition, mobility, and autonomy sufficient to execute long-horizon plans.
  • Agency: control over allocation of time, attention, and capital; reduced dependence on external calendars, institutions, and approval pathways.
  • Compounding: mechanisms where outputs become inputs (skill, distribution, ownership, credibility, data, tooling), increasing rate-of-progress over time.

What belongs here (scope)

This archive stores artifacts that are expensive to regenerate:

  • Decision memos: what I did, why, what I expected to happen, what would have changed the decision.
  • Postmortems: what failed, what was mis-modeled, what signals were available, what I ignored.
  • Mechanistic notes: biology, AI, systems—anything where the causal graph matters.
  • Strategy primitives: sequencing, leverage, bottlenecks, incentive alignment, gatekeeper maps.
  • Tooling: templates, checklists, heuristics with explicit domains of validity.
  • Contradictions: unresolved tensions are kept visible; they are not “smoothed” for aesthetics.

Output formats (so future-me can parse quickly)

Most entries conform to one of these shapes:

  1. Thesis + ledger
  • Claim
  • Why I think it
  • Evidence / counterevidence
  • Predictions
  • Failure modes
  • Next measurement
  1. Decision record
  • Context
  • Options considered
  • Selection criteria
  • Decision
  • Kill criteria
  • Post-decision observation plan
  1. Model
  • Definitions
  • Variables
  • Causal edges
  • Assumptions
  • Where it breaks

Anti-goals (what this is not)

  • Not a personal brand narrative.
  • Not a public-facing curriculum or portfolio (unless it incidentally becomes one).
  • Not a place for “belief signaling.”
  • Not a comfort object.
  • Not a diary of emotions without decision relevance.

If I can’t translate a paragraph into action, measurement, or an update rule, it probably doesn’t belong.

My default failure modes (so I can actively counter them)

I assume I am vulnerable to:

  • Aesthetic rigor: mistaking technical elegance for market relevance.
  • Overfitting to intelligence: preferring problems that reward cognition over problems that reward distribution and sequencing.
  • Local optima: polishing subcomponents while neglecting the bottleneck.
  • Narrative closure: stopping search once a story feels internally consistent.
  • Hidden dilution: “winning” outcomes that don’t produce ownership or durable leverage.

This vault exists partly as a detection system for these patterns.

Standard of evidence (how strict I am)

I use different thresholds depending on consequence:

  • Low consequence / reversible: cheap tests, fast iteration, loose priors.
  • High consequence / hard-to-reverse: pre-mortems, explicit tail-risk accounting, redundancy, slower updates.

I do not require certainty to act. I require bounded downside and a plan to learn.

Maintenance rules (so the archive doesn’t rot)

  • I keep timestamps and preserve old beliefs with revisions rather than overwriting.
  • Each major idea has a last reviewed date and a next check trigger.
  • Any repeated mistake gets a named pattern and a counter-procedure.

Compact statement

This blog is a high-integrity repository of my models and decisions—optimized for functional lifespan extension with agency, executed via staged capital formation and subsequent allocation + influence. It is written in mechanism-first language so future-me can retrieve decision logic, update cleanly, and continue execution without recomputation.

If others read this, it’s incidental. The primary reader is me, later.