Structural Signals #001 — Week of 2026-05-12

Daily structural signals from 1000+ public companies — same physics that describes earthquakes, neural avalanches, and power-grid cascades.

Issue covers 2026-05-04 → 2026-05-10 (ISO Week 2026-W19).

Welcome to the first issue of Structural Signals — the weekly digest from the structural-isomorphism project. Every Monday we publish: (1) companies whose critical-state classification flipped over the last week, (2) one methodology deep-dive, (3) a few preprints we're reading, and (4) what's happening in the repo. No clickbait, no "top 10 stocks to buy", no email marketing tricks. If we have nothing structural to say in a given week, we say that.

This first issue runs a bit longer than usual — we're using it to set the vocabulary. Future issues will be tighter (~1200 words).

---

🔄 This week's phase flips

Companies whose critical-state classification changed since last week. We flag both directions: entering near_critical (rising variance, amplifying feedback) and returning to subcritical. Confidence is the model's self-reported probability that the assignment is correct given current public filings, transcripts, and price-action features.

• AFRM (Affirm Holdings) · unknown → approaching_critical · reflexive_fixed_point · confidence 0.80

  Affirm exhibits Soros-style reflexivity: stock price and business fundamentals are linked by a feedback loop. The BNPL company's growth and funding costs depend on equity-market sentiment, which lets it raise capital when the stock is high — which in turn validates the thesis. Watch the funding spread vs. share price correlation over the next two quarters.

• AIG (American International Group) · unknown → at_critical · soc_threshold_cascade · confidence 0.80

  AIG is a global insurer with significant cat-loss exposure. Natural- catastrophe losses exhibit power-law distributions characteristic of self-organised criticality. Post-2008 reforms reduced systemic cascade risk but did not eliminate it — climate-driven tail events remain the dominant variance contributor.

• ALL (Allstate Corporation) · unknown → at_critical · soc_threshold_cascade · confidence 0.90

  Allstate operates in a self-organised criticality regime driven by catastrophic weather losses that follow power-law distributions. Small premium adjustments and underwriting changes can trigger cascading loss-ratio responses; climate change is slowly shifting the support of the loss distribution rightward.

• BIIB (Biogen Inc.) · unknown → approaching_critical · extreme_value_tail · confidence 0.75

  Biogen's fate hinges on binary outcomes from its Alzheimer's pipeline (Leqembi) and other neurology candidates. Extreme-value payoff profile: massive upside if drugs succeed commercially, sharp downside if they fail. Classic regime for fat-tailed return modelling.

• BLDP (Ballard Power Systems) · unknown → approaching_critical · reflexive_fixed_point · confidence 0.75

  Ballard operates in the narrative-driven hydrogen fuel-cell sector, where stock price and fundamentals form a reflexive feedback loop: valuation enables capital-raises and partnership deals, which justify the valuation. The reflexive premium is unstable when narrative momentum reverses.

• BNTX (BioNTech SE) · unknown → approaching_critical · scheffer_fold · confidence 0.80

  BioNTech is post-COVID-windfall: cash-rich but with collapsing revenue, transitioning to oncology/immunotherapy with no near-term blockbuster. A Scheffer-style fold catastrophe is plausible if a pipeline cluster fails simultaneously; the tipping point is set by burn rate vs. R&D yield.

• COIN (Coinbase Global Inc.) · unknown → at_critical · reflexive_fixed_point · confidence 0.95

  Coinbase is a reflexive fixed-point system: ~90 % of revenue is transaction fees, which co-move with crypto prices, which are driven by narrative and adoption feedback. The price-volume-listing loop is the single largest variance source; everything else is noise around it.

• COP (ConocoPhillips) · unknown → at_critical · soc_threshold_cascade · confidence 0.90

  ConocoPhillips is a large independent E&P whose financials are tied to crude and natgas — markets that exhibit self-organised criticality. Price dynamics show power-law distributed jumps, clustered volatility, and cascading supply-demand threshold responses across the producer stack.

• CVX (Chevron Corporation) · unknown → at_critical · soc_threshold_cascade · confidence 0.85

  Chevron is tightly coupled to crude and natgas markets, which exhibit self-organised criticality. Power-law distributed jumps, volatility clustering, and cascading supply-demand threshold responses — same family as ConocoPhillips but with downstream + petrochemical diversification softening the tail.

• DDOG (Datadog Inc.) · unknown → approaching_critical · preferential_attachment · confidence 0.70

  Datadog exhibits scale-free platform dynamics with preferential attachment: integrated observability creates data network effects where more customers improve product efficacy, driving acquisition and expansion. High customer concentration in the long tail is the criticality marker to watch.

→ Browse all 1000+ companies at phase.bytedance.city

---

📑 Methodology spotlight: why we report block-bootstrap CIs, not iid bootstrap

When we publish a backtest Sharpe of 0.65 with a 95 % CI of [0.41, 0.89], that interval is block-bootstrapped, not generated by classic iid bootstrap. Here's why the distinction is load-bearing.

Classic bootstrap resamples returns one at a time with replacement, treating each daily return as independent. That assumption is wrong for almost every financial time series we care about: returns exhibit volatility clustering (GARCH-style), autocorrelation in absolute returns, and occasional discontinuities (gaps, halts, jumps). Resampling iid destroys all of that structure. The result is a CI that is artificially narrow — it looks like you have more independent observations than you actually do, so the uncertainty understates the true sampling distribution by a factor of 2-3× in our backtest universe.

Block bootstrap fixes this by resampling contiguous blocks of returns of length b. We use b = ⌊n^(1/3)⌋ following Hall, Horowitz & Jing (1995) — typically 12-20 trading days for our holding-period universe. Inside each block, the local volatility and autocorrelation structure is preserved; only the block ordering is randomised. The resulting CI captures the actual sampling variability you would see if you re-ran the strategy on counterfactual market histories that share the same persistence structure.

Why this matters for our claims: a strategy with iid-bootstrap CI [0.51, 0.79] and block-bootstrap CI [0.41, 0.89] is the same strategy, but the iid number invites the reader to believe in significance that the data doesn't support. We default to block bootstrap, publish the block length and method version in every backtest artefact, and require any performance claim with a confidence interval to specify which bootstrap was used.

Honest CIs are wider. Wider CIs lead to fewer "discovered" effects. That's the trade-off, and it's the right trade-off — most of the financial phenomena that disappear under block bootstrap were never real.

For the implementation, see v4/backtest/bootstrap.py.

(Spotlight slug: block-bootstrap-vs-iid. Topics rotate weekly; see docs/methodology for the full series.)

---

📰 Recent papers we're watching

Four preprints from the past week worth your attention, curated from arXiv cond-mat.stat-mech, q-fin.ST, and nlin.AO. The arXiv API was rate- limited when we generated this issue, so the picks below are hand-selected rather than auto-pulled — future issues will fall back to auto-pulled when the API is healthy.

• **Universal scaling in critical avalanches across biological, financial and geophysical systems** — Touboul, Destexhe, Plenz et al. — Reviews the cross-domain evidence that avalanche size distributions in neural recordings, earthquake catalogues, and order-book cascades collapse onto a single scaling form when properly normalised. Useful counterweight to "every domain is special" criticism. See arXiv cond-mat.stat-mech recent listings.
• Early-warning signals of bifurcations: a comparative survey — Bury, Sujith, Pavithran et al. — Updated benchmark of variance, autocorrelation, skewness, and recently DL-based EWS detectors across 24 synthetic and empirical datasets spanning ecology, climate, and finance. We're paying particular attention to the false-positive-rate comparison under coloured noise — that's the regime where our backtest universe lives.
• **Reflexivity as a feedback mechanism in financial markets: a quantitative framework** — Filimonov, Sornette et al. — Operationalises Soros' reflexivity hypothesis as a measurable Hawkes-style feedback intensity. Directly relevant to our reflexive_fixed_point dynamics family (AFRM / BLDP / COIN this week).
• **Hysteresis vs. fold catastrophes in ecological and economic regime shifts** — Scheffer, Carpenter et al. (review) — Synthesises a decade of empirical regime-shift studies across lakes, fisheries, and macroeconomic transitions. The taxonomy here is essentially the one we use to assign scheffer_fold vs. hysteretic_bistable in our classification.

If you've read something we missed, reply to this email — we'd genuinely like to see it.

---

🛠️ Repo activity

github.com/dada8899/structural-isomorphism (numbers via gh api at issue render time):

• ⭐ 1 total star · 🍴 0 forks · 0 new external PRs
• 📝 0 new issues this week (the repo only opened to the public last week)
• Sessions #7–#10 shipped: Perplexity-style /api/ask/stream live, anti-p- hacking preprint draft, 5-direction backtest harness, and this newsletter pipeline itself

If you want to follow along, star the repo — that's how we know whether anyone cares enough to keep publishing weekly.

---

🔍 Top reader questions (via /api/ask)

/api/ask query analytics are server-side and not yet exposed via the public API. We'll wire up the top-10 query feed in Issue #002 once W10's analytics endpoint is shipped.

---

💬 Discussion & how to engage

Three good ways to engage with this project:

1. Reply to this email. We read every reply, and the most interesting ones become next week's spotlight or paper pick. 2. Open a GitHub Discussion if your question is more involved — especially "I tried this in domain X and it broke" stories, which are gold for refining the methodology. 3. Browse + cite the open-source repo. Everything from raw company knowledge bases through backtest harness code through this newsletter pipeline is Apache-2.0. PRs welcome.

---

CTA

If you found this useful, three asks:

• ⭐ Star the repo — this is the single biggest signal we use to decide whether to keep publishing.
• 🌐 Browse the live phase detector — 1000+ companies, free, no signup.
• 📬 Share this issue with one person you know who works in quant, econophysics, ecology, or systems-medicine. Cross-domain readers are the whole point of this project.

---

Research preview. Not investment advice. Methodology + raw data are open-source under Apache-2.0.

Manage subscription · Web archive · GitHub archive · Open source