For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
A high-bar external application form arrives looking like a deliverable: a thing the applicant owes a gatekeeper, a hoop measured in hours of preparation, a transaction whose currency is access. Under that frame the form is overhead, and the smart applicant minimizes the cost of clearing it.
The frame is wrong by exactly its sign. The form is the instrument. The clarity is the asset. The submission is incidental.
The structural insight is small and load-bearing: the form-shape is not a list of questions. The question selection, the word limits, the order, the implied taxonomy — the whole structure — is the evaluator's taste already compressed into the artifact. Generations of evaluators shaped the form to filter the applicants they wanted to filter. The filtering work is encoded in the form itself, regardless of who reads the answers later.
This makes the form a dipole-calibration instrument with a synthetic evaluator. The architecture that lets a self-modifying agent acquire capability through sparse correction against a high-floor evaluator works here, with the form-shape standing in for the human evaluator. The applicant calibrates against the form's structure; the form's structure is what the evaluator would have said, transposed into question design. The clarification depends on the form-shape, not on the eventual reader.
This is why the YC application "helps you think" even before submission, and why an applicant who never submits but completes the form rigorously can still come away with strategic clarity they did not have before. The questions, the taxonomies, the limits, are the artifacts of a high-floor evaluator's compressed taste. Translating the applicant's thinking into the form's register is calibration against that taste.
Everything else in the piece is a corollary.
A working knowledge practice — a personal graph, a research notebook, a folder of drafts — optimizes for compounding. The practice converges; the convergence is the point; the compounding is the moat. But convergence within a private dialect makes the dialect itself invisible to internal review. The graph cannot run a check from outside itself. The form, being the evaluator's taste-as-structure, is outside; the form forces translation into a register the practice did not build and cannot edit.
Three properties of the encoded evaluator transpose into structural pressures on the applicant:
The asymmetry is structural. The applicant pays in time and clarity; the evaluator pays nothing for being hard to satisfy. The applicant's pain is exactly the evaluator's leverage.
The mechanism collapses if the translation is outsourced. A model that completes the form on the applicant's behalf — pulling from the applicant's repository in the applicant's own dialect — produces a fluent rendering of the practice rather than an external pressure on it. The dialect-incompatibility evaporates. The form becomes another node in the practice. The instrument is real; the operator performing the translation is what makes it work.
The same boundary explains why a softened imagined evaluator collapses the regimen: if the applicant's standard for "what the evaluator would expect" is set by the applicant's own taste, the form-shape's pressure relaxes back to the practice's. The regimen is generative only when the imagined evaluator stays high-floor.
The procedural consequence is non-obvious. The optimal way to use a high-bar form is two stages.
Stage one — substrate run. Treat the form as an experiment object. Produce drafts of every answer with full internal-practice machinery: the dialect at full strength, the graph behind every claim, the founder profile reconciled, the company frame mapped, the privacy posture identified, the gaps surfaced. The artifacts are not the application. They are the substrate the application will be written against.
Stage two — operator translation. The operator takes the substrate out of the experiment and writes the live application directly. Not by pasting from the substrate. By writing fresh, with the substrate as prior. The translation is where the clarification lands, because translation forces the operator to render the substrate in the form's register — the register the substrate cannot be written in without losing its compounding properties.
This pattern has a fourth closure mode the conventional template (submit / decline / let-deadline-pass) does not name: operator-takes-artifacts. The experiment delivers a substrate; the operator delivers the application; the experiment never produces the deliverable. The substrate has done its work the moment the operator can write live answers from a clarified position.
The closure mode generalizes. A tenure dossier, a board update, a court filing, an IRB protocol, an investor memo — any sufficiently load-bearing form admits the same two-stage structure. The substrate is not the deliverable. The substrate is what makes the deliverable writable.
If the form is the instrument, the operator does not need to wait for an external deadline to use it. A folder of un-submitted application drafts — the YC application written every six months regardless of intent to apply, the grant application written for a grant the applicant will not pursue, the keynote talk drafted for a conference the applicant will not attend — produces clarification at intervals shorter than the natural cadence of external opportunities. The discipline is to keep the imagined evaluator high-floor; if the standard softens, the form-shape's pressure relaxes back to the practice's. The regimen is generative only in proportion to the operator's willingness to ship the live versions when real deadlines do arrive — without consequence-deployment, the regimen converges to clarification without action, which is the same failure mode the strategic-thesis null hypothesis names: the practice runs, the work is real, nothing in the world updates on it.
The default frame on application forms is that they are bureaucratic overhead, designed to filter applicants for the gatekeeper's convenience. The frame is half right: the forms do filter. But they filter the applicant's own thinking before they filter the applicant. The hard form is a dipole — structured external pressure against which internal compression resolves into something the operator could not have produced from inside the same dialect. The instrument generates clarity; the submission generates consequence; both are needed.
P.S. — Graph: