A tumour is not just growing. It is maintaining itself — spending energy to stay alive, to build its blood supply, to hide from the immune system, to resist treatment. Every living system does this. A flame, a bacterium, a city — anything that persists in a changing environment does so by actively maintaining its own structure.
Generative Geometry proposes that all such systems maintain themselves through the same structural process: a cycle of 16 positions, in a fixed order, at every scale. A tumour runs this cycle. So does the immune system fighting it. So does the clinic treating it.
The Drug Response Simulator tests this claim against published clinical trials. One equation derives predicted response rates from the structure of the tumour's maintenance cycle. 138 structural groups across 16 cancer types. Mean absolute error: 3.8 percentage points across 72 validated trials. No per-trial fitting. No machine learning. One structural formula. And at the patient level, two immune measurements predict individual immunotherapy response with AUC 0.81.
Every dissipative system — every system that maintains itself by spending energy — traces the same 16-position path from first disturbance to full conservation. In cancer, these positions map to the complete trajectory from the first mutagenic event to established resistance.
The 16 positions fall into four regimes of four. Each regime has a distinct character — a different kind of work the system is doing.
A clinical stage in the DRS is defined by three numbers that capture where the tumour is in its lifecycle:
| Parameter | Symbol | What it measures |
|---|---|---|
| Depth | n | How many nested layers of maintenance the tumour runs |
| Entrenchment | τ | How long the tumour has been maintaining itself |
| Prior lines | p | How many previous treatments the tumour has survived |
The same drugs at different stages produce completely different results — not because the drugs changed, but because the tumour did.
This is why early intervention is structurally different from late intervention. At depth 1 and τ = 0, even modest agents produce high response. At depth 3 and τ = 12, even the strongest combination faces severe structural penalties. The reduction is exponential, not linear.
The 16 positions describe where the tumour is in its lifecycle. But at every position, the tumour is also running a maintenance cycle — a cycle within the cycle. And that inner cycle has the same structure: four operations, in the same order.
This is fractal depth. The same four-fold pattern repeats at every scale. The outer cycle has 16 positions. Each position contains an inner cycle of 4 operations. This inner cycle is what the tumour is doing right now to stay alive — regardless of where it is in the larger lifecycle.
The DRS calls these four inner operations the sub-phases:
The tumour needs all four running simultaneously. Block one, and it compensates through the other three. Block two in different sub-phases, and the compensation narrows. Block three, and the maintenance cycle is structurally compromised. This is why combination therapy works — and why the pattern of the combination matters more than the strength of any single drug.
Every cancer drug, regardless of its specific molecular target, does two things: it disrupts one of the four sub-phases, and it does so in a particular way. The DRS classifies agents into 138 structural groups across four levels of address resolution.
Every drug has a four-level address that tells you everything about what it does:
| Level | What it reads | Example |
|---|---|---|
| L1 — Position | Which sub-phase does it target? | Encounter (SP3) |
| L2 — Function | Which mechanism class? | Checkpoint inhibitor |
| L3 — Target | Which specific molecular target? | Anti-PD-1 |
| L4 — Role | Which pharmacodynamic role? | Binding / Potency / Selectivity / Duration |
Drugs with the same address get the same M value — because the address IS the identity. Nivolumab and pembrolizumab in melanoma both have M = 0.631 because they share the same address at L1–L3. Their L4 addresses differ (pembrolizumab binds 100× stronger), but clinical dosing saturates the target, so L4 differences wash out. The address predicts this.
This is why the v2 model uses 138 structural groups instead of individually calibrated agents. Deeper address resolution replaces calibration. At L2, 34% of M variance is explained. At L3, 61%. At L4, 83%. Each level of address you read eliminates that level of trial dependency.
The four functions are derived from two operations (hold or cross) at two levels (same level or one level deeper):
A fifth function, Observer, belongs to the intervener’s surveillance loop — the physician, not the drug. Imaging, biomarkers, and monitoring are Observer functions. They do not appear as agents in the formula but they determine when to act and what to change.
Each sub-phase admits two actions — one that resists the tumour’s maintenance (hold) and one that enables a structural change (cross):
| Sub-phase | Resist (hold) | Enable (cross) |
|---|---|---|
| Signal | Prevent — block before the step begins | Provoke — surface a signal the system ignores |
| Structure | Transform — build alternative, make the current path obsolete | Accelerate — remove obstacles, speed construction |
| Encounter | Regain control — restore coupling between cycle and governor | Catalyse — force the encounter the system is avoiding |
| Conservation | Slow — extend time, let the containing system respond | Consolidate — lock in the change before the system reverts |
The combination of agents produces one of four strategies, determined by the structural depth of the intervention:
| Strategy | Depth | What it does |
|---|---|---|
| Dissolution | 0 | Prevent the cycle from starting |
| Disruption | 1 | Break one layer of maintenance |
| Rejection | 2 | Break two layers — the encounter fails |
| Occupation | 3+ | Replace the system’s conservation with your own |
Each agent has a potency value M — a number between 0 and 1 representing the structural overlap between the drug's address and the tumour's address. In v2, M is determined by the agent's structural group rather than individual calibration. Drugs in the same group share the same M.
M changes per cancer type. The same drug has different overlap depending on the tumour biology. Nivolumab is M=0.631 in melanoma (highly immunogenic) but M=0.576 in NSCLC. This reflects the different system addresses, not different drug properties.
The Drug Reference shows every agent’s M value, confidence score, calibration source, and structural classification across all 16 cancer types.
Read it as: how much of the drug's potency actually gets through.
M — the structural overlap between the drug's address and the tumour's address. A higher M means a better match.
D — the conservation drain. Everything the tumour does to resist, suppress, repair, and maintain itself against the drug. High D means most of the drug's potency is absorbed by the tumour's conservation layer.
(1−D)/(1+D) — the drain relationship. This is not linear subtraction. When D = 0.5, you lose 67% of potency. When D = 0.8, you lose 89%. A small reduction in drain can be more powerful than a large increase in potency.
For drug combinations, the formula extends to include sub-phase coverage (k), structural depth (n), temporal resistance (τ), and prior-line fatigue (p). The full formula in the DRS computes these from the system state and agent profiles automatically.
k is the number of different sub-phases covered. Two Signal agents (both targeting the growth signal): k = 1. One Signal agent + one Encounter agent: k = 2. One agent in each sub-phase: k = 4.
σ is the cascade bonus:
| Coverage | σ value | Effect |
|---|---|---|
| 1/4 — one type | 1.000 | Baseline |
| 2/4 — two types | 1.050 | +5% exponent boost |
| 3/4 — three types | 1.070 | +7% exponent boost |
| 4/4 — all four types | 1.080 | +8% exponent boost |
These look small. They are not. Because σ is in the exponent, even 5% compounds dramatically. A residual of 0.4 raised to 2.0 gives 0.16. Raised to 2.1 gives 0.14. That difference is real percentage points in response rate.
Across 16 cancer types, the formula identifies cases where novel combinations outperform standard of care. In 91% of cases, the gain comes from covering a previously uncovered sub-phase — not from using a stronger drug. The most commonly missed: Conservation. Survival pathway agents are absent from first-line standard of care in 9 of 16 cancers.
If the combination matters, does the order matter? Two strategies:
Potency-first: start with the highest-M agent. Maximise immediate response.
Coverage-first: start with the agent that opens the most new coverage. Maximise structural leverage.
Across all 16 cancer types at first-line, the two strategies produce the same ordering. The strongest drugs naturally sit in different sub-phases — potency and coverage don't conflict.
This convergence breaks at second line and beyond, where potency-first can miss the optimal combination by 5 to 42 percentage points.
The four drug functions — Sentinel, Miner, Architect, Catalyst — operate at the molecular level. But there is a fifth function that operates at the clinical decision level: the Observer.
The Observer is the physician. Not a drug. The Observer monitors the tumour’s response, reads where the maintenance load has shifted, and decides what to do next. The Observer adds no chemistry — only timing intelligence.
When you select two or more agents in the DRS, the prediction panel shows the Observer panel: the optimal strike order with intermediate ORRs at each step. This is the sequence a physician would follow if monitoring between each strike.
The formula computes the same final ORR whether agents are given simultaneously or sequentially — the equilibrium blockade is the same either way. But the path matters. The Observer panel shows:
Optimal strike order: which agent to give first, second, third. The ordering that produces the highest intermediate response at each step.
Intermediate ORRs: the predicted response after each strike lands. This is what the physician would measure before deciding to continue.
Coverage at each step: how many sub-phases are blocked after each addition.
The value of this information is clinical: if the first strike produces a strong response, the Observer can pause and monitor before adding toxicity. If the first strike produces a weak response, the Observer knows to escalate immediately to the next sub-phase.
Block one sub-phase and the tumour shifts its maintenance load to the remaining three. The maintenance must continue — it just reroutes. The Observer reads this shift and directs the next strike at the sub-phase that now carries the highest load. Each strike exploits the gap the previous strike created.
The formula predicts where the gap will appear before the tumour adapts — because the maintenance cycle’s structure is fixed. The Observer’s job is to act on that prediction.
When agents span different sub-phases, the best ordering produces measurably higher intermediate ORRs than the worst — even though the final endpoint is the same. The Observer panel quantifies this advantage.
Strike 1: Encorafenib (Architect, Structure · Transform, M=0.679). ORR: 52.0%. Coverage 1/4. Load shifts to Signal, Encounter, Conservation.
Strike 2: Nivolumab (Sentinel, Encounter · Regain control, M=0.641). ORR: 82.1%. Coverage 2/4. The σ cascade kicks in.
Strike 3: Lenvatinib (Sentinel, Conservation · Slow, M=0.580). ORR: 91.5%. Coverage 3/4. Three sub-phases blocked.
The Observer’s value: at step 1, the physician already sees 52% response. Enough to confirm the approach is working before adding the next agent. Without sequencing, all three go in simultaneously — same endpoint, but no intermediate information.
The Observer panel in the main DRS shows the optimal strike order. For a detailed step-by-step view — watching the maintenance load redistribute after each strike, seeing the per-sub-phase blockade, and following the recommended next strike — use the dedicated simulator:
The DRS predicts population response rates. But the same formula predicts individual patient outcomes when you resolve the system address deeper — from cancer type down to the patient's immune microenvironment.
At the patient level, the formula becomes:
D — Conservation drain. The immune mechanism holding back the treatment. Measured from the patient's pre-treatment tumour biopsy. In NSCLC: CCR8+ Treg fraction. In melanoma: HAVCR2+LAG3 exhaustion markers.
S — Signal quality. Whether the immune system can perceive the tumour. In NSCLC: FGFBP2+ NK cell fraction. In BCC: NK cell fraction.
Validated across three independent scRNA-seq datasets:
| Cancer | Patients | D reads from | AUC |
|---|---|---|---|
| NSCLC | 159 | CCR8+ Tregs | 0.808 |
| Melanoma | 19 | HAVCR2+LAG3 exhaustion | 0.889 |
| BCC | 11 | CD8 exhausted T-cells | 0.767 |
For context: PD-L1 achieves AUC 0.64. Senior oncologists achieve 0.72. The SCORPIO ML system (trained on 9,745 patients) achieves 0.76. This formula uses two pre-treatment measurements and no machine learning.
Full write-up: A Different Way to Look at Cancer
Select a cancer type. Choose a stage. Pick agents one at a time and watch the predicted ORR change. Notice three things:
Coverage vs potency: does adding a new sub-phase produce a bigger jump than adding a stronger agent in the same one?
Observer value: when you select two or more agents, scroll down in the prediction panel to see the static vs adaptive comparison. How much does sequencing add?
Gaps: which sub-phase is missing from the current combination? The coverage bar and the “How to improve” advice tell you where the structural opportunity is.
Research papers:
The Geometry of Intervention — full derivation of the blockade formula
First Principles of Change — the axiomatic foundation
There Is Only One Way to Grow — 24 dissipative systems, 384/384 positional matches
The book: Riding Change: How Change Moves, and How to Move With It (2026)
Contact: raimo@generativegeometry.science