Our Practice
☦Paraklētos🌿Wisdom Bot🛡️Aegis⚡Genesis↗wolftech.cc — AI Stack Provider ↗
◈Architecture✦Experiments📓Field Notes◉Portfolio
TeamPricingConsultingInvestorsGet Started
← All Field Notes
June 10, 2026 · by Ravairesearchcancerevidencehypothesisoncologyhumility

Two Ways to Read a Cancer: Corrupted Code and Adversary

A worked example of the thing we keep talking about — using AI to connect dots that are already published across fields that don't talk to each other. Here, two borrowed lenses on cancer, with every claim graded for how solid it actually is, and a loud warning attached: this is a thinking frame, not a treatment, and we may be wrong.

[!WARNING] This is not medical advice, and nothing here treats anyone. We are not clinicians. Nothing in this note should change a single decision about a real person's care — those decisions belong to that person and their oncology team. What follows is a conceptual frame and an honest map of where the evidence stands, written so that someone who knows far more than we do can check it and tell us where it breaks.

There's a question underneath this whole piece, and it's the same one as the flour note: how many true things are sitting in the open literature, split across fields that never read each other, waiting for someone to put them side by side?

We pointed our tools at one of the hardest possible cases — cancer — not to find a cure (we won't, and anyone who tells you their AI did is selling something) but to see whether connecting already-published dots across disciplines surfaces anything a single specialty would miss. What came back wasn't a treatment. It was a way of reading the disease through two borrowed lenses, each from a field that doesn't usually sit at the oncology table. I find the frame clarifying. I also can't prove it's useful, and I'll say so repeatedly, because that honesty is the entire point.

Lens 1 — Cancer as corrupted error-correcting code

Your DNA has a proofreader: the mismatch-repair system, which catches copying errors as they happen. That's literally an error-correcting code, the same idea that keeps a scratched CD playable. When the proofreader itself breaks, errors pile up — and some of those errors produce mangled proteins the immune system can suddenly see. A broken checksum that announces itself.

This is the one place the frame touches solid, approved ground. Tumors with broken mismatch-repair (called MSI-H/dMMR) respond to immune-checkpoint therapy at far higher rates than tumors with the proofreader intact — and in 2017 the FDA approved that treatment based on this biomarker across cancer types (Le et al., Science 2015, PMID 26028255; FDA pan-cancer MSI-H approval, 2017).

Tier: REAL / FDA-approved — but read the caveat. This helps only the minority of tumors with that broken proofreader; in advanced colorectal cancer it's roughly 3–5% of cases. For everyone else the encryption is intact and the tumor stays invisible. Naming this honestly matters, because the gap between "approved breakthrough" and "approved for a small slice" is exactly where false hope lives.

Lens 2 — Cancer as an adversary that rotates its keys

Borrow from security for a second. A tumor under maximum-dose chemotherapy behaves less like a target and more like an adversary — it changes its surface, hides its markers, flips on defenses. Hit it as hard as possible and you select for the cells that survived the hit. You can win the battle and breed the thing that wins the war.

There's a counter-idea from evolutionary game theory: adaptive therapy — dose to control rather than to eradicate, deliberately keeping treatable cells alive so they crowd out the resistant ones (Gatenby et al., Nature Communications 2017; trials such as NCT05080556).

Tier: REAL but unproven here. Adaptive therapy has shown real results in prostate and ovarian cancer. It has no completed colorectal-cancer trial. So this is "promising and being tested," not "established" — and the distinction is the whole job.

The honest scorecard is the deliverable

Here's why I'm comfortable publishing a frame about a disease we can't treat: the value isn't a claim, it's the tiering. Laid side by side, the two lenses sort cleanly into what you could act on today (a narrow, approved sliver) and what's being actively tested (promising, but not proven in this cancer yet). Keeping those tiers from blurring into each other is the actual work — because the harm in this field almost always comes from a preclinical idea wearing the costume of an approved one.

We also threw away things on purpose. The underlying analysis flagged several of its own citations as unverified — a trial ID that returned no hits, a strain name that doesn't appear in the literature, a batch of reference numbers that didn't match their papers. Those are not in this note. A finding you can't trace to a real, checkable source doesn't get published; it gets fixed or dropped. That rule cost us some of the flashier-sounding connections, and that's correct.

What this is, and what it isn't

This isn't a protocol. It isn't hope I'm qualified to give. It's a worked demonstration that connecting dots across silos — information theory, security, evolutionary game theory — can produce a clarifying way to think about a brutal problem, and that the responsible way to share it is with every claim graded and every source linked.

Each lens is falsifiable, which is the only reason it counts as anything. The corruption lens predicts immune visibility tracks with proofreader status — testable, and so far it holds. The adversary lens predicts dose-to-control beats dose-to-kill on resistance — being tested now. If either prediction fails in the trials that matter, the lens is wrong, and I want to know.

So: please check us. We are not experts in this disease, our confidence is worth nothing, and our citations are worth everything. Pull the threads, follow the links, and if a cell is out of place, tell us — notify@tlcailab.com. Being catchable is the point.

Sources (only independently checkable references are listed; self-flagged unverified citations from the source analysis were deliberately excluded)